diff --git a/Changes.md b/Changes.md --- a/Changes.md +++ b/Changes.md @@ -1,5 +1,10 @@ # Change log for the `llvm-extra` package +## 0.9 + +* `Extension`: Move to new package `llvm-extension`. + We now implement advanced instructions using generic LLVM intrinsics. + ## 0.8.1 * `FastMath`: support for simplified arithmetic primitives diff --git a/Makefile b/Makefile --- a/Makefile +++ b/Makefile @@ -1,18 +1,10 @@ .PHONY: sharedobj -ghci: - ghci -Wall -i:src:x86/none src/Run.hs - -ghci-cpuid: - ghci -Wall -i:src:x86/cpuid src/Run.hs - testbuild: - runhaskell Setup.lhs configure --user -fbuildExamples -fbuildTools -fcpuid - runhaskell Setup.lhs build - runhaskell Setup.lhs haddock - - runhaskell Setup.lhs configure --user -f-cpuid + runhaskell Setup.lhs configure --user -fbuildExamples --enable-tests runhaskell Setup.lhs build +# runhaskell Setup.lhs haddock + ./dist/build/llvm-extra-test/llvm-extra-test llvmversion = 2.6 diff --git a/Problems.txt b/Problems.txt deleted file mode 100644 --- a/Problems.txt +++ /dev/null @@ -1,91 +0,0 @@ -LLVM-2.5 running GHCi - -First I can load, say, Array.hs in ghci. -When running 'main' I get the error that LLVMSystem.so cannot be found. -No LLVM*.so file on my machine, cannot be found in a Suse package. -Building .so file manually using gcc as in ./make-so.sh. - -Then I get a problem with pthread.so not being a shared object file -but a script. -However, we do not need pthread anyway, -thus removing it from ~/.ghc/i386-linux-6.10.4/package.conf solves that problem. -This is a known issue: - http://hackage.haskell.org/trac/ghc/ticket/2615#comment:16 - -In GHC-6.12.3 you must also run - ghc-pkg recache --user - after removing pthread from ~/.ghc/i386-linux-6.12.3/package.conf.d/llvm-*.conf - -Now when running 'main' I get the error, -that something about CurrentEngine cannot be found. -It means, we must also include /usr/lib/llvm/LLVM*.o files. -But in what order? -Seems there is no working order, -but the one given by - llvm-config --libs -is close to what we need. -Problem: GHCi cannot cope with weak symbol _ZTIN4llvm12X86SubtargetE ("V" in nm) - -This is a known issue due to - http://hackage.haskell.org/trac/ghc/ticket/3333#comment:3 - -I have used gcc to build a monolithic libLLVM.so -containing all libLLVM*.a and libLLVM*.o files of /usr/lib/llvm/. -Then I reduced the occurrences of LLVM in package.conf -in the extraLibraries field to "LLVM" -and the ldOptions to -lLLVM. - - -Now I can call LLVM in GHCi - but only until I do :reload. -After reload the next attempt to play something will let GHCi quit with: - -ghci: JITEmitter.cpp:110: ::JITResolver::JITResolver(llvm::JIT&): Assertion `TheJITResolver == 0 && "Multiple JIT resolvers?"' failed. - -Maybe this can be handled in LLVM-2.6 -where the JIT must be initialized explicitly. - - -LLVM-2.6: - -If I do as described above, -then when linking the Array example -I get errors for undefined symbols like -/usr/local/lib/libLLVM.so: undefined reference to `AutoGeneratedSwitch_emit_dash_llvm' -/usr/local/lib/libLLVM.so: undefined reference to `AutoGeneratedList_Wl_comma_' -... - -These can be avoided by excluding libplugin_llvmc_Base.a -and libplugin_llvmc_Clang.a from the libLLVM.so conglomerate. -Then the example can be compiled but it aborts with PassRegistrar error as below. - -I GHCi I get «unknown symbol `LLVMGetBitcodeModuleProviderInContext'» -when running 'main' in GHCi. -Additionally to -lLLVM for our custom libLLVM.so -I have to add -lLTO to ldOptions. -Then all symbols can be found. -However, when running LLVM.initializeNativeTarget -GHCi quits with -ghci: Pass.cpp:152: void::PassRegistrar::RegisterPass(const llvm::PassInfo&): Assertion `Inserted && "Pass registered multiple times!"' failed. -Also running LLVM.Target.X86.initializeTarget or Array.renderRamp -leads to this error. - - -LLVM-2.8 - -If I remove 'pthread' from Extra-Libraries and add 'LLVM-2.8rc' to Extra-GHCi-Libraries in - $HOME/.ghc/i386-linux-6.12.3/package.conf.d/llvm-0.9.*-*.conf - and run - ghc-pkg recache --user - I am even able to run 'llvm' example programs from within GHCi! - -Sometimes it happends that llvm.buildinfo contains -ld-options: /usr/local/lib/libLLVM-2.8rc.so -and then llvm-0.9.*-*.conf contains this as well. -This leads to a failure of a LLVM runtime assertion: - -llvm/examples$ DotProd.exe -Two passes with the same argument (-preverify) attempted to be registered! -UNREACHABLE executed! -Aborted - -ld-options field must be empty! diff --git a/llvm-extra.cabal b/llvm-extra.cabal --- a/llvm-extra.cabal +++ b/llvm-extra.cabal @@ -1,6 +1,7 @@ +Cabal-Version: 2.2 Name: llvm-extra -Version: 0.8.3 -License: BSD3 +Version: 0.9 +License: BSD-3-Clause License-File: LICENSE Author: Henning Thielemann Maintainer: Henning Thielemann @@ -33,13 +34,6 @@ . * complex Haskell values mapped to LLVM values in "LLVM.Extra.Multi.Value" . - * automatic adaption to target specific extensions, - currently used for access of vector operations - that are specific to an SSE level on x86 processors - in "LLVM.Extra.Extension" - (On x86 architectures we depend on the cpuid package - that is needed for automatic detection of available features.) - . * advanced vector operations such as sum of all vector elements, cumulative sum, floor, non-negative fraction, absolute value @@ -48,35 +42,20 @@ * type classes for handling scalar and vector operations in a uniform way in "LLVM.Extra.ScalarOrVector" - . - * a Makefile and a description - of how to run LLVM code from within GHCi. Stability: Experimental -Tested-With: GHC==7.0.4, GHC==7.4.2, GHC==7.8.2 -Tested-With: GHC==8.4.3, GHC==8.6.1 -Cabal-Version: 1.14 +Tested-With: GHC==7.0.4, GHC==7.4.2, GHC==7.8.4 +Tested-With: GHC==8.4.4, GHC==8.6.5, GHC==8.8.1 Build-Type: Simple Extra-Source-Files: Makefile Changes.md - Problems.txt - x86/cpuid/LLVM/Extra/ExtensionCheck/X86.hs - x86/none/LLVM/Extra/ExtensionCheck/X86.hs Flag buildExamples description: Build example executables default: False -Flag buildTools - description: Build intrinsic translator - default: False - -Flag cpuid - description: Use CPUID for host feature detection if available on the architecture - default: True - Source-Repository this - Tag: 0.8.3 + Tag: 0.9 Type: darcs Location: http://code.haskell.org/~thielema/llvm-extra/ @@ -86,9 +65,8 @@ Library Build-Depends: - -- llvm must be imported with restrictive version bounds, - -- because we import implicitly and unqualified - llvm-tf >=3.1.2 && <3.2, + private, + llvm-tf >=9.0 && <9.1, tfp >=1.0 && <1.1, non-empty >=0.2.1 && <0.4, containers >=0.1 && <0.7, @@ -98,23 +76,9 @@ transformers >=0.1.1 && <0.6, tagged >=0.7 && <0.9, utility-ht >=0.0.11 && <0.1, - prelude-compat >=0.0 && <0.0.1 - - Build-Depends: + prelude-compat >=0.0 && <0.0.1, base >=3 && <5 - If (arch(i386) || arch(x86_64)) && flag(cpuid) - Build-Depends: - unsafe >=0.0 && <0.1, - cpuid >=0.2.3 && <0.3 - Hs-Source-Dirs: x86/cpuid - Else - -- Instead of calling the cpuid instruction directly - -- we may ask LLVM's Subtarget detection. - -- This would also enable cross compilation. - -- However in LLVM-2.6 this is only available in the C++ interface. - Hs-Source-Dirs: x86/none - Default-Language: Haskell98 GHC-Options: -Wall Hs-source-dirs: src @@ -127,9 +91,6 @@ LLVM.Extra.Either LLVM.Extra.Class LLVM.Extra.Control - LLVM.Extra.Extension - LLVM.Extra.Extension.X86 - LLVM.Extra.ExtensionCheck.X86 LLVM.Extra.Array LLVM.Extra.Scalar LLVM.Extra.Vector @@ -145,13 +106,26 @@ LLVM.Extra.Multi.Vector.Instance LLVM.Extra.Multi.Class Other-Modules: - LLVM.Extra.ArithmeticPrivate LLVM.Extra.MaybePrivate LLVM.Extra.EitherPrivate LLVM.Extra.MemoryPrivate - LLVM.Extra.Extension.X86Auto LLVM.Extra.Multi.Value.Private +Library private + Build-Depends: + llvm-tf, + tfp, + non-empty, + utility-ht, + base + + Default-Language: Haskell98 + GHC-Options: -Wall + Hs-source-dirs: private + Exposed-Modules: + LLVM.Extra.ScalarOrVectorPrivate + LLVM.Extra.ArithmeticPrivate + Executable tone-llvm If flag(buildExamples) Build-Depends: @@ -169,16 +143,21 @@ GHC-Options: -Wall Main-Is: src/Array.hs -Executable prepare-intrinsics - If flag(buildTools) - Build-Depends: - parsec >=2.1 && <3.2, - containers >=0.1 && <0.7, - transformers, - utility-ht >=0.0.1 && <0.1, - base >=3 && <5 - Else - Buildable: False +Test-Suite llvm-extra-test + Type: exitcode-stdio-1.0 + Build-Depends: + QuickCheck >=2.11 && <3, + private, + llvm-extra, + llvm-tf, + tfp, + storable-tuple >=0.0.3 && <0.1, + utility-ht >=0.0.1 && <0.1, + base >=3 && <5 Default-Language: Haskell98 GHC-Options: -Wall - Main-Is: src/PrepareIntrinsics.hs + Hs-Source-Dirs: test + Main-Is: Main.hs + Other-Modules: + Test.Vector + LLVM.Extra.VectorAlt diff --git a/private/LLVM/Extra/ArithmeticPrivate.hs b/private/LLVM/Extra/ArithmeticPrivate.hs new file mode 100644 --- /dev/null +++ b/private/LLVM/Extra/ArithmeticPrivate.hs @@ -0,0 +1,127 @@ +{-# LANGUAGE TypeFamilies #-} +module LLVM.Extra.ArithmeticPrivate where + +import qualified LLVM.Util.Intrinsic as Intrinsic + +import qualified LLVM.Core as LLVM +import LLVM.Core + (CodeGenFunction, valueOf, Value, + CmpPredicate(CmpLE, CmpGE), FPPredicate, CmpRet, CmpResult, + IsConst, IsPrimitive, IsArithmetic, IsInteger, IsFloating, + getElementPtr, ) + +import Foreign.Ptr (Ptr, ) +import Data.Word (Word32, ) +import Data.Int (Int32, ) + +import Prelude hiding (and, or, sqrt, sin, cos, exp, log, abs, min, max, ) + + +add :: + (IsArithmetic a) => + Value a -> Value a -> CodeGenFunction r (Value a) +add = LLVM.add + +sub :: + (IsArithmetic a) => + Value a -> Value a -> CodeGenFunction r (Value a) +sub = LLVM.sub + + +inc :: + (IsArithmetic a, IsConst a, Num a) => + Value a -> CodeGenFunction r (Value a) +inc x = add x (valueOf 1) + +dec :: + (IsArithmetic a, IsConst a, Num a) => + Value a -> CodeGenFunction r (Value a) +dec x = sub x (valueOf 1) + +advanceArrayElementPtr :: + Value (Ptr a) -> + CodeGenFunction r (Value (Ptr a)) +advanceArrayElementPtr p = + getElementPtr p (valueOf 1 :: Value Word32, ()) + +decreaseArrayElementPtr :: + Value (Ptr a) -> + CodeGenFunction r (Value (Ptr a)) +decreaseArrayElementPtr p = + getElementPtr p (valueOf (-1) :: Value Int32, ()) + + + +mul :: + (IsArithmetic a) => + Value a -> Value a -> CodeGenFunction r (Value a) +mul = LLVM.mul + + +{- | +This would also work for vectors, +but LLVM-3.1 crashes when actually doing this. +-} +min :: (CmpRet a) => Value a -> Value a -> CodeGenFunction r (Value a) +min = cmpSelect (cmp CmpLE) + +max :: (CmpRet a) => Value a -> Value a -> CodeGenFunction r (Value a) +max = cmpSelect (cmp CmpGE) + +abs :: (IsArithmetic a, CmpRet a) => + Value a -> CodeGenFunction r (Value a) +abs x = do + b <- cmp LLVM.CmpGE x (LLVM.value LLVM.zero) + LLVM.select b x =<< LLVM.neg x + + +signumGen :: + (CmpRet a, IsPrimitive a) => + Value a -> Value a -> + Value a -> CodeGenFunction r (Value a) +signumGen minusOne one x = do + let zero = LLVM.value LLVM.zero + negative <- cmp LLVM.CmpLT x zero + positive <- cmp LLVM.CmpGT x zero + LLVM.select negative minusOne + =<< LLVM.select positive one zero + +signum :: + (Num a, CmpRet a, IsConst a, IsPrimitive a) => + Value a -> CodeGenFunction r (Value a) +signum = signumGen (LLVM.valueOf (-1)) (LLVM.valueOf 1) + + +cmpSelect :: + (CmpRet a) => + (Value a -> Value a -> CodeGenFunction r (Value (CmpResult a))) -> + (Value a -> Value a -> CodeGenFunction r (Value a)) +cmpSelect f x y = + f x y >>= \b -> LLVM.select b x y + + +fcmp :: + (IsFloating a, CmpRet a, CmpResult a ~ b) => + FPPredicate -> Value a -> Value a -> CodeGenFunction r (Value b) +fcmp = LLVM.fcmp + +cmp :: + (CmpRet a, CmpResult a ~ b) => + CmpPredicate -> Value a -> Value a -> CodeGenFunction r (Value b) +cmp = LLVM.cmp + + + +and :: + (IsInteger a) => + Value a -> Value a -> CodeGenFunction r (Value a) +and = LLVM.and + +or :: + (IsInteger a) => + Value a -> Value a -> CodeGenFunction r (Value a) +or = LLVM.or + + +fraction :: (IsFloating a) => Value a -> CodeGenFunction r (Value a) +fraction x = sub x =<< Intrinsic.floor x diff --git a/private/LLVM/Extra/ScalarOrVectorPrivate.hs b/private/LLVM/Extra/ScalarOrVectorPrivate.hs new file mode 100644 --- /dev/null +++ b/private/LLVM/Extra/ScalarOrVectorPrivate.hs @@ -0,0 +1,141 @@ +{-# LANGUAGE TypeFamilies #-} +module LLVM.Extra.ScalarOrVectorPrivate where + +import qualified LLVM.Extra.ArithmeticPrivate as A + +import qualified Type.Data.Num.Decimal as TypeNum +import Type.Data.Num.Decimal (D1) + +import qualified LLVM.Core as LLVM +import LLVM.Core + (Value, ConstValue, valueOf, + CmpRet, ShapeOf, + Vector, WordN, IntN, FP128, + IsConst, IsInteger, CodeGenFunction) + +import qualified Data.NonEmpty as NonEmpty +import Data.Word (Word8, Word16, Word32, Word64) +import Data.Int (Int8, Int16, Int32, Int64) + +import Prelude hiding (replicate) + + +type family Scalar vector :: * + +type instance Scalar Float = Float +type instance Scalar Double = Double +type instance Scalar FP128 = FP128 +type instance Scalar Bool = Bool +type instance Scalar Int8 = Int8 +type instance Scalar Int16 = Int16 +type instance Scalar Int32 = Int32 +type instance Scalar Int64 = Int64 +type instance Scalar Word8 = Word8 +type instance Scalar Word16 = Word16 +type instance Scalar Word32 = Word32 +type instance Scalar Word64 = Word64 +type instance Scalar (IntN d) = IntN d +type instance Scalar (WordN d) = WordN d +type instance Scalar (Vector n a) = a + + +class Replicate vector where + -- | an alternative is using the 'Vector.Constant' vector type + replicate :: Value (Scalar vector) -> CodeGenFunction r (Value vector) + replicateConst :: ConstValue (Scalar vector) -> ConstValue vector + +instance Replicate Float where replicate = return; replicateConst = id; +instance Replicate Double where replicate = return; replicateConst = id; +instance Replicate FP128 where replicate = return; replicateConst = id; +instance Replicate Bool where replicate = return; replicateConst = id; +instance Replicate Int8 where replicate = return; replicateConst = id; +instance Replicate Int16 where replicate = return; replicateConst = id; +instance Replicate Int32 where replicate = return; replicateConst = id; +instance Replicate Int64 where replicate = return; replicateConst = id; +instance Replicate Word8 where replicate = return; replicateConst = id; +instance Replicate Word16 where replicate = return; replicateConst = id; +instance Replicate Word32 where replicate = return; replicateConst = id; +instance Replicate Word64 where replicate = return; replicateConst = id; +instance Replicate (IntN d) where replicate = return; replicateConst = id; +instance Replicate (WordN d) where replicate = return; replicateConst = id; +instance + (TypeNum.Positive n, LLVM.IsPrimitive a) => + Replicate (Vector n a) where + replicate x = do + v <- singleton x + LLVM.shufflevector v (LLVM.value LLVM.undef) LLVM.zero + replicateConst x = LLVM.constCyclicVector $ NonEmpty.Cons x [] + +singleton :: + (LLVM.IsPrimitive a) => + Value a -> CodeGenFunction r (Value (Vector D1 a)) +singleton x = + LLVM.insertelement (LLVM.value LLVM.undef) x (valueOf 0) + + +uaddSat, usubSat :: + (IsInteger v, CmpRet v, Replicate v, Scalar v ~ a, IsConst a, Bounded a) => + Value v -> Value v -> CodeGenFunction r (Value v) +uaddSat x y = do + z <- A.add x y + wrong <- A.cmp LLVM.CmpLT z x + maxBnd <- replicate $ valueOf maxBound + LLVM.select wrong maxBnd z +usubSat x y = do + z <- A.sub x y + wrong <- A.cmp LLVM.CmpGT z x + LLVM.select wrong (LLVM.value LLVM.zero) z + +saddSat, ssubSat :: + (IsInteger v, CmpRet v, Replicate v, ShapeOf v ~ shape, + LLVM.ShapedType shape Bool ~ bv, ShapeOf bv ~ shape, CmpRet bv, + Scalar v ~ a, IsConst a, Bounded a) => + Value v -> Value v -> CodeGenFunction r (Value v) + +saddSat x y = do + z <- A.add x y + nonNegX <- A.cmp LLVM.CmpGE x $ LLVM.value LLVM.zero + nonNegY <- A.cmp LLVM.CmpGE y $ LLVM.value LLVM.zero + distinctSign <- A.cmp LLVM.CmpNE nonNegX nonNegY + overflow <- A.cmp LLVM.CmpLT z x + underflow <- A.cmp LLVM.CmpGT z x + maxBnd <- replicate $ valueOf maxBound + minBnd <- replicate $ valueOf minBound + maxSat <- LLVM.select overflow maxBnd z + minSat <- LLVM.select underflow minBnd z + saturated <- LLVM.select nonNegX maxSat minSat + LLVM.select distinctSign z saturated + +ssubSat x y = do + z <- A.sub x y + nonNegX <- A.cmp LLVM.CmpGE x $ LLVM.value LLVM.zero + nonNegY <- A.cmp LLVM.CmpGE y $ LLVM.value LLVM.zero + sameSign <- A.cmp LLVM.CmpEQ nonNegX nonNegY + overflow <- A.cmp LLVM.CmpLT z x + underflow <- A.cmp LLVM.CmpGT z x + maxBnd <- replicate $ valueOf maxBound + minBnd <- replicate $ valueOf minBound + maxSat <- LLVM.select overflow maxBnd z + minSat <- LLVM.select underflow minBnd z + saturated <- LLVM.select nonNegX maxSat minSat + LLVM.select sameSign z saturated + +saddSatLogical :: + (IsInteger v, CmpRet v, Replicate v, ShapeOf v ~ shape, + LLVM.ShapedType shape Bool ~ bv, ShapeOf bv ~ shape, CmpRet bv, + IsInteger bv, + Scalar v ~ a, IsConst a, Bounded a) => + Value v -> Value v -> CodeGenFunction r (Value v) +saddSatLogical x y = do + z <- A.add x y + nonNegX <- A.cmp LLVM.CmpGE x $ LLVM.value LLVM.zero + nonNegY <- A.cmp LLVM.CmpGE y $ LLVM.value LLVM.zero + distinctSign <- A.cmp LLVM.CmpNE nonNegX nonNegY + minBnd <- replicate $ valueOf minBound + maxBnd <- replicate $ valueOf maxBound + bounds <- LLVM.select nonNegX maxBnd minBnd + overflow <- A.cmp LLVM.CmpLT z y + underflow <- A.cmp LLVM.CmpGT z y + xflow <- LLVM.select nonNegX overflow underflow + correctSum <- A.or distinctSign xflow + LLVM.select correctSum z bounds diff --git a/src/Array.hs b/src/Array.hs --- a/src/Array.hs +++ b/src/Array.hs @@ -3,15 +3,13 @@ {-# LANGUAGE ForeignFunctionInterface #-} module Main where -import LLVM.Extra.Control (arrayLoop, ) import qualified LLVM.Extra.ScalarOrVector as SoV import qualified LLVM.Extra.Vector as Vector -import qualified LLVM.Extra.Extension.X86 as X86 -import qualified LLVM.Extra.Extension as Ext - +import qualified LLVM.Extra.Iterator as Iter import qualified LLVM.Extra.Class as Class import qualified LLVM.Extra.Arithmetic as A +import LLVM.Extra.Control (arrayLoop, ) import qualified LLVM.ExecutionEngine as EE import qualified LLVM.Core as LLVM @@ -32,8 +30,10 @@ import qualified Data.Empty as Empty import Data.NonEmpty ((!:), ) -import Control.Monad.Trans.State (StateT(StateT), runStateT, ) -import Control.Monad (liftM2, ) +import Control.Monad.Trans.State (StateT(StateT), runStateT) +import Control.Monad.HT ((<=<)) +import Control.Monad (liftM2) +import Control.Applicative (liftA2) @@ -59,9 +59,7 @@ because 'frem' is only available in the FPU. -} fractionVector0 :: - (LLVM.IsFloating c, LLVM.ABinOp a (Value (Vector D4 Float)), - LLVM.ABinOpResult a (Value (Vector D4 Float)) ~ (v c)) => - a -> CodeGenFunction r (v c) + Value (Vector D4 Float) -> CodeGenFunction r (Value (Vector D4 Float)) fractionVector0 x = frem x =<< constVec 1 @@ -138,24 +136,6 @@ A.mul (valueOf 0.25) =<< A.add s0 s1 -{- -Needs the horizontal add instruction from the SSSE3 extension in ix86 CPUs. --} -mixHorizontal :: Value (Vector D4 Float) -> CodeGenFunction r (Value Float) -mixHorizontal y = do - z <- Ext.runUnsafe X86.haddps y (value undef) - s <- Ext.runUnsafe X86.haddps z (value undef) - A.mul (valueOf 0.25) =<< extractelement s (valueOf 0) - -{- -Needs the dot product instruction from the SSE4 extension in ix86 CPUs. --} -mixDotProduct :: Value (Vector D4 Float) -> CodeGenFunction r (Value Float) -mixDotProduct y = do - x <- SoV.replicate (valueOf 0.25) - z <- Ext.runUnsafe X86.dpps x y (valueOf 0xF1) - extractelement z (valueOf 0) - mChorusVector :: CodeGenModule (Function @@ -166,11 +146,29 @@ const1 <- constVec 1 const2 <- constVec (-2) s <- arrayLoop size ptr (value (zero :: Vec)) $ \ ptri phase -> do - flip store ptri =<< mixHorizontal =<< add const1 =<< mul const2 phase + flip store ptri =<< mixGeneric =<< add const1 =<< mul const2 phase Vector.fraction =<< A.add phase freq ss <- extractelement s (valueOf 0) ret ss +mChorusVectorIterator :: + CodeGenModule + (Function + (Word32 -> Ptr Float -> Float -> Float -> Float -> Float -> IO Float)) +mChorusVectorIterator = + createFunction ExternalLinkage $ \ size ptr f0 f1 f2 f3 -> do + freq <- Vector.assemble [f0,f1,f2,f3] + const1 <- constVec 1 + const2 <- constVec (-2) + Iter.mapM_ id $ Iter.take size $ + liftA2 + (\ptri phase -> + flip store ptri =<< mixGeneric =<< add const1 =<< mul const2 phase) + (Iter.arrayPtrs ptr) + (Iter.iterate (Vector.fraction <=< A.add freq) (value (zero :: Vec))) + ret (value zero :: Value Float) + + waveSaw :: Value Float -> CodeGenFunction r (Value Float) waveSaw t = A.sub (valueOf 1) =<< @@ -263,7 +261,7 @@ renderChorus :: IO () renderChorus = do - fill <- generateFunction derefChorusPtr mChorusVector + fill <- generateFunction derefChorusPtr mChorusVectorIterator IO.withFile "speedtest.f32" IO.WriteMode $ \h -> let len = 10000000 in allocaArray len $ \ ptr -> diff --git a/src/LLVM/Extra/Arithmetic.hs b/src/LLVM/Extra/Arithmetic.hs --- a/src/LLVM/Extra/Arithmetic.hs +++ b/src/LLVM/Extra/Arithmetic.hs @@ -1,4 +1,5 @@ {-# LANGUAGE TypeFamilies #-} +{-# LANGUAGE UndecidableInstances #-} {-# LANGUAGE FlexibleContexts #-} module LLVM.Extra.Arithmetic ( -- * arithmetic: generalized and improved type inference @@ -24,16 +25,16 @@ Transcendental (pi, sin, cos, exp, log, pow), ) where +import qualified LLVM.Util.Intrinsic as Intrinsic import LLVM.Extra.ArithmeticPrivate (inc, dec, advanceArrayElementPtr, decreaseArrayElementPtr, ) import qualified LLVM.Extra.Class as Class import qualified LLVM.Extra.ScalarOrVector as SoV -import qualified LLVM.Util.Proxy as LP import qualified LLVM.Core as LLVM import LLVM.Core (CodeGenFunction, value, Value, ConstValue, - IsType, IsInteger, IsFloating, IsArithmetic, IsFirstClass, ) + IsInteger, IsFloating, IsArithmetic) import Control.Monad (liftM2, liftM3, ) @@ -275,55 +276,12 @@ inv = LLVM.inv -valueTypeName :: - (IsType a) => - Value a -> String -valueTypeName = - LLVM.intrinsicTypeName . ((\_ -> LP.Proxy) :: Value a -> LP.Proxy a) - -callIntrinsic1 :: - (IsFirstClass a) => - String -> Value a -> CodeGenFunction r (Value a) -callIntrinsic1 fn x = do - op <- LLVM.externFunction ("llvm." ++ fn ++ "." ++ valueTypeName x) - LLVM.call op x >>= addReadNone - -callIntrinsic2 :: - (IsFirstClass a) => - String -> Value a -> Value a -> CodeGenFunction r (Value a) -callIntrinsic2 fn x y = do - op <- LLVM.externFunction ("llvm." ++ fn ++ "." ++ valueTypeName x) - LLVM.call op x y >>= addReadNone - - -{- -If we add the attribute, then LLVM-2.8 complains: - -$ ./dist/build/synthi-llvm-test/synthi-llvm-test -Attribute readnone only applies to the function! - %97 = call readnone float @llvm.sin.f32(float %96) -Attribute readnone only applies to the function! - %99 = call readnone float @llvm.exp.f32(float %98) -Attribute readnone only applies to the function! - %102 = call readnone float @llvm.cos.f32(float %101) -Broken module found, compilation aborted! -Stack dump: -0. Running pass 'Function Pass Manager' on module '_module'. -1. Running pass 'Module Verifier' on function '@fillsignal' -make: *** [test] Abgebrochen --} -addReadNone :: Value a -> CodeGenFunction r (Value a) -addReadNone x = do --- LLVM.addAttributes x 0 [LLVM.ReadNoneAttribute] - return x - - class Field a => Algebraic a where sqrt :: a -> CodeGenFunction r a instance (IsFloating a) => Algebraic (Value a) where - sqrt = callIntrinsic1 "sqrt" + sqrt = Intrinsic.call1 "sqrt" class Algebraic a => Transcendental a where @@ -333,8 +291,8 @@ instance (IsFloating a, SoV.TranscendentalConstant a) => Transcendental (Value a) where pi = return $ value SoV.constPi - sin = callIntrinsic1 "sin" - cos = callIntrinsic1 "cos" - exp = callIntrinsic1 "exp" - log = callIntrinsic1 "log" - pow = callIntrinsic2 "pow" + sin = Intrinsic.call1 "sin" + cos = Intrinsic.call1 "cos" + exp = Intrinsic.call1 "exp" + log = Intrinsic.call1 "log" + pow = Intrinsic.call2 "pow" diff --git a/src/LLVM/Extra/ArithmeticPrivate.hs b/src/LLVM/Extra/ArithmeticPrivate.hs deleted file mode 100644 --- a/src/LLVM/Extra/ArithmeticPrivate.hs +++ /dev/null @@ -1,124 +0,0 @@ -{-# LANGUAGE TypeFamilies #-} -module LLVM.Extra.ArithmeticPrivate where - -import qualified LLVM.Core as LLVM -import LLVM.Core - (CodeGenFunction, valueOf, Value, - CmpPredicate(CmpLE, CmpGE), FPPredicate, CmpRet, CmpResult, - IsConst, IsFirstClass, IsArithmetic, IsInteger, IsFloating, - getElementPtr, ) - -import Foreign.Ptr (Ptr, ) -import Data.Word (Word32, ) -import Data.Int (Int32, ) - -import Prelude hiding (and, or, sqrt, sin, cos, exp, log, abs, min, max, ) - - -add :: - (IsArithmetic a) => - Value a -> Value a -> CodeGenFunction r (Value a) -add = LLVM.add - -sub :: - (IsArithmetic a) => - Value a -> Value a -> CodeGenFunction r (Value a) -sub = LLVM.sub - - -inc :: - (IsArithmetic a, IsConst a, Num a) => - Value a -> CodeGenFunction r (Value a) -inc x = add x (valueOf 1) - -dec :: - (IsArithmetic a, IsConst a, Num a) => - Value a -> CodeGenFunction r (Value a) -dec x = sub x (valueOf 1) - -advanceArrayElementPtr :: - Value (Ptr a) -> - CodeGenFunction r (Value (Ptr a)) -advanceArrayElementPtr p = - getElementPtr p (valueOf 1 :: Value Word32, ()) - -decreaseArrayElementPtr :: - Value (Ptr a) -> - CodeGenFunction r (Value (Ptr a)) -decreaseArrayElementPtr p = - getElementPtr p (valueOf (-1) :: Value Int32, ()) - - - -mul :: - (IsArithmetic a) => - Value a -> Value a -> CodeGenFunction r (Value a) -mul = LLVM.mul - - -{- | -This would also work for vectors, -but LLVM-3.1 crashes when actually doing this. --} -min :: (IsFirstClass a, CmpRet a) => - Value a -> Value a -> CodeGenFunction r (Value a) -min = cmpSelect (cmp CmpLE) - -max :: (IsFirstClass a, CmpRet a) => - Value a -> Value a -> CodeGenFunction r (Value a) -max = cmpSelect (cmp CmpGE) - -abs :: (IsArithmetic a, CmpRet a) => - Value a -> CodeGenFunction r (Value a) -abs x = max x =<< LLVM.neg x - - -signumGen :: - (LLVM.IsFirstClass a, - LLVM.CmpRet a, LLVM.CmpResult a ~ Bool) => - LLVM.Value a -> LLVM.Value a -> - Value a -> CodeGenFunction r (Value a) -signumGen minusOne one x = do - let zero = LLVM.value LLVM.zero - negative <- cmp LLVM.CmpLT x zero - positive <- cmp LLVM.CmpGT x zero - LLVM.select negative minusOne - =<< LLVM.select positive one zero - -signum :: - (Num a, - LLVM.IsConst a, LLVM.IsFirstClass a, - LLVM.CmpRet a, LLVM.CmpResult a ~ Bool) => - Value a -> CodeGenFunction r (Value a) -signum = signumGen (LLVM.valueOf (-1)) (LLVM.valueOf 1) - - -cmpSelect :: - (IsFirstClass a, CmpRet a) => - (Value a -> Value a -> CodeGenFunction r (Value (CmpResult a))) -> - (Value a -> Value a -> CodeGenFunction r (Value a)) -cmpSelect f x y = - f x y >>= \b -> LLVM.select b x y - - -fcmp :: - (IsFloating a, CmpRet a, CmpResult a ~ b) => - FPPredicate -> Value a -> Value a -> CodeGenFunction r (Value b) -fcmp = LLVM.fcmp - -cmp :: - (CmpRet a, CmpResult a ~ b) => - CmpPredicate -> Value a -> Value a -> CodeGenFunction r (Value b) -cmp = LLVM.cmp - - - -and :: - (IsInteger a) => - Value a -> Value a -> CodeGenFunction r (Value a) -and = LLVM.and - -or :: - (IsInteger a) => - Value a -> Value a -> CodeGenFunction r (Value a) -or = LLVM.or diff --git a/src/LLVM/Extra/Control.hs b/src/LLVM/Extra/Control.hs --- a/src/LLVM/Extra/Control.hs +++ b/src/LLVM/Extra/Control.hs @@ -28,8 +28,8 @@ br, condBr, Value, value, valueOf, phi, addPhiInputs, - CmpPredicate(CmpGT), CmpRet, CmpResult, - IsInteger, IsType, IsConst, IsFirstClass, + CmpPredicate(CmpGT), CmpRet, + IsInteger, IsType, IsConst, IsPrimitive, CodeGenFunction, CodeGenModule, newModule, defineModule, writeBitcodeToFile, ) @@ -51,7 +51,7 @@ -} arrayLoop :: (Phi a, IsType b, - Num i, IsConst i, IsInteger i, CmpRet i, CmpResult i ~ Bool) => + Num i, IsConst i, IsInteger i, CmpRet i, IsPrimitive i) => Value i -> Value (Ptr b) -> a -> (Value (Ptr b) -> a -> CodeGenFunction r a) -> CodeGenFunction r a @@ -64,7 +64,7 @@ arrayLoop2 :: (Phi s, IsType a, IsType b, - Num i, IsConst i, IsInteger i, CmpRet i, CmpResult i ~ Bool) => + Num i, IsConst i, IsInteger i, CmpRet i, IsPrimitive i) => Value i -> Value (Ptr a) -> Value (Ptr b) -> s -> (Value (Ptr a) -> Value (Ptr b) -> s -> CodeGenFunction r s) -> CodeGenFunction r s @@ -79,7 +79,7 @@ arrayLoopWithExit :: (Phi s, IsType a, - Num i, IsConst i, IsInteger i, CmpRet i, CmpResult i ~ Bool) => + Num i, IsConst i, IsInteger i, CmpRet i, IsPrimitive i) => Value i -> Value (Ptr a) -> s -> (Value (Ptr a) -> s -> CodeGenFunction r (Value Bool, s)) -> CodeGenFunction r (Value i, s) @@ -107,7 +107,7 @@ -} _arrayLoopWithExitDecLoop :: (Phi a, IsType b, - Num i, IsConst i, IsInteger i, CmpRet i, CmpResult i ~ Bool) => + Num i, IsConst i, IsInteger i, CmpRet i, IsPrimitive i) => Value i -> Value (Ptr b) -> a -> (Value (Ptr b) -> a -> CodeGenFunction r (Value Bool, a)) -> CodeGenFunction r (Value i, a) @@ -153,7 +153,7 @@ arrayLoop2WithExit :: (Phi s, IsType a, IsType b, - Num i, IsConst i, IsInteger i, CmpRet i, CmpResult i ~ Bool) => + Num i, IsConst i, IsInteger i, CmpRet i, IsPrimitive i) => Value i -> Value (Ptr a) -> Value (Ptr b) -> s -> (Value (Ptr a) -> Value (Ptr b) -> s -> CodeGenFunction r (Value Bool, s)) -> CodeGenFunction r (Value i, s) @@ -168,7 +168,7 @@ fixedLengthLoop :: (Phi s, - Num i, IsConst i, IsInteger i, CmpRet i, CmpResult i ~ Bool) => + Num i, IsConst i, IsInteger i, CmpRet i, IsPrimitive i) => Value i -> s -> (s -> CodeGenFunction r s) -> CodeGenFunction r s @@ -317,7 +317,7 @@ class Phi a => Select a where select :: Value Bool -> a -> a -> CodeGenFunction r a -instance (IsFirstClass a, CmpRet a, CmpResult a ~ Bool) => Select (Value a) where +instance (CmpRet a, IsPrimitive a) => Select (Value a) where select = LLVM.select instance Select () where diff --git a/src/LLVM/Extra/Extension.hs b/src/LLVM/Extra/Extension.hs deleted file mode 100644 --- a/src/LLVM/Extra/Extension.hs +++ /dev/null @@ -1,178 +0,0 @@ -{-# LANGUAGE TypeFamilies #-} -{-# LANGUAGE FlexibleInstances #-} -{-# LANGUAGE FlexibleContexts #-} -{-# LANGUAGE GeneralizedNewtypeDeriving #-} -{-# LANGUAGE Rank2Types #-} -module LLVM.Extra.Extension ( - T, Result, CallArgs, - Subtarget(Subtarget), wrap, - intrinsic, intrinsicAttr, - run, runWhen, runUnsafe, - with, with2, with3, - ) where - -import qualified LLVM.Core as LLVM -import LLVM.Core - (Value, CodeGenFunction, externFunction, call, - addAttributes, Attribute {- (ReadNoneAttribute) -}, ) - -import Data.Map (Map, ) -import qualified Data.Map as Map - -import Control.Monad.Trans.Writer (Writer, writer, runWriter, ) -import Control.Monad (join, ) -import Control.Applicative (Applicative, pure, (<*>), ) - -import Prelude2010 hiding (replicate, sum, map, zipWith, ) -import Prelude () - - -data Subtarget = - Subtarget { - targetName, name :: String, - check :: forall r. CodeGenFunction r Bool - } - - -{- | -This is an Applicative functor that registers, -what extensions are needed in order to run the contained instructions. -You can escape from the functor by calling 'run' -and providing a generic implementation. - -We use an applicative functor -since with a monadic interface -we had to create the specialised code in every case, -in order to see which extensions where used -in the course of creating the instructions. - -We use only one (unparameterized) type for all extensions, -since this is the most simple solution. -Alternatively we could use a type parameter -where class constraints show what extensions are needed. -This would be just like exceptions that are explicit in the type signature -as in the control-monad-exception package. -However we would still need to lift all basic LLVM instructions to the new monad. --} -newtype T a = - Cons (Writer (Map String Subtarget) a) - deriving (Functor, Applicative) - -{- | -Declare that a certain plain LLVM instruction -depends on a particular extension. -This can be useful if you rely on the data layout -of a certain architecture when doing a bitcast, -or if you know that LLVM translates a certain generic operation -to something especially optimal for the declared extension. --} -wrap :: Subtarget -> a -> T a -wrap tar cgf = - Cons $ - writer (cgf, Map.singleton (name tar) tar) - - -type family Result g :: * -type instance Result (a -> g) = Result g -type instance Result (CodeGenFunction r a) = r - - -{- | Analogous to 'LLVM.FunctionArgs' - -The type parameter @r@ and its functional dependency are necessary -since @g@ must be a function of the form @a -> ... -> c -> CodeGenFunction r d@ -and we must ensure that the explicit @r@ and the implicit @r@ in the @g@ do match. --} -class CallArgs g where - buildIntrinsic :: [Attribute] -> CodeGenFunction (Result g) g -> g - -instance (CallArgs g) => - CallArgs (Value a -> g) where - buildIntrinsic attrs g x = - buildIntrinsic attrs (fmap ($x) g) - -instance CallArgs (CodeGenFunction r (Value a)) where - buildIntrinsic attrs g = do - z <- join g - addAttributes z 0 attrs - return z - -{- | -Create an intrinsic and register the needed extension. -We cannot immediately check whether the signature matches -or whether the right extension is given. -However, when resolving intrinsics -LLVM will not find the intrinsic if the extension is wrong, -and it also checks the signature. --} -intrinsic :: - (LLVM.IsFunction f, LLVM.CallArgs f g (Result g), CallArgs g) => - Subtarget -> String -> T g -intrinsic = - intrinsicAttr [{- ReadNoneAttribute -}] - -intrinsicAttr :: - (LLVM.IsFunction f, LLVM.CallArgs f g (Result g), CallArgs g) => - [Attribute] -> Subtarget -> String -> T g -intrinsicAttr attrs tar intr = - wrap tar $ - buildIntrinsic attrs $ - fmap call $ - externFunction $ - "llvm." ++ targetName tar ++ "." ++ name tar ++ "." ++ intr - - -infixl 1 `run` - -{- | -@run generic specific@ generates the @specific@ code -if the required extensions are available on the host processor -and @generic@ otherwise. --} -run :: - CodeGenFunction r a -> - T (CodeGenFunction r a) -> - CodeGenFunction r a -run alt (Cons m) = do - let (a,s) = runWriter m - b <- mapM check (Map.elems s) - if and b - then a - else alt - -{- | -Convenient variant of 'run': -Only run the code with extended instructions -if an additional condition is satisfied. --} -runWhen :: - Bool -> - CodeGenFunction r a -> - T (CodeGenFunction r a) -> - CodeGenFunction r a -runWhen c alt (Cons m) = do - let (a,s) = runWriter m - b <- mapM check (Map.elems s) - if c && and b - then a - else alt - -{- | -Only for debugging purposes. --} -runUnsafe :: - T a -> a -runUnsafe (Cons m) = - fst $ runWriter m - - -with :: (Functor f) => f a -> (a -> b) -> f b -with = flip fmap - -with2 :: (Applicative f) => f a -> f b -> (a -> b -> c) -> f c -with2 a b f = - pure f <*> a <*> b - -with3 :: (Applicative f) => f a -> f b -> f c -> (a -> b -> c -> d) -> f d -with3 a b c f = - pure f <*> a <*> b <*> c diff --git a/src/LLVM/Extra/Extension/X86.hs b/src/LLVM/Extra/Extension/X86.hs deleted file mode 100644 --- a/src/LLVM/Extra/Extension/X86.hs +++ /dev/null @@ -1,367 +0,0 @@ -{-# LANGUAGE TypeFamilies #-} -{-# LANGUAGE FlexibleContexts #-} -{- | -Some special operations on X86 processors. -If you want to use them in algorithms -you will always have to prepare an alternative implementation -in terms of plain LLVM instructions. -You will then run them with 'Ext.run' -and this driver function then selects the most advanced of both implementations. -Functions that are written this way can be found in "LLVM.Extra.Vector". -Availability of extensions is checked with the @CPUID@ instruction. -However this does only work if you compile code for the host machine, -that is cross compilation will fail! -For cross compilation we would need access to the SubTarget detection of LLVM -that is only available in the C++ interface in version 2.6. --} -module LLVM.Extra.Extension.X86 ( - X86.maxss, X86.minss, X86.maxps, X86.minps, - X86.maxsd, X86.minsd, X86.maxpd, X86.minpd, - cmpss, cmpps, cmpsd, cmppd, cmpps256, cmppd256, - pcmpgtb, pcmpgtw, pcmpgtd, pcmpgtq, - pcmpugtb, pcmpugtw, pcmpugtd, pcmpugtq, - pminsb, pminsw, pminsd, - pmaxsb, pmaxsw, pmaxsd, - pminub, pminuw, pminud, - pmaxub, pmaxuw, pmaxud, - pabsb, pabsw, pabsd, - pmuludq, pmuldq, - pmulld, - cvtps2dq, cvtpd2dq, - cvtdq2ps, cvtdq2pd, - ldmxcsr, stmxcsr, withMXCSR, - X86.haddps, X86.haddpd, X86.dpps, X86.dppd, - roundss, X86.roundps, roundsd, X86.roundpd, - absss, abssd, absps, abspd, - X86.psllw128, X86.pslld128, X86.psllq128, - X86.psrlw128, X86.psrld128, X86.psrlq128, - X86.psraw128, X86.psrad128, - X86.paddsb128, X86.paddsw128, X86.paddusb128, X86.paddusw128, - X86.psubsb128, X86.psubsw128, X86.psubusb128, X86.psubusw128, - ) where - -import qualified LLVM.Extra.Extension.X86Auto as X86 -import qualified LLVM.Extra.Extension as Ext -import LLVM.Extra.Extension.X86Auto ( - V2Double, V4Float, - V2Int64, V2Word64, - V4Int32, V4Word32, - V8Int16, V8Word16, - V16Int8, V16Word8, - ) -import LLVM.Extra.ExtensionCheck.X86 - (sse1, sse2, sse41, sse42, ) - -import qualified LLVM.Extra.ArithmeticPrivate as A -import qualified LLVM.Core as LLVM -import LLVM.Core - (Value, Vector, valueOf, constOf, vector, - CodeGenFunction, FPPredicate, ) - -import qualified Type.Data.Num.Decimal as TypeNum - -import qualified Data.NonEmpty.Class as NonEmptyC -import qualified Data.Empty as Empty -import Data.NonEmpty ((!:), ) - -import Data.Bits (clearBit, complement, ) -import Data.Word (Word8, Word32, Word64, ) - -import qualified Control.Monad.HT as M -import Control.Monad.HT ((<=<), ) -import Control.Applicative (pure, ) - -import Foreign.Ptr (Ptr, ) - -import Prelude2010 -import Prelude () - - -switchFPPred :: - (Num i, LLVM.IsConst i, LLVM.IsInteger i, LLVM.IsPrimitive i, - LLVM.IsFirstClass v, - TypeNum.Positive n, - LLVM.IsSized v, LLVM.IsSized (Vector n i), - LLVM.SizeOf v ~ LLVM.SizeOf (Vector n i)) => - (Value v -> Value v -> Value Word8 -> CodeGenFunction r (Value v)) -> - FPPredicate -> Value v -> Value v -> CodeGenFunction r (Value (Vector n i)) -switchFPPred g p x y = - let f i x0 y0 = LLVM.bitcast =<< g x0 y0 (valueOf i) - in case p of - LLVM.FPFalse -> return (LLVM.value LLVM.zero) - LLVM.FPOEQ -> f 0 x y - LLVM.FPOGT -> f 1 y x - LLVM.FPOGE -> f 2 y x - LLVM.FPOLT -> f 1 x y - LLVM.FPOLE -> f 2 x y - LLVM.FPONE -> M.liftJoin2 A.and (f 7 x y) (f 4 x y) - LLVM.FPORD -> f 7 x y - LLVM.FPUNO -> f 3 x y - LLVM.FPUEQ -> M.liftJoin2 A.or (f 3 x y) (f 0 x y) - LLVM.FPUGT -> f 6 x y - LLVM.FPUGE -> f 5 x y - LLVM.FPULT -> f 6 y x - LLVM.FPULE -> f 5 y x - LLVM.FPUNE -> f 4 x y - LLVM.FPT -> return (valueOf $ pure (-1)) - -cmpss :: Ext.T (FPPredicate -> V4Float -> V4Float -> CodeGenFunction r V4Int32) -cmpss = fmap switchFPPred X86.cmpss - -cmpps :: Ext.T (FPPredicate -> V4Float -> V4Float -> CodeGenFunction r V4Int32) -cmpps = fmap switchFPPred X86.cmpps - -cmpsd :: Ext.T (FPPredicate -> V2Double -> V2Double -> CodeGenFunction r V2Int64) -cmpsd = fmap switchFPPred X86.cmpsd - -cmppd :: Ext.T (FPPredicate -> V2Double -> V2Double -> CodeGenFunction r V2Int64) -cmppd = fmap switchFPPred X86.cmppd - -cmpps256 :: Ext.T (FPPredicate -> X86.V8Float -> X86.V8Float -> CodeGenFunction r X86.V8Int32) -cmpps256 = fmap switchFPPred X86.cmpps256 - -cmppd256 :: Ext.T (FPPredicate -> X86.V4Double -> X86.V4Double -> CodeGenFunction r X86.V4Int64) -cmppd256 = fmap switchFPPred X86.cmppd256 - - -pcmpgtb :: Ext.T (V16Int8 -> V16Int8 -> CodeGenFunction r V16Int8) -pcmpgtb = Ext.intrinsic sse2 "pcmpgt.b" - -pcmpgtw :: Ext.T (V8Int16 -> V8Int16 -> CodeGenFunction r V8Int16) -pcmpgtw = Ext.intrinsic sse2 "pcmpgt.w" - -pcmpgtd :: Ext.T (V4Int32 -> V4Int32 -> CodeGenFunction r V4Int32) -pcmpgtd = Ext.intrinsic sse2 "pcmpgt.d" - -pcmpgtq :: Ext.T (V2Int64 -> V2Int64 -> CodeGenFunction r V2Int64) -pcmpgtq = Ext.intrinsic sse42 "pcmpgtq" - - -pcmpuFromPcmp :: - (TypeNum.Positive n, - LLVM.IsPrimitive s, - LLVM.IsPrimitive u, LLVM.IsArithmetic u, LLVM.IsConst u, - Bounded u, Integral u, - LLVM.IsSized (Vector n s), LLVM.IsSized (Vector n u), - LLVM.SizeOf (Vector n s) ~ LLVM.SizeOf (Vector n u)) => - Ext.T (Value (Vector n s) -> Value (Vector n s) -> CodeGenFunction r (Value (Vector n s))) -> - Ext.T (Value (Vector n u) -> Value (Vector n u) -> CodeGenFunction r (Value (Vector n u))) -pcmpuFromPcmp pcmp = - Ext.with pcmp $ \cmp x y -> do - let offset = valueOf $ pure (1 + div maxBound 2) - xa <- LLVM.bitcast =<< A.sub x offset - ya <- LLVM.bitcast =<< A.sub y offset - LLVM.bitcast =<< cmp xa ya - -pcmpugtb :: Ext.T (V16Word8 -> V16Word8 -> CodeGenFunction r V16Word8) -pcmpugtb = pcmpuFromPcmp pcmpgtb - -pcmpugtw :: Ext.T (V8Word16 -> V8Word16 -> CodeGenFunction r V8Word16) -pcmpugtw = pcmpuFromPcmp pcmpgtw - -pcmpugtd :: Ext.T (V4Word32 -> V4Word32 -> CodeGenFunction r V4Word32) -pcmpugtd = pcmpuFromPcmp pcmpgtd - -pcmpugtq :: Ext.T (V2Word64 -> V2Word64 -> CodeGenFunction r V2Word64) -pcmpugtq = pcmpuFromPcmp pcmpgtq - - -pminsb, pmaxsb :: Ext.T (V16Int8 -> V16Int8 -> CodeGenFunction r V16Int8) -pminsb = X86.pminsb128 -pmaxsb = X86.pmaxsb128 - -pminsw, pmaxsw :: Ext.T (V8Int16 -> V8Int16 -> CodeGenFunction r V8Int16) -pminsw = X86.pminsw128 -pmaxsw = X86.pmaxsw128 - -pminsd, pmaxsd :: Ext.T (V4Int32 -> V4Int32 -> CodeGenFunction r V4Int32) -pminsd = X86.pminsd128 -pmaxsd = X86.pmaxsd128 - - -pminub, pmaxub :: Ext.T (V16Word8 -> V16Word8 -> CodeGenFunction r V16Word8) -pminub = X86.pminub128 -pmaxub = X86.pmaxub128 - -pminuw, pmaxuw :: Ext.T (V8Word16 -> V8Word16 -> CodeGenFunction r V8Word16) -pminuw = X86.pminuw128 -pmaxuw = X86.pmaxuw128 - -pminud, pmaxud :: Ext.T (V4Word32 -> V4Word32 -> CodeGenFunction r V4Word32) -pminud = X86.pminud128 -pmaxud = X86.pmaxud128 - - -pabsb :: Ext.T (V16Int8 -> CodeGenFunction r V16Int8) -pabsb = X86.pabsb128 - -pabsw :: Ext.T (V8Int16 -> CodeGenFunction r V8Int16) -pabsw = X86.pabsw128 - -pabsd :: Ext.T (V4Int32 -> CodeGenFunction r V4Int32) -pabsd = X86.pabsd128 - - -pmuludq :: Ext.T (V4Word32 -> V4Word32 -> CodeGenFunction r V2Word64) -pmuludq = X86.pmuludq128 - -pmuldq :: Ext.T (V4Int32 -> V4Int32 -> CodeGenFunction r V2Int64) -pmuldq = X86.pmuldq128 - -pmulld :: Ext.T (V4Word32 -> V4Word32 -> CodeGenFunction r V4Word32) -pmulld = Ext.wrap sse41 LLVM.mul --- pmulld = Ext.intrinsic sse41 "pmulld" - - -cvtps2dq :: Ext.T (V4Float -> CodeGenFunction r V4Int32) -cvtps2dq = X86.cvtps2dq - --- | the upper two integers are set to zero, there is no instruction that converts to Int64 -cvtpd2dq :: Ext.T (V2Double -> CodeGenFunction r V4Int32) -cvtpd2dq = X86.cvtpd2dq - - -cvtdq2ps :: Ext.T (V4Int32 -> CodeGenFunction r V4Float) -cvtdq2ps = X86.cvtdq2ps - --- | the upper two integers are ignored, there is no instruction that converts from Int64 -cvtdq2pd :: Ext.T (V4Int32 -> CodeGenFunction r V2Double) -cvtdq2pd = X86.cvtdq2pd - - -valueUnit :: Value () -> () -valueUnit _ = () - -{- | -MXCSR is not really supported by LLVM-2.6. -LLVM does not know about the dependency of all floating point operations -on this status register. --} -ldmxcsr :: Ext.T (Value (Ptr Word32) -> CodeGenFunction r ()) -ldmxcsr = - fmap (fmap valueUnit .) $ Ext.intrinsicAttr [] sse1 "ldmxcsr" - -stmxcsr :: Ext.T (Value (Ptr Word32) -> CodeGenFunction r ()) -stmxcsr = - fmap (fmap valueUnit .) $ Ext.intrinsicAttr [] sse1 "stmxcsr" - -withMXCSR :: Word32 -> Ext.T (CodeGenFunction r a -> CodeGenFunction r a) -withMXCSR mxcsr = - Ext.with2 ldmxcsr stmxcsr $ \ ld st f -> do - mxcsrOld <- LLVM.alloca - st mxcsrOld - mxcsrFloor <- LLVM.alloca - LLVM.store (valueOf $ mxcsr) mxcsrFloor -{- unfortunately, createGlobal is a function CodeGenModule monad - mxcsrFloor <- - LLVM.createGlobal True LLVM.InternalLinkage mxcsr --} - ld mxcsrFloor - r <- f - ld mxcsrOld - return r - -{- -[maxsd, minsd, maxpd, minpd] = - map (Ext.intrinsic sse2) - ["max.ss", "min.ss", "max.ps", "min.ps"] --} - -roundss :: Ext.T (V4Float -> Value Word32 -> CodeGenFunction r V4Float) -roundss = - fmap (\f -> f (LLVM.value LLVM.undef)) X86.roundss - -roundsd :: Ext.T (V2Double -> Value Word32 -> CodeGenFunction r V2Double) -roundsd = - fmap (\f -> f (LLVM.value LLVM.undef)) X86.roundsd - - - -{- -Not an LLVM intrinsic but implementation specific: -We expect that floating point values are in IEEE format -and thus the most significant bit is the sign. -The absolute value can be computed very efficiently by clearing the sign bit. -Actually, LLVM's codegen implements neg by an XOR on the sign bit. --} -absss :: Ext.T (V4Float -> CodeGenFunction r V4Float) -absss = - Ext.wrap sse1 $ - LLVM.bitcast - <=< A.and (LLVM.valueOf $ vector $ - (flip clearBit 31 $ complement 0) !: NonEmptyC.repeat (complement 0) - :: V4Word32) - <=< LLVM.bitcast - -{- -This function works on a single Float, -but I like to do the masking in an XMM register -because usually the value is there anyway. - -absss = - flip LLVM.extractelement (valueOf 0) - . flip asTypeOf (undefined :: V4Float) - <=< LLVM.bitcast --- <=< A.and (LLVM.value $ constVector [constOf 0x7FFFFFFF] :: V4Word32) --- <=< A.and (LLVM.value $ constVector [constOf 0x7FFFFFFF, LLVM.undef, LLVM.undef, LLVM.undef] :: V4Word32) - <=< A.and (LLVM.value $ constVector [constOf 0x7FFFFFFF, LLVM.zero, LLVM.zero, LLVM.zero] :: V4Word32) - <=< LLVM.bitcast - . flip asTypeOf (undefined :: V4Float) - <=< flip (LLVM.insertelement (LLVM.value LLVM.undef)) (valueOf 0) --} -{- This moves the value to a general purpose register and performs the bit masking there -absss = - LLVM.bitcast - <=< A.and (valueOf 0x7FFFFFFF :: Value Word32) - <=< LLVM.bitcast --} - -abssd :: Ext.T (V2Double -> CodeGenFunction r V2Double) -abssd = - Ext.wrap sse2 $ - LLVM.bitcast - <=< A.and (LLVM.valueOf $ vector $ - (flip clearBit 63 $ complement 0) !: complement 0 !: Empty.Cons - :: V2Word64) - <=< LLVM.bitcast - - -mask :: - (TypeNum.Positive n, LLVM.IsConst w, LLVM.IsPrimitive w, LLVM.IsInteger w) => - w -> Value (Vector n w) -> CodeGenFunction r (Value (Vector n w)) -mask x = - A.and (LLVM.valueOf $ pure x) - -absps :: - (TypeNum.Positive n) => - Ext.T (Value (Vector n Float) -> CodeGenFunction r (Value (Vector n Float))) -absps = - Ext.wrap sse1 $ - LLVM.bitcastElements - <=< mask (flip clearBit 31 $ complement 0 :: Word32) - <=< LLVM.bitcastElements - -abspd :: - (TypeNum.Positive n) => - Ext.T (Value (Vector n Double) -> CodeGenFunction r (Value (Vector n Double))) -abspd = - Ext.wrap sse2 $ - LLVM.bitcastElements - <=< mask (flip clearBit 63 $ complement 0 :: Word64) - <=< LLVM.bitcastElements - -{- | -cumulative sum: -@(a,b,c,d) -> (a,a+b,a+b+c,a+b+c+d)@ - -I try to cleverly use horizontal add, -but the generic version in the Vector module is better. --} -_cumulate1s :: Ext.T (V4Float -> CodeGenFunction r V4Float) -_cumulate1s = Ext.with X86.haddps $ \haddp x -> do - y <- haddp x (LLVM.value LLVM.undef) - z <- LLVM.shufflevector x y $ - constOf $ vector $ 0!:4!:2!:5!:Empty.Cons - offset <- LLVM.shufflevector y (LLVM.value LLVM.zero) $ - constOf $ vector $ 4!:5!:0!:0!:Empty.Cons - A.add z offset diff --git a/src/LLVM/Extra/Extension/X86Auto.hs b/src/LLVM/Extra/Extension/X86Auto.hs deleted file mode 100644 --- a/src/LLVM/Extra/Extension/X86Auto.hs +++ /dev/null @@ -1,2699 +0,0 @@ -{- Do not edit! This file was created with the PrepareIntrinsics tool. -} -module LLVM.Extra.Extension.X86Auto where - -import qualified LLVM.Extra.Extension as Ext -import qualified LLVM.Extra.ExtensionCheck.X86 as ExtX86 -import qualified LLVM.Core as LLVM -import qualified Type.Data.Num.Decimal as TypeNum -import qualified Data.Int as I -import qualified Data.Word as W -import Foreign.Ptr (Ptr, ) - -type MMX = LLVM.Value (LLVM.Vector TypeNum.D8 W.Word8) -type V16Float = LLVM.Value (LLVM.Vector TypeNum.D16 Float) -type V16Int16 = LLVM.Value (LLVM.Vector TypeNum.D16 I.Int16) -type V16Int32 = LLVM.Value (LLVM.Vector TypeNum.D16 I.Int32) -type V16Int8 = LLVM.Value (LLVM.Vector TypeNum.D16 I.Int8) -type V16Word16 = LLVM.Value (LLVM.Vector TypeNum.D16 W.Word16) -type V16Word8 = LLVM.Value (LLVM.Vector TypeNum.D16 W.Word8) -type V2Double = LLVM.Value (LLVM.Vector TypeNum.D2 Double) -type V2Int64 = LLVM.Value (LLVM.Vector TypeNum.D2 I.Int64) -type V2Word64 = LLVM.Value (LLVM.Vector TypeNum.D2 W.Word64) -type V32Int16 = LLVM.Value (LLVM.Vector TypeNum.D32 I.Int16) -type V32Int8 = LLVM.Value (LLVM.Vector TypeNum.D32 I.Int8) -type V32Word8 = LLVM.Value (LLVM.Vector TypeNum.D32 W.Word8) -type V4Double = LLVM.Value (LLVM.Vector TypeNum.D4 Double) -type V4Float = LLVM.Value (LLVM.Vector TypeNum.D4 Float) -type V4Int32 = LLVM.Value (LLVM.Vector TypeNum.D4 I.Int32) -type V4Int64 = LLVM.Value (LLVM.Vector TypeNum.D4 I.Int64) -type V4Word32 = LLVM.Value (LLVM.Vector TypeNum.D4 W.Word32) -type V4Word64 = LLVM.Value (LLVM.Vector TypeNum.D4 W.Word64) -type V64Int8 = LLVM.Value (LLVM.Vector TypeNum.D64 I.Int8) -type V8Double = LLVM.Value (LLVM.Vector TypeNum.D8 Double) -type V8Float = LLVM.Value (LLVM.Vector TypeNum.D8 Float) -type V8Int16 = LLVM.Value (LLVM.Vector TypeNum.D8 I.Int16) -type V8Int32 = LLVM.Value (LLVM.Vector TypeNum.D8 I.Int32) -type V8Int64 = LLVM.Value (LLVM.Vector TypeNum.D8 I.Int64) -type V8Word16 = LLVM.Value (LLVM.Vector TypeNum.D8 W.Word16) -type V8Word32 = LLVM.Value (LLVM.Vector TypeNum.D8 W.Word32) - -{- -readeflags_u32 :: Ext.T (LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -readeflags_u32 = Ext.intrinsic ExtX86.flags "read.u32" - -readeflags_u64 :: Ext.T (LLVM.CodeGenFunction r (LLVM.Value I.Int64)) -readeflags_u64 = Ext.intrinsic ExtX86.flags "read.u64" - -writeeflags_u32 :: Ext.T (LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (LLVM.Value ())) -writeeflags_u32 = Ext.intrinsic ExtX86.flags "write.u32" - -writeeflags_u64 :: Ext.T (LLVM.Value I.Int64 -> LLVM.CodeGenFunction r (LLVM.Value ())) -writeeflags_u64 = Ext.intrinsic ExtX86.flags "write.u64" - -rdtsc :: Ext.T (LLVM.CodeGenFunction r (LLVM.Value I.Int64)) -rdtsc = Ext.intrinsic ExtX86. "int.x86.rdtsc" - -rdtscp :: Ext.T (LLVM.Value (Ptr ()) -> LLVM.CodeGenFunction r (LLVM.Value I.Int64)) -rdtscp = Ext.intrinsic ExtX86. "int.x86.rdtscp" - -rdpmc :: Ext.T (LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (LLVM.Value I.Int64)) -rdpmc = Ext.intrinsic ExtX86. "int.x86.rdpmc" --} - -pavgusb :: Ext.T (MMX -> MMX -> LLVM.CodeGenFunction r (MMX)) -pavgusb = Ext.intrinsic ExtX86.amd3dnow "pavgusb" - -pf2id :: Ext.T (MMX -> LLVM.CodeGenFunction r (MMX)) -pf2id = Ext.intrinsic ExtX86.amd3dnow "pf2id" - -pfacc :: Ext.T (MMX -> MMX -> LLVM.CodeGenFunction r (MMX)) -pfacc = Ext.intrinsic ExtX86.amd3dnow "pfacc" - -pfadd :: Ext.T (MMX -> MMX -> LLVM.CodeGenFunction r (MMX)) -pfadd = Ext.intrinsic ExtX86.amd3dnow "pfadd" - -pfcmpeq :: Ext.T (MMX -> MMX -> LLVM.CodeGenFunction r (MMX)) -pfcmpeq = Ext.intrinsic ExtX86.amd3dnow "pfcmpeq" - -pfcmpge :: Ext.T (MMX -> MMX -> LLVM.CodeGenFunction r (MMX)) -pfcmpge = Ext.intrinsic ExtX86.amd3dnow "pfcmpge" - -pfcmpgt :: Ext.T (MMX -> MMX -> LLVM.CodeGenFunction r (MMX)) -pfcmpgt = Ext.intrinsic ExtX86.amd3dnow "pfcmpgt" - -pfmax :: Ext.T (MMX -> MMX -> LLVM.CodeGenFunction r (MMX)) -pfmax = Ext.intrinsic ExtX86.amd3dnow "pfmax" - -pfmin :: Ext.T (MMX -> MMX -> LLVM.CodeGenFunction r (MMX)) -pfmin = Ext.intrinsic ExtX86.amd3dnow "pfmin" - -pfmul :: Ext.T (MMX -> MMX -> LLVM.CodeGenFunction r (MMX)) -pfmul = Ext.intrinsic ExtX86.amd3dnow "pfmul" - -pfrcp :: Ext.T (MMX -> LLVM.CodeGenFunction r (MMX)) -pfrcp = Ext.intrinsic ExtX86.amd3dnow "pfrcp" - -pfrcpit1 :: Ext.T (MMX -> MMX -> LLVM.CodeGenFunction r (MMX)) -pfrcpit1 = Ext.intrinsic ExtX86.amd3dnow "pfrcpit1" - -pfrcpit2 :: Ext.T (MMX -> MMX -> LLVM.CodeGenFunction r (MMX)) -pfrcpit2 = Ext.intrinsic ExtX86.amd3dnow "pfrcpit2" - -pfrsqrt :: Ext.T (MMX -> LLVM.CodeGenFunction r (MMX)) -pfrsqrt = Ext.intrinsic ExtX86.amd3dnow "pfrsqrt" - -pfrsqit1 :: Ext.T (MMX -> MMX -> LLVM.CodeGenFunction r (MMX)) -pfrsqit1 = Ext.intrinsic ExtX86.amd3dnow "pfrsqit1" - -pfsub :: Ext.T (MMX -> MMX -> LLVM.CodeGenFunction r (MMX)) -pfsub = Ext.intrinsic ExtX86.amd3dnow "pfsub" - -pfsubr :: Ext.T (MMX -> MMX -> LLVM.CodeGenFunction r (MMX)) -pfsubr = Ext.intrinsic ExtX86.amd3dnow "pfsubr" - -pi2fd :: Ext.T (MMX -> LLVM.CodeGenFunction r (MMX)) -pi2fd = Ext.intrinsic ExtX86.amd3dnow "pi2fd" - -pmulhrw :: Ext.T (MMX -> MMX -> LLVM.CodeGenFunction r (MMX)) -pmulhrw = Ext.intrinsic ExtX86.amd3dnow "pmulhrw" - -pf2iw :: Ext.T (MMX -> LLVM.CodeGenFunction r (MMX)) -pf2iw = Ext.intrinsic ExtX86.amd3dnowa "pf2iw" - -pfnacc :: Ext.T (MMX -> MMX -> LLVM.CodeGenFunction r (MMX)) -pfnacc = Ext.intrinsic ExtX86.amd3dnowa "pfnacc" - -pfpnacc :: Ext.T (MMX -> MMX -> LLVM.CodeGenFunction r (MMX)) -pfpnacc = Ext.intrinsic ExtX86.amd3dnowa "pfpnacc" - -pi2fw :: Ext.T (MMX -> LLVM.CodeGenFunction r (MMX)) -pi2fw = Ext.intrinsic ExtX86.amd3dnowa "pi2fw" - -addss :: Ext.T (V4Float -> V4Float -> LLVM.CodeGenFunction r (V4Float)) -addss = Ext.intrinsic ExtX86.sse1 "add.ss" - -subss :: Ext.T (V4Float -> V4Float -> LLVM.CodeGenFunction r (V4Float)) -subss = Ext.intrinsic ExtX86.sse1 "sub.ss" - -mulss :: Ext.T (V4Float -> V4Float -> LLVM.CodeGenFunction r (V4Float)) -mulss = Ext.intrinsic ExtX86.sse1 "mul.ss" - -divss :: Ext.T (V4Float -> V4Float -> LLVM.CodeGenFunction r (V4Float)) -divss = Ext.intrinsic ExtX86.sse1 "div.ss" - -sqrtss :: Ext.T (V4Float -> LLVM.CodeGenFunction r (V4Float)) -sqrtss = Ext.intrinsic ExtX86.sse1 "sqrt.ss" - -sqrtps :: Ext.T (V4Float -> LLVM.CodeGenFunction r (V4Float)) -sqrtps = Ext.intrinsic ExtX86.sse1 "sqrt.ps" - -rcpss :: Ext.T (V4Float -> LLVM.CodeGenFunction r (V4Float)) -rcpss = Ext.intrinsic ExtX86.sse1 "rcp.ss" - -rcpps :: Ext.T (V4Float -> LLVM.CodeGenFunction r (V4Float)) -rcpps = Ext.intrinsic ExtX86.sse1 "rcp.ps" - -rsqrtss :: Ext.T (V4Float -> LLVM.CodeGenFunction r (V4Float)) -rsqrtss = Ext.intrinsic ExtX86.sse1 "rsqrt.ss" - -rsqrtps :: Ext.T (V4Float -> LLVM.CodeGenFunction r (V4Float)) -rsqrtps = Ext.intrinsic ExtX86.sse1 "rsqrt.ps" - -minss :: Ext.T (V4Float -> V4Float -> LLVM.CodeGenFunction r (V4Float)) -minss = Ext.intrinsic ExtX86.sse1 "min.ss" - -minps :: Ext.T (V4Float -> V4Float -> LLVM.CodeGenFunction r (V4Float)) -minps = Ext.intrinsic ExtX86.sse1 "min.ps" - -maxss :: Ext.T (V4Float -> V4Float -> LLVM.CodeGenFunction r (V4Float)) -maxss = Ext.intrinsic ExtX86.sse1 "max.ss" - -maxps :: Ext.T (V4Float -> V4Float -> LLVM.CodeGenFunction r (V4Float)) -maxps = Ext.intrinsic ExtX86.sse1 "max.ps" - -cmpss :: Ext.T (V4Float -> V4Float -> LLVM.Value W.Word8 -> LLVM.CodeGenFunction r (V4Float)) -cmpss = Ext.intrinsic ExtX86.sse1 "cmp.ss" - -cmpps :: Ext.T (V4Float -> V4Float -> LLVM.Value W.Word8 -> LLVM.CodeGenFunction r (V4Float)) -cmpps = Ext.intrinsic ExtX86.sse1 "cmp.ps" - -comieq :: Ext.T (V4Float -> V4Float -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -comieq = Ext.intrinsic ExtX86.sse1 "comieq.ss" - -comilt :: Ext.T (V4Float -> V4Float -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -comilt = Ext.intrinsic ExtX86.sse1 "comilt.ss" - -comile :: Ext.T (V4Float -> V4Float -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -comile = Ext.intrinsic ExtX86.sse1 "comile.ss" - -comigt :: Ext.T (V4Float -> V4Float -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -comigt = Ext.intrinsic ExtX86.sse1 "comigt.ss" - -comige :: Ext.T (V4Float -> V4Float -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -comige = Ext.intrinsic ExtX86.sse1 "comige.ss" - -comineq :: Ext.T (V4Float -> V4Float -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -comineq = Ext.intrinsic ExtX86.sse1 "comineq.ss" - -ucomieq :: Ext.T (V4Float -> V4Float -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -ucomieq = Ext.intrinsic ExtX86.sse1 "ucomieq.ss" - -ucomilt :: Ext.T (V4Float -> V4Float -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -ucomilt = Ext.intrinsic ExtX86.sse1 "ucomilt.ss" - -ucomile :: Ext.T (V4Float -> V4Float -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -ucomile = Ext.intrinsic ExtX86.sse1 "ucomile.ss" - -ucomigt :: Ext.T (V4Float -> V4Float -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -ucomigt = Ext.intrinsic ExtX86.sse1 "ucomigt.ss" - -ucomige :: Ext.T (V4Float -> V4Float -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -ucomige = Ext.intrinsic ExtX86.sse1 "ucomige.ss" - -ucomineq :: Ext.T (V4Float -> V4Float -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -ucomineq = Ext.intrinsic ExtX86.sse1 "ucomineq.ss" - -cvtss2si :: Ext.T (V4Float -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -cvtss2si = Ext.intrinsic ExtX86.sse1 "cvtss2si" - -cvtss2si64 :: Ext.T (V4Float -> LLVM.CodeGenFunction r (LLVM.Value I.Int64)) -cvtss2si64 = Ext.intrinsic ExtX86.sse1 "cvtss2si64" - -cvttss2si :: Ext.T (V4Float -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -cvttss2si = Ext.intrinsic ExtX86.sse1 "cvttss2si" - -cvttss2si64 :: Ext.T (V4Float -> LLVM.CodeGenFunction r (LLVM.Value I.Int64)) -cvttss2si64 = Ext.intrinsic ExtX86.sse1 "cvttss2si64" - -cvtsi2ss :: Ext.T (V4Float -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V4Float)) -cvtsi2ss = Ext.intrinsic ExtX86.sse1 "cvtsi2ss" - -cvtsi642ss :: Ext.T (V4Float -> LLVM.Value I.Int64 -> LLVM.CodeGenFunction r (V4Float)) -cvtsi642ss = Ext.intrinsic ExtX86.sse1 "cvtsi642ss" - -cvtps2pi :: Ext.T (V4Float -> LLVM.CodeGenFunction r (MMX)) -cvtps2pi = Ext.intrinsic ExtX86.sse1 "cvtps2pi" - -cvttps2pi :: Ext.T (V4Float -> LLVM.CodeGenFunction r (MMX)) -cvttps2pi = Ext.intrinsic ExtX86.sse1 "cvttps2pi" - -cvtpi2ps :: Ext.T (V4Float -> MMX -> LLVM.CodeGenFunction r (V4Float)) -cvtpi2ps = Ext.intrinsic ExtX86.sse1 "cvtpi2ps" - -storeups :: Ext.T (LLVM.Value (Ptr ()) -> V4Float -> LLVM.CodeGenFunction r (LLVM.Value ())) -storeups = Ext.intrinsic ExtX86.sse1 "storeu.ps" - -sfence :: Ext.T (LLVM.CodeGenFunction r (LLVM.Value ())) -sfence = Ext.intrinsic ExtX86.sse1 "sfence" - -movmskps :: Ext.T (V4Float -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -movmskps = Ext.intrinsic ExtX86.sse1 "movmsk.ps" - -addsd :: Ext.T (V2Double -> V2Double -> LLVM.CodeGenFunction r (V2Double)) -addsd = Ext.intrinsic ExtX86.sse2 "add.sd" - -subsd :: Ext.T (V2Double -> V2Double -> LLVM.CodeGenFunction r (V2Double)) -subsd = Ext.intrinsic ExtX86.sse2 "sub.sd" - -mulsd :: Ext.T (V2Double -> V2Double -> LLVM.CodeGenFunction r (V2Double)) -mulsd = Ext.intrinsic ExtX86.sse2 "mul.sd" - -divsd :: Ext.T (V2Double -> V2Double -> LLVM.CodeGenFunction r (V2Double)) -divsd = Ext.intrinsic ExtX86.sse2 "div.sd" - -sqrtsd :: Ext.T (V2Double -> LLVM.CodeGenFunction r (V2Double)) -sqrtsd = Ext.intrinsic ExtX86.sse2 "sqrt.sd" - -sqrtpd :: Ext.T (V2Double -> LLVM.CodeGenFunction r (V2Double)) -sqrtpd = Ext.intrinsic ExtX86.sse2 "sqrt.pd" - -minsd :: Ext.T (V2Double -> V2Double -> LLVM.CodeGenFunction r (V2Double)) -minsd = Ext.intrinsic ExtX86.sse2 "min.sd" - -minpd :: Ext.T (V2Double -> V2Double -> LLVM.CodeGenFunction r (V2Double)) -minpd = Ext.intrinsic ExtX86.sse2 "min.pd" - -maxsd :: Ext.T (V2Double -> V2Double -> LLVM.CodeGenFunction r (V2Double)) -maxsd = Ext.intrinsic ExtX86.sse2 "max.sd" - -maxpd :: Ext.T (V2Double -> V2Double -> LLVM.CodeGenFunction r (V2Double)) -maxpd = Ext.intrinsic ExtX86.sse2 "max.pd" - -cmpsd :: Ext.T (V2Double -> V2Double -> LLVM.Value W.Word8 -> LLVM.CodeGenFunction r (V2Double)) -cmpsd = Ext.intrinsic ExtX86.sse2 "cmp.sd" - -cmppd :: Ext.T (V2Double -> V2Double -> LLVM.Value W.Word8 -> LLVM.CodeGenFunction r (V2Double)) -cmppd = Ext.intrinsic ExtX86.sse2 "cmp.pd" - -comisdeq :: Ext.T (V2Double -> V2Double -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -comisdeq = Ext.intrinsic ExtX86.sse2 "comieq.sd" - -comisdlt :: Ext.T (V2Double -> V2Double -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -comisdlt = Ext.intrinsic ExtX86.sse2 "comilt.sd" - -comisdle :: Ext.T (V2Double -> V2Double -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -comisdle = Ext.intrinsic ExtX86.sse2 "comile.sd" - -comisdgt :: Ext.T (V2Double -> V2Double -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -comisdgt = Ext.intrinsic ExtX86.sse2 "comigt.sd" - -comisdge :: Ext.T (V2Double -> V2Double -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -comisdge = Ext.intrinsic ExtX86.sse2 "comige.sd" - -comisdneq :: Ext.T (V2Double -> V2Double -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -comisdneq = Ext.intrinsic ExtX86.sse2 "comineq.sd" - -ucomisdeq :: Ext.T (V2Double -> V2Double -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -ucomisdeq = Ext.intrinsic ExtX86.sse2 "ucomieq.sd" - -ucomisdlt :: Ext.T (V2Double -> V2Double -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -ucomisdlt = Ext.intrinsic ExtX86.sse2 "ucomilt.sd" - -ucomisdle :: Ext.T (V2Double -> V2Double -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -ucomisdle = Ext.intrinsic ExtX86.sse2 "ucomile.sd" - -ucomisdgt :: Ext.T (V2Double -> V2Double -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -ucomisdgt = Ext.intrinsic ExtX86.sse2 "ucomigt.sd" - -ucomisdge :: Ext.T (V2Double -> V2Double -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -ucomisdge = Ext.intrinsic ExtX86.sse2 "ucomige.sd" - -ucomisdneq :: Ext.T (V2Double -> V2Double -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -ucomisdneq = Ext.intrinsic ExtX86.sse2 "ucomineq.sd" - -paddsb128 :: Ext.T (V16Int8 -> V16Int8 -> LLVM.CodeGenFunction r (V16Int8)) -paddsb128 = Ext.intrinsic ExtX86.sse2 "padds.b" - -paddsw128 :: Ext.T (V8Int16 -> V8Int16 -> LLVM.CodeGenFunction r (V8Int16)) -paddsw128 = Ext.intrinsic ExtX86.sse2 "padds.w" - -paddusb128 :: Ext.T (V16Word8 -> V16Word8 -> LLVM.CodeGenFunction r (V16Word8)) -paddusb128 = Ext.intrinsic ExtX86.sse2 "paddus.b" - -paddusw128 :: Ext.T (V8Word16 -> V8Word16 -> LLVM.CodeGenFunction r (V8Word16)) -paddusw128 = Ext.intrinsic ExtX86.sse2 "paddus.w" - -psubsb128 :: Ext.T (V16Int8 -> V16Int8 -> LLVM.CodeGenFunction r (V16Int8)) -psubsb128 = Ext.intrinsic ExtX86.sse2 "psubs.b" - -psubsw128 :: Ext.T (V8Int16 -> V8Int16 -> LLVM.CodeGenFunction r (V8Int16)) -psubsw128 = Ext.intrinsic ExtX86.sse2 "psubs.w" - -psubusb128 :: Ext.T (V16Word8 -> V16Word8 -> LLVM.CodeGenFunction r (V16Word8)) -psubusb128 = Ext.intrinsic ExtX86.sse2 "psubus.b" - -psubusw128 :: Ext.T (V8Word16 -> V8Word16 -> LLVM.CodeGenFunction r (V8Word16)) -psubusw128 = Ext.intrinsic ExtX86.sse2 "psubus.w" - -pmulhuw128 :: Ext.T (V8Word16 -> V8Word16 -> LLVM.CodeGenFunction r (V8Word16)) -pmulhuw128 = Ext.intrinsic ExtX86.sse2 "pmulhu.w" - -pmulhw128 :: Ext.T (V8Int16 -> V8Int16 -> LLVM.CodeGenFunction r (V8Int16)) -pmulhw128 = Ext.intrinsic ExtX86.sse2 "pmulh.w" - -pmuludq128 :: Ext.T (V4Word32 -> V4Word32 -> LLVM.CodeGenFunction r (V2Word64)) -pmuludq128 = Ext.intrinsic ExtX86.sse2 "pmulu.dq" - -pmaddwd128 :: Ext.T (V8Int16 -> V8Int16 -> LLVM.CodeGenFunction r (V4Int32)) -pmaddwd128 = Ext.intrinsic ExtX86.sse2 "pmadd.wd" - -pavgb128 :: Ext.T (V16Int8 -> V16Int8 -> LLVM.CodeGenFunction r (V16Int8)) -pavgb128 = Ext.intrinsic ExtX86.sse2 "pavg.b" - -pavgw128 :: Ext.T (V8Int16 -> V8Int16 -> LLVM.CodeGenFunction r (V8Int16)) -pavgw128 = Ext.intrinsic ExtX86.sse2 "pavg.w" - -pmaxub128 :: Ext.T (V16Word8 -> V16Word8 -> LLVM.CodeGenFunction r (V16Word8)) -pmaxub128 = Ext.intrinsic ExtX86.sse2 "pmaxu.b" - -pmaxsw128 :: Ext.T (V8Int16 -> V8Int16 -> LLVM.CodeGenFunction r (V8Int16)) -pmaxsw128 = Ext.intrinsic ExtX86.sse2 "pmaxs.w" - -pminub128 :: Ext.T (V16Word8 -> V16Word8 -> LLVM.CodeGenFunction r (V16Word8)) -pminub128 = Ext.intrinsic ExtX86.sse2 "pminu.b" - -pminsw128 :: Ext.T (V8Int16 -> V8Int16 -> LLVM.CodeGenFunction r (V8Int16)) -pminsw128 = Ext.intrinsic ExtX86.sse2 "pmins.w" - -psadbw128 :: Ext.T (V16Int8 -> V16Int8 -> LLVM.CodeGenFunction r (V2Int64)) -psadbw128 = Ext.intrinsic ExtX86.sse2 "psad.bw" - -psllw128 :: Ext.T (V8Int16 -> V8Int16 -> LLVM.CodeGenFunction r (V8Int16)) -psllw128 = Ext.intrinsic ExtX86.sse2 "psll.w" - -pslld128 :: Ext.T (V4Int32 -> V4Int32 -> LLVM.CodeGenFunction r (V4Int32)) -pslld128 = Ext.intrinsic ExtX86.sse2 "psll.d" - -psllq128 :: Ext.T (V2Int64 -> V2Int64 -> LLVM.CodeGenFunction r (V2Int64)) -psllq128 = Ext.intrinsic ExtX86.sse2 "psll.q" - -psrlw128 :: Ext.T (V8Int16 -> V8Int16 -> LLVM.CodeGenFunction r (V8Int16)) -psrlw128 = Ext.intrinsic ExtX86.sse2 "psrl.w" - -psrld128 :: Ext.T (V4Int32 -> V4Int32 -> LLVM.CodeGenFunction r (V4Int32)) -psrld128 = Ext.intrinsic ExtX86.sse2 "psrl.d" - -psrlq128 :: Ext.T (V2Int64 -> V2Int64 -> LLVM.CodeGenFunction r (V2Int64)) -psrlq128 = Ext.intrinsic ExtX86.sse2 "psrl.q" - -psraw128 :: Ext.T (V8Int16 -> V8Int16 -> LLVM.CodeGenFunction r (V8Int16)) -psraw128 = Ext.intrinsic ExtX86.sse2 "psra.w" - -psrad128 :: Ext.T (V4Int32 -> V4Int32 -> LLVM.CodeGenFunction r (V4Int32)) -psrad128 = Ext.intrinsic ExtX86.sse2 "psra.d" - -psllwi128 :: Ext.T (V8Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V8Int16)) -psllwi128 = Ext.intrinsic ExtX86.sse2 "pslli.w" - -pslldi128 :: Ext.T (V4Int32 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V4Int32)) -pslldi128 = Ext.intrinsic ExtX86.sse2 "pslli.d" - -psllqi128 :: Ext.T (V2Int64 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V2Int64)) -psllqi128 = Ext.intrinsic ExtX86.sse2 "pslli.q" - -psrlwi128 :: Ext.T (V8Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V8Int16)) -psrlwi128 = Ext.intrinsic ExtX86.sse2 "psrli.w" - -psrldi128 :: Ext.T (V4Int32 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V4Int32)) -psrldi128 = Ext.intrinsic ExtX86.sse2 "psrli.d" - -psrlqi128 :: Ext.T (V2Int64 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V2Int64)) -psrlqi128 = Ext.intrinsic ExtX86.sse2 "psrli.q" - -psrawi128 :: Ext.T (V8Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V8Int16)) -psrawi128 = Ext.intrinsic ExtX86.sse2 "psrai.w" - -psradi128 :: Ext.T (V4Int32 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V4Int32)) -psradi128 = Ext.intrinsic ExtX86.sse2 "psrai.d" - -cvtdq2pd :: Ext.T (V4Int32 -> LLVM.CodeGenFunction r (V2Double)) -cvtdq2pd = Ext.intrinsic ExtX86.sse2 "cvtdq2pd" - -cvtdq2ps :: Ext.T (V4Int32 -> LLVM.CodeGenFunction r (V4Float)) -cvtdq2ps = Ext.intrinsic ExtX86.sse2 "cvtdq2ps" - -cvtpd2dq :: Ext.T (V2Double -> LLVM.CodeGenFunction r (V4Int32)) -cvtpd2dq = Ext.intrinsic ExtX86.sse2 "cvtpd2dq" - -cvttpd2dq :: Ext.T (V2Double -> LLVM.CodeGenFunction r (V4Int32)) -cvttpd2dq = Ext.intrinsic ExtX86.sse2 "cvttpd2dq" - -cvtpd2ps :: Ext.T (V2Double -> LLVM.CodeGenFunction r (V4Float)) -cvtpd2ps = Ext.intrinsic ExtX86.sse2 "cvtpd2ps" - -cvtps2dq :: Ext.T (V4Float -> LLVM.CodeGenFunction r (V4Int32)) -cvtps2dq = Ext.intrinsic ExtX86.sse2 "cvtps2dq" - -cvttps2dq :: Ext.T (V4Float -> LLVM.CodeGenFunction r (V4Int32)) -cvttps2dq = Ext.intrinsic ExtX86.sse2 "cvttps2dq" - -cvtps2pd :: Ext.T (V4Float -> LLVM.CodeGenFunction r (V2Double)) -cvtps2pd = Ext.intrinsic ExtX86.sse2 "cvtps2pd" - -cvtsd2si :: Ext.T (V2Double -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -cvtsd2si = Ext.intrinsic ExtX86.sse2 "cvtsd2si" - -cvtsd2si64 :: Ext.T (V2Double -> LLVM.CodeGenFunction r (LLVM.Value I.Int64)) -cvtsd2si64 = Ext.intrinsic ExtX86.sse2 "cvtsd2si64" - -cvttsd2si :: Ext.T (V2Double -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -cvttsd2si = Ext.intrinsic ExtX86.sse2 "cvttsd2si" - -cvttsd2si64 :: Ext.T (V2Double -> LLVM.CodeGenFunction r (LLVM.Value I.Int64)) -cvttsd2si64 = Ext.intrinsic ExtX86.sse2 "cvttsd2si64" - -cvtsi2sd :: Ext.T (V2Double -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V2Double)) -cvtsi2sd = Ext.intrinsic ExtX86.sse2 "cvtsi2sd" - -cvtsi642sd :: Ext.T (V2Double -> LLVM.Value I.Int64 -> LLVM.CodeGenFunction r (V2Double)) -cvtsi642sd = Ext.intrinsic ExtX86.sse2 "cvtsi642sd" - -cvtsd2ss :: Ext.T (V4Float -> V2Double -> LLVM.CodeGenFunction r (V4Float)) -cvtsd2ss = Ext.intrinsic ExtX86.sse2 "cvtsd2ss" - -cvtss2sd :: Ext.T (V2Double -> V4Float -> LLVM.CodeGenFunction r (V2Double)) -cvtss2sd = Ext.intrinsic ExtX86.sse2 "cvtss2sd" - -cvtpd2pi :: Ext.T (V2Double -> LLVM.CodeGenFunction r (MMX)) -cvtpd2pi = Ext.intrinsic ExtX86.sse1 "cvtpd2pi" - -cvttpd2pi :: Ext.T (V2Double -> LLVM.CodeGenFunction r (MMX)) -cvttpd2pi = Ext.intrinsic ExtX86.sse1 "cvttpd2pi" - -cvtpi2pd :: Ext.T (MMX -> LLVM.CodeGenFunction r (V2Double)) -cvtpi2pd = Ext.intrinsic ExtX86.sse1 "cvtpi2pd" - -storeupd :: Ext.T (LLVM.Value (Ptr ()) -> V2Double -> LLVM.CodeGenFunction r (LLVM.Value ())) -storeupd = Ext.intrinsic ExtX86.sse2 "storeu.pd" - -storedqu :: Ext.T (LLVM.Value (Ptr ()) -> V16Int8 -> LLVM.CodeGenFunction r (LLVM.Value ())) -storedqu = Ext.intrinsic ExtX86.sse2 "storeu.dq" - -storelv4si :: Ext.T (LLVM.Value (Ptr ()) -> V4Int32 -> LLVM.CodeGenFunction r (LLVM.Value ())) -storelv4si = Ext.intrinsic ExtX86.sse2 "storel.dq" - -packsswb128 :: Ext.T (V8Int16 -> V8Int16 -> LLVM.CodeGenFunction r (V16Int8)) -packsswb128 = Ext.intrinsic ExtX86.sse2 "packsswb.128" - -packssdw128 :: Ext.T (V4Int32 -> V4Int32 -> LLVM.CodeGenFunction r (V8Int16)) -packssdw128 = Ext.intrinsic ExtX86.sse2 "packssdw.128" - -packuswb128 :: Ext.T (V8Word16 -> V8Word16 -> LLVM.CodeGenFunction r (V16Word8)) -packuswb128 = Ext.intrinsic ExtX86.sse2 "packuswb.128" - -movmskpd :: Ext.T (V2Double -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -movmskpd = Ext.intrinsic ExtX86.sse2 "movmsk.pd" - -pmovmskb128 :: Ext.T (V16Int8 -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -pmovmskb128 = Ext.intrinsic ExtX86.sse2 "pmovmskb.128" - -maskmovdqu :: Ext.T (V16Int8 -> V16Int8 -> LLVM.Value (Ptr ()) -> LLVM.CodeGenFunction r (LLVM.Value ())) -maskmovdqu = Ext.intrinsic ExtX86.sse2 "maskmov.dqu" - -clflush :: Ext.T (LLVM.Value (Ptr ()) -> LLVM.CodeGenFunction r (LLVM.Value ())) -clflush = Ext.intrinsic ExtX86.sse2 "clflush" - -lfence :: Ext.T (LLVM.CodeGenFunction r (LLVM.Value ())) -lfence = Ext.intrinsic ExtX86.sse2 "lfence" - -mfence :: Ext.T (LLVM.CodeGenFunction r (LLVM.Value ())) -mfence = Ext.intrinsic ExtX86.sse2 "mfence" - -pause :: Ext.T (LLVM.CodeGenFunction r (LLVM.Value ())) -pause = Ext.intrinsic ExtX86.sse2 "pause" - -addsubps :: Ext.T (V4Float -> V4Float -> LLVM.CodeGenFunction r (V4Float)) -addsubps = Ext.intrinsic ExtX86.sse3 "addsub.ps" - -addsubpd :: Ext.T (V2Double -> V2Double -> LLVM.CodeGenFunction r (V2Double)) -addsubpd = Ext.intrinsic ExtX86.sse3 "addsub.pd" - -haddps :: Ext.T (V4Float -> V4Float -> LLVM.CodeGenFunction r (V4Float)) -haddps = Ext.intrinsic ExtX86.sse3 "hadd.ps" - -haddpd :: Ext.T (V2Double -> V2Double -> LLVM.CodeGenFunction r (V2Double)) -haddpd = Ext.intrinsic ExtX86.sse3 "hadd.pd" - -hsubps :: Ext.T (V4Float -> V4Float -> LLVM.CodeGenFunction r (V4Float)) -hsubps = Ext.intrinsic ExtX86.sse3 "hsub.ps" - -hsubpd :: Ext.T (V2Double -> V2Double -> LLVM.CodeGenFunction r (V2Double)) -hsubpd = Ext.intrinsic ExtX86.sse3 "hsub.pd" - -lddqu :: Ext.T (LLVM.Value (Ptr ()) -> LLVM.CodeGenFunction r (V16Int8)) -lddqu = Ext.intrinsic ExtX86.sse3 "ldu.dq" - -monitor :: Ext.T (LLVM.Value (Ptr ()) -> LLVM.Value I.Int32 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (LLVM.Value ())) -monitor = Ext.intrinsic ExtX86.sse3 "monitor" - -mwait :: Ext.T (LLVM.Value I.Int32 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (LLVM.Value ())) -mwait = Ext.intrinsic ExtX86.sse3 "mwait" - -phaddw :: Ext.T (MMX -> MMX -> LLVM.CodeGenFunction r (MMX)) -phaddw = Ext.intrinsic ExtX86.ssse3 "phadd.w" - -phaddw128 :: Ext.T (V8Int16 -> V8Int16 -> LLVM.CodeGenFunction r (V8Int16)) -phaddw128 = Ext.intrinsic ExtX86.ssse3 "phadd.w.128" - -phaddd :: Ext.T (MMX -> MMX -> LLVM.CodeGenFunction r (MMX)) -phaddd = Ext.intrinsic ExtX86.ssse3 "phadd.d" - -phaddd128 :: Ext.T (V4Int32 -> V4Int32 -> LLVM.CodeGenFunction r (V4Int32)) -phaddd128 = Ext.intrinsic ExtX86.ssse3 "phadd.d.128" - -phaddsw :: Ext.T (MMX -> MMX -> LLVM.CodeGenFunction r (MMX)) -phaddsw = Ext.intrinsic ExtX86.ssse3 "phadd.sw" - -phaddsw128 :: Ext.T (V8Int16 -> V8Int16 -> LLVM.CodeGenFunction r (V8Int16)) -phaddsw128 = Ext.intrinsic ExtX86.ssse3 "phadd.sw.128" - -phsubw :: Ext.T (MMX -> MMX -> LLVM.CodeGenFunction r (MMX)) -phsubw = Ext.intrinsic ExtX86.ssse3 "phsub.w" - -phsubw128 :: Ext.T (V8Int16 -> V8Int16 -> LLVM.CodeGenFunction r (V8Int16)) -phsubw128 = Ext.intrinsic ExtX86.ssse3 "phsub.w.128" - -phsubd :: Ext.T (MMX -> MMX -> LLVM.CodeGenFunction r (MMX)) -phsubd = Ext.intrinsic ExtX86.ssse3 "phsub.d" - -phsubd128 :: Ext.T (V4Int32 -> V4Int32 -> LLVM.CodeGenFunction r (V4Int32)) -phsubd128 = Ext.intrinsic ExtX86.ssse3 "phsub.d.128" - -phsubsw :: Ext.T (MMX -> MMX -> LLVM.CodeGenFunction r (MMX)) -phsubsw = Ext.intrinsic ExtX86.ssse3 "phsub.sw" - -phsubsw128 :: Ext.T (V8Int16 -> V8Int16 -> LLVM.CodeGenFunction r (V8Int16)) -phsubsw128 = Ext.intrinsic ExtX86.ssse3 "phsub.sw.128" - -pmaddubsw :: Ext.T (MMX -> MMX -> LLVM.CodeGenFunction r (MMX)) -pmaddubsw = Ext.intrinsic ExtX86.ssse3 "pmadd.ub.sw" - -pmaddubsw128 :: Ext.T (V16Word8 -> V16Word8 -> LLVM.CodeGenFunction r (V8Word16)) -pmaddubsw128 = Ext.intrinsic ExtX86.ssse3 "pmadd.ub.sw.128" - -pmulhrsw :: Ext.T (MMX -> MMX -> LLVM.CodeGenFunction r (MMX)) -pmulhrsw = Ext.intrinsic ExtX86.ssse3 "pmul.hr.sw" - -pmulhrsw128 :: Ext.T (V8Int16 -> V8Int16 -> LLVM.CodeGenFunction r (V8Int16)) -pmulhrsw128 = Ext.intrinsic ExtX86.ssse3 "pmul.hr.sw.128" - -pshufb :: Ext.T (MMX -> MMX -> LLVM.CodeGenFunction r (MMX)) -pshufb = Ext.intrinsic ExtX86.ssse3 "pshuf.b" - -pshufb128 :: Ext.T (V16Int8 -> V16Int8 -> LLVM.CodeGenFunction r (V16Int8)) -pshufb128 = Ext.intrinsic ExtX86.ssse3 "pshuf.b.128" - -pshufd :: Ext.T (V4Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int32)) -pshufd = Ext.intrinsic ExtX86.sse2 "pshuf.d" - -pshuflw :: Ext.T (V8Int16 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int16)) -pshuflw = Ext.intrinsic ExtX86.sse2 "pshufl.w" - -pshufhw :: Ext.T (V8Int16 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int16)) -pshufhw = Ext.intrinsic ExtX86.sse2 "pshufh.w" - -pshufw :: Ext.T (MMX -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (MMX)) -pshufw = Ext.intrinsic ExtX86.sse1 "pshuf.w" - -psignb :: Ext.T (MMX -> MMX -> LLVM.CodeGenFunction r (MMX)) -psignb = Ext.intrinsic ExtX86.ssse3 "psign.b" - -psignb128 :: Ext.T (V16Int8 -> V16Int8 -> LLVM.CodeGenFunction r (V16Int8)) -psignb128 = Ext.intrinsic ExtX86.ssse3 "psign.b.128" - -psignw :: Ext.T (MMX -> MMX -> LLVM.CodeGenFunction r (MMX)) -psignw = Ext.intrinsic ExtX86.ssse3 "psign.w" - -psignw128 :: Ext.T (V8Int16 -> V8Int16 -> LLVM.CodeGenFunction r (V8Int16)) -psignw128 = Ext.intrinsic ExtX86.ssse3 "psign.w.128" - -psignd :: Ext.T (MMX -> MMX -> LLVM.CodeGenFunction r (MMX)) -psignd = Ext.intrinsic ExtX86.ssse3 "psign.d" - -psignd128 :: Ext.T (V4Int32 -> V4Int32 -> LLVM.CodeGenFunction r (V4Int32)) -psignd128 = Ext.intrinsic ExtX86.ssse3 "psign.d.128" - -pabsb :: Ext.T (MMX -> LLVM.CodeGenFunction r (MMX)) -pabsb = Ext.intrinsic ExtX86.ssse3 "pabs.b" - -pabsb128 :: Ext.T (V16Int8 -> LLVM.CodeGenFunction r (V16Int8)) -pabsb128 = Ext.intrinsic ExtX86.ssse3 "pabs.b.128" - -pabsw :: Ext.T (MMX -> LLVM.CodeGenFunction r (MMX)) -pabsw = Ext.intrinsic ExtX86.ssse3 "pabs.w" - -pabsw128 :: Ext.T (V8Int16 -> LLVM.CodeGenFunction r (V8Int16)) -pabsw128 = Ext.intrinsic ExtX86.ssse3 "pabs.w.128" - -pabsd :: Ext.T (MMX -> LLVM.CodeGenFunction r (MMX)) -pabsd = Ext.intrinsic ExtX86.ssse3 "pabs.d" - -pabsd128 :: Ext.T (V4Int32 -> LLVM.CodeGenFunction r (V4Int32)) -pabsd128 = Ext.intrinsic ExtX86.ssse3 "pabs.d.128" - -roundss :: Ext.T (V4Float -> V4Float -> LLVM.Value W.Word32 -> LLVM.CodeGenFunction r (V4Float)) -roundss = Ext.intrinsic ExtX86.sse41 "round.ss" - -roundps :: Ext.T (V4Float -> LLVM.Value W.Word32 -> LLVM.CodeGenFunction r (V4Float)) -roundps = Ext.intrinsic ExtX86.sse41 "round.ps" - -roundsd :: Ext.T (V2Double -> V2Double -> LLVM.Value W.Word32 -> LLVM.CodeGenFunction r (V2Double)) -roundsd = Ext.intrinsic ExtX86.sse41 "round.sd" - -roundpd :: Ext.T (V2Double -> LLVM.Value W.Word32 -> LLVM.CodeGenFunction r (V2Double)) -roundpd = Ext.intrinsic ExtX86.sse41 "round.pd" - -pmovsxbd128 :: Ext.T (V16Int8 -> LLVM.CodeGenFunction r (V4Int32)) -pmovsxbd128 = Ext.intrinsic ExtX86.sse41 "pmovsxbd" - -pmovsxbq128 :: Ext.T (V16Int8 -> LLVM.CodeGenFunction r (V2Int64)) -pmovsxbq128 = Ext.intrinsic ExtX86.sse41 "pmovsxbq" - -pmovsxbw128 :: Ext.T (V16Int8 -> LLVM.CodeGenFunction r (V8Int16)) -pmovsxbw128 = Ext.intrinsic ExtX86.sse41 "pmovsxbw" - -pmovsxdq128 :: Ext.T (V4Int32 -> LLVM.CodeGenFunction r (V2Int64)) -pmovsxdq128 = Ext.intrinsic ExtX86.sse41 "pmovsxdq" - -pmovsxwd128 :: Ext.T (V8Int16 -> LLVM.CodeGenFunction r (V4Int32)) -pmovsxwd128 = Ext.intrinsic ExtX86.sse41 "pmovsxwd" - -pmovsxwq128 :: Ext.T (V8Int16 -> LLVM.CodeGenFunction r (V2Int64)) -pmovsxwq128 = Ext.intrinsic ExtX86.sse41 "pmovsxwq" - -pmovzxbd128 :: Ext.T (V16Int8 -> LLVM.CodeGenFunction r (V4Int32)) -pmovzxbd128 = Ext.intrinsic ExtX86.sse41 "pmovzxbd" - -pmovzxbq128 :: Ext.T (V16Int8 -> LLVM.CodeGenFunction r (V2Int64)) -pmovzxbq128 = Ext.intrinsic ExtX86.sse41 "pmovzxbq" - -pmovzxbw128 :: Ext.T (V16Int8 -> LLVM.CodeGenFunction r (V8Int16)) -pmovzxbw128 = Ext.intrinsic ExtX86.sse41 "pmovzxbw" - -pmovzxdq128 :: Ext.T (V4Int32 -> LLVM.CodeGenFunction r (V2Int64)) -pmovzxdq128 = Ext.intrinsic ExtX86.sse41 "pmovzxdq" - -pmovzxwd128 :: Ext.T (V8Int16 -> LLVM.CodeGenFunction r (V4Int32)) -pmovzxwd128 = Ext.intrinsic ExtX86.sse41 "pmovzxwd" - -pmovzxwq128 :: Ext.T (V8Int16 -> LLVM.CodeGenFunction r (V2Int64)) -pmovzxwq128 = Ext.intrinsic ExtX86.sse41 "pmovzxwq" - -phminposuw128 :: Ext.T (V8Word16 -> LLVM.CodeGenFunction r (V8Word16)) -phminposuw128 = Ext.intrinsic ExtX86.sse41 "phminposuw" - -pmaxsb128 :: Ext.T (V16Int8 -> V16Int8 -> LLVM.CodeGenFunction r (V16Int8)) -pmaxsb128 = Ext.intrinsic ExtX86.sse41 "pmaxsb" - -pmaxsd128 :: Ext.T (V4Int32 -> V4Int32 -> LLVM.CodeGenFunction r (V4Int32)) -pmaxsd128 = Ext.intrinsic ExtX86.sse41 "pmaxsd" - -pmaxud128 :: Ext.T (V4Word32 -> V4Word32 -> LLVM.CodeGenFunction r (V4Word32)) -pmaxud128 = Ext.intrinsic ExtX86.sse41 "pmaxud" - -pmaxuw128 :: Ext.T (V8Word16 -> V8Word16 -> LLVM.CodeGenFunction r (V8Word16)) -pmaxuw128 = Ext.intrinsic ExtX86.sse41 "pmaxuw" - -pminsb128 :: Ext.T (V16Int8 -> V16Int8 -> LLVM.CodeGenFunction r (V16Int8)) -pminsb128 = Ext.intrinsic ExtX86.sse41 "pminsb" - -pminsd128 :: Ext.T (V4Int32 -> V4Int32 -> LLVM.CodeGenFunction r (V4Int32)) -pminsd128 = Ext.intrinsic ExtX86.sse41 "pminsd" - -pminud128 :: Ext.T (V4Word32 -> V4Word32 -> LLVM.CodeGenFunction r (V4Word32)) -pminud128 = Ext.intrinsic ExtX86.sse41 "pminud" - -pminuw128 :: Ext.T (V8Word16 -> V8Word16 -> LLVM.CodeGenFunction r (V8Word16)) -pminuw128 = Ext.intrinsic ExtX86.sse41 "pminuw" - -aesimc128 :: Ext.T (V2Int64 -> LLVM.CodeGenFunction r (V2Int64)) -aesimc128 = Ext.intrinsic ExtX86.aes "aesimc" - -aesenc128 :: Ext.T (V2Int64 -> V2Int64 -> LLVM.CodeGenFunction r (V2Int64)) -aesenc128 = Ext.intrinsic ExtX86.aes "aesenc" - -aesenclast128 :: Ext.T (V2Int64 -> V2Int64 -> LLVM.CodeGenFunction r (V2Int64)) -aesenclast128 = Ext.intrinsic ExtX86.aes "aesenclast" - -aesdec128 :: Ext.T (V2Int64 -> V2Int64 -> LLVM.CodeGenFunction r (V2Int64)) -aesdec128 = Ext.intrinsic ExtX86.aes "aesdec" - -aesdeclast128 :: Ext.T (V2Int64 -> V2Int64 -> LLVM.CodeGenFunction r (V2Int64)) -aesdeclast128 = Ext.intrinsic ExtX86.aes "aesdeclast" - -aeskeygenassist128 :: Ext.T (V2Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Int64)) -aeskeygenassist128 = Ext.intrinsic ExtX86.aes "aeskeygenassist" - -{- -pclmulqdq128 :: Ext.T (V2Int64 -> V2Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Int64)) -pclmulqdq128 = Ext.intrinsic ExtX86. "int.x86.pclmulqdq" --} - -packusdw128 :: Ext.T (V4Word32 -> V4Word32 -> LLVM.CodeGenFunction r (V8Word16)) -packusdw128 = Ext.intrinsic ExtX86.sse41 "packusdw" - -pmuldq128 :: Ext.T (V4Int32 -> V4Int32 -> LLVM.CodeGenFunction r (V2Int64)) -pmuldq128 = Ext.intrinsic ExtX86.sse41 "pmuldq" - -extractps128 :: Ext.T (V4Float -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -extractps128 = Ext.intrinsic ExtX86.sse41 "extractps" - -insertps128 :: Ext.T (V4Float -> V4Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Float)) -insertps128 = Ext.intrinsic ExtX86.sse41 "insertps" - -pblendvb128 :: Ext.T (V16Int8 -> V16Int8 -> V16Int8 -> LLVM.CodeGenFunction r (V16Int8)) -pblendvb128 = Ext.intrinsic ExtX86.sse41 "pblendvb" - -blendvpd :: Ext.T (V2Double -> V2Double -> V2Double -> LLVM.CodeGenFunction r (V2Double)) -blendvpd = Ext.intrinsic ExtX86.sse41 "blendvpd" - -blendvps :: Ext.T (V4Float -> V4Float -> V4Float -> LLVM.CodeGenFunction r (V4Float)) -blendvps = Ext.intrinsic ExtX86.sse41 "blendvps" - -dppd :: Ext.T (V2Double -> V2Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Double)) -dppd = Ext.intrinsic ExtX86.sse41 "dppd" - -dpps :: Ext.T (V4Float -> V4Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Float)) -dpps = Ext.intrinsic ExtX86.sse41 "dpps" - -mpsadbw128 :: Ext.T (V16Int8 -> V16Int8 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int16)) -mpsadbw128 = Ext.intrinsic ExtX86.sse41 "mpsadbw" - -movntdqa :: Ext.T (LLVM.Value (Ptr ()) -> LLVM.CodeGenFunction r (V2Int64)) -movntdqa = Ext.intrinsic ExtX86.sse41 "movntdqa" - -ptestz128 :: Ext.T (V2Int64 -> V2Int64 -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -ptestz128 = Ext.intrinsic ExtX86.sse41 "ptestz" - -ptestc128 :: Ext.T (V2Int64 -> V2Int64 -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -ptestc128 = Ext.intrinsic ExtX86.sse41 "ptestc" - -ptestnzc128 :: Ext.T (V2Int64 -> V2Int64 -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -ptestnzc128 = Ext.intrinsic ExtX86.sse41 "ptestnzc" - -crc32qi :: Ext.T (LLVM.Value I.Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -crc32qi = Ext.intrinsic ExtX86.sse42 "crc32.32.8" - -crc32hi :: Ext.T (LLVM.Value I.Int32 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -crc32hi = Ext.intrinsic ExtX86.sse42 "crc32.32.16" - -crc32si :: Ext.T (LLVM.Value I.Int32 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -crc32si = Ext.intrinsic ExtX86.sse42 "crc32.32.32" - -crc32di :: Ext.T (LLVM.Value I.Int64 -> LLVM.Value I.Int64 -> LLVM.CodeGenFunction r (LLVM.Value I.Int64)) -crc32di = Ext.intrinsic ExtX86.sse42 "crc32.64.64" - -pcmpistrm128 :: Ext.T (V16Int8 -> V16Int8 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V16Int8)) -pcmpistrm128 = Ext.intrinsic ExtX86.sse42 "pcmpistrm128" - -pcmpistri128 :: Ext.T (V16Int8 -> V16Int8 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -pcmpistri128 = Ext.intrinsic ExtX86.sse42 "pcmpistri128" - -pcmpistria128 :: Ext.T (V16Int8 -> V16Int8 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -pcmpistria128 = Ext.intrinsic ExtX86.sse42 "pcmpistria128" - -pcmpistric128 :: Ext.T (V16Int8 -> V16Int8 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -pcmpistric128 = Ext.intrinsic ExtX86.sse42 "pcmpistric128" - -pcmpistrio128 :: Ext.T (V16Int8 -> V16Int8 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -pcmpistrio128 = Ext.intrinsic ExtX86.sse42 "pcmpistrio128" - -pcmpistris128 :: Ext.T (V16Int8 -> V16Int8 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -pcmpistris128 = Ext.intrinsic ExtX86.sse42 "pcmpistris128" - -pcmpistriz128 :: Ext.T (V16Int8 -> V16Int8 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -pcmpistriz128 = Ext.intrinsic ExtX86.sse42 "pcmpistriz128" - -pcmpestrm128 :: Ext.T (V16Int8 -> LLVM.Value I.Int32 -> V16Int8 -> LLVM.Value I.Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V16Int8)) -pcmpestrm128 = Ext.intrinsic ExtX86.sse42 "pcmpestrm128" - -pcmpestri128 :: Ext.T (V16Int8 -> LLVM.Value I.Int32 -> V16Int8 -> LLVM.Value I.Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -pcmpestri128 = Ext.intrinsic ExtX86.sse42 "pcmpestri128" - -pcmpestria128 :: Ext.T (V16Int8 -> LLVM.Value I.Int32 -> V16Int8 -> LLVM.Value I.Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -pcmpestria128 = Ext.intrinsic ExtX86.sse42 "pcmpestria128" - -pcmpestric128 :: Ext.T (V16Int8 -> LLVM.Value I.Int32 -> V16Int8 -> LLVM.Value I.Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -pcmpestric128 = Ext.intrinsic ExtX86.sse42 "pcmpestric128" - -pcmpestrio128 :: Ext.T (V16Int8 -> LLVM.Value I.Int32 -> V16Int8 -> LLVM.Value I.Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -pcmpestrio128 = Ext.intrinsic ExtX86.sse42 "pcmpestrio128" - -pcmpestris128 :: Ext.T (V16Int8 -> LLVM.Value I.Int32 -> V16Int8 -> LLVM.Value I.Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -pcmpestris128 = Ext.intrinsic ExtX86.sse42 "pcmpestris128" - -pcmpestriz128 :: Ext.T (V16Int8 -> LLVM.Value I.Int32 -> V16Int8 -> LLVM.Value I.Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -pcmpestriz128 = Ext.intrinsic ExtX86.sse42 "pcmpestriz128" - -extrqi :: Ext.T (V2Int64 -> LLVM.Value I.Int8 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Int64)) -extrqi = Ext.intrinsic ExtX86.sse4a "extrqi" - -extrq :: Ext.T (V2Int64 -> V16Int8 -> LLVM.CodeGenFunction r (V2Int64)) -extrq = Ext.intrinsic ExtX86.sse4a "extrq" - -insertqi :: Ext.T (V2Int64 -> V2Int64 -> LLVM.Value I.Int8 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Int64)) -insertqi = Ext.intrinsic ExtX86.sse4a "insertqi" - -insertq :: Ext.T (V2Int64 -> V2Int64 -> LLVM.CodeGenFunction r (V2Int64)) -insertq = Ext.intrinsic ExtX86.sse4a "insertq" - -movntss :: Ext.T (LLVM.Value (Ptr ()) -> V4Float -> LLVM.CodeGenFunction r (LLVM.Value ())) -movntss = Ext.intrinsic ExtX86.sse4a "movnt.ss" - -movntsd :: Ext.T (LLVM.Value (Ptr ()) -> V2Double -> LLVM.CodeGenFunction r (LLVM.Value ())) -movntsd = Ext.intrinsic ExtX86.sse4a "movnt.sd" - -addsubpd256 :: Ext.T (V4Double -> V4Double -> LLVM.CodeGenFunction r (V4Double)) -addsubpd256 = Ext.intrinsic ExtX86.avx "addsub.pd.256" - -addsubps256 :: Ext.T (V8Float -> V8Float -> LLVM.CodeGenFunction r (V8Float)) -addsubps256 = Ext.intrinsic ExtX86.avx "addsub.ps.256" - -maxpd256 :: Ext.T (V4Double -> V4Double -> LLVM.CodeGenFunction r (V4Double)) -maxpd256 = Ext.intrinsic ExtX86.avx "max.pd.256" - -maxps256 :: Ext.T (V8Float -> V8Float -> LLVM.CodeGenFunction r (V8Float)) -maxps256 = Ext.intrinsic ExtX86.avx "max.ps.256" - -minpd256 :: Ext.T (V4Double -> V4Double -> LLVM.CodeGenFunction r (V4Double)) -minpd256 = Ext.intrinsic ExtX86.avx "min.pd.256" - -minps256 :: Ext.T (V8Float -> V8Float -> LLVM.CodeGenFunction r (V8Float)) -minps256 = Ext.intrinsic ExtX86.avx "min.ps.256" - -sqrtpd256 :: Ext.T (V4Double -> LLVM.CodeGenFunction r (V4Double)) -sqrtpd256 = Ext.intrinsic ExtX86.avx "sqrt.pd.256" - -sqrtps256 :: Ext.T (V8Float -> LLVM.CodeGenFunction r (V8Float)) -sqrtps256 = Ext.intrinsic ExtX86.avx "sqrt.ps.256" - -rsqrtps256 :: Ext.T (V8Float -> LLVM.CodeGenFunction r (V8Float)) -rsqrtps256 = Ext.intrinsic ExtX86.avx "rsqrt.ps.256" - -rcpps256 :: Ext.T (V8Float -> LLVM.CodeGenFunction r (V8Float)) -rcpps256 = Ext.intrinsic ExtX86.avx "rcp.ps.256" - -roundpd256 :: Ext.T (V4Double -> LLVM.Value W.Word32 -> LLVM.CodeGenFunction r (V4Double)) -roundpd256 = Ext.intrinsic ExtX86.avx "round.pd.256" - -roundps256 :: Ext.T (V8Float -> LLVM.Value W.Word32 -> LLVM.CodeGenFunction r (V8Float)) -roundps256 = Ext.intrinsic ExtX86.avx "round.ps.256" - -haddpd256 :: Ext.T (V4Double -> V4Double -> LLVM.CodeGenFunction r (V4Double)) -haddpd256 = Ext.intrinsic ExtX86.avx "hadd.pd.256" - -hsubps256 :: Ext.T (V8Float -> V8Float -> LLVM.CodeGenFunction r (V8Float)) -hsubps256 = Ext.intrinsic ExtX86.avx "hsub.ps.256" - -hsubpd256 :: Ext.T (V4Double -> V4Double -> LLVM.CodeGenFunction r (V4Double)) -hsubpd256 = Ext.intrinsic ExtX86.avx "hsub.pd.256" - -haddps256 :: Ext.T (V8Float -> V8Float -> LLVM.CodeGenFunction r (V8Float)) -haddps256 = Ext.intrinsic ExtX86.avx "hadd.ps.256" - -vpermilvarpd :: Ext.T (V2Double -> V2Int64 -> LLVM.CodeGenFunction r (V2Double)) -vpermilvarpd = Ext.intrinsic ExtX86.avx "vpermilvar.pd" - -vpermilvarps :: Ext.T (V4Float -> V4Int32 -> LLVM.CodeGenFunction r (V4Float)) -vpermilvarps = Ext.intrinsic ExtX86.avx "vpermilvar.ps" - -vpermilvarpd256 :: Ext.T (V4Double -> V4Int64 -> LLVM.CodeGenFunction r (V4Double)) -vpermilvarpd256 = Ext.intrinsic ExtX86.avx "vpermilvar.pd.256" - -vpermilvarps256 :: Ext.T (V8Float -> V8Int32 -> LLVM.CodeGenFunction r (V8Float)) -vpermilvarps256 = Ext.intrinsic ExtX86.avx "vpermilvar.ps.256" - -vperm2f128_pd256 :: Ext.T (V4Double -> V4Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Double)) -vperm2f128_pd256 = Ext.intrinsic ExtX86.avx "vperm2f128.pd.256" - -vperm2f128_ps256 :: Ext.T (V8Float -> V8Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Float)) -vperm2f128_ps256 = Ext.intrinsic ExtX86.avx "vperm2f128.ps.256" - -vperm2f128_si256 :: Ext.T (V8Int32 -> V8Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int32)) -vperm2f128_si256 = Ext.intrinsic ExtX86.avx "vperm2f128.si.256" - -vpermi2vard128_mask :: Ext.T (V4Int32 -> V4Int32 -> V4Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int32)) -vpermi2vard128_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermi2var.d.128" - -vpermi2vard256_mask :: Ext.T (V8Int32 -> V8Int32 -> V8Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int32)) -vpermi2vard256_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermi2var.d.256" - -vpermi2vard512_mask :: Ext.T (V16Int32 -> V16Int32 -> V16Int32 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int32)) -vpermi2vard512_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermi2var.d.512" - -vpermi2varhi128_mask :: Ext.T (V8Int16 -> V8Int16 -> V8Int16 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int16)) -vpermi2varhi128_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermi2var.hi.128" - -vpermi2varhi256_mask :: Ext.T (V16Int16 -> V16Int16 -> V16Int16 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int16)) -vpermi2varhi256_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermi2var.hi.256" - -vpermi2varhi512_mask :: Ext.T (V32Int16 -> V32Int16 -> V32Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V32Int16)) -vpermi2varhi512_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermi2var.hi.512" - -vpermi2varpd128_mask :: Ext.T (V2Double -> V2Int64 -> V2Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Double)) -vpermi2varpd128_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermi2var.pd.128" - -vpermi2varpd256_mask :: Ext.T (V4Double -> V4Int64 -> V4Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Double)) -vpermi2varpd256_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermi2var.pd.256" - -vpermi2varpd512_mask :: Ext.T (V8Double -> V8Int64 -> V8Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Double)) -vpermi2varpd512_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermi2var.pd.512" - -vpermi2varps128_mask :: Ext.T (V4Float -> V4Int32 -> V4Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Float)) -vpermi2varps128_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermi2var.ps.128" - -vpermi2varps256_mask :: Ext.T (V8Float -> V8Int32 -> V8Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Float)) -vpermi2varps256_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermi2var.ps.256" - -vpermi2varps512_mask :: Ext.T (V16Float -> V16Int32 -> V16Float -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Float)) -vpermi2varps512_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermi2var.ps.512" - -vpermi2varq128_mask :: Ext.T (V2Int64 -> V2Int64 -> V2Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Int64)) -vpermi2varq128_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermi2var.q.128" - -vpermi2varq256_mask :: Ext.T (V4Int64 -> V4Int64 -> V4Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int64)) -vpermi2varq256_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermi2var.q.256" - -vpermi2varq512_mask :: Ext.T (V8Int64 -> V8Int64 -> V8Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int64)) -vpermi2varq512_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermi2var.q.512" - -vpermt2vard512_mask :: Ext.T (V16Int32 -> V16Int32 -> V16Int32 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int32)) -vpermt2vard512_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermt2var.d.512" - -vpermt2varq512_mask :: Ext.T (V8Int64 -> V8Int64 -> V8Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int64)) -vpermt2varq512_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermt2var.q.512" - -vpermt2varps512_mask :: Ext.T (V16Int32 -> V16Float -> V16Float -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Float)) -vpermt2varps512_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermt2var.ps.512" - -vpermt2varpd512_mask :: Ext.T (V8Int64 -> V8Double -> V8Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Double)) -vpermt2varpd512_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermt2var.pd.512" - -vpermt2vard128_mask :: Ext.T (V4Int32 -> V4Int32 -> V4Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int32)) -vpermt2vard128_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermt2var.d.128" - -vpermt2vard128_maskz :: Ext.T (V4Int32 -> V4Int32 -> V4Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int32)) -vpermt2vard128_maskz = Ext.intrinsic ExtX86.avx512 "maskz.vpermt2var.d.128" - -vpermt2vard256_mask :: Ext.T (V8Int32 -> V8Int32 -> V8Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int32)) -vpermt2vard256_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermt2var.d.256" - -vpermt2vard256_maskz :: Ext.T (V8Int32 -> V8Int32 -> V8Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int32)) -vpermt2vard256_maskz = Ext.intrinsic ExtX86.avx512 "maskz.vpermt2var.d.256" - -vpermt2vard512_maskz :: Ext.T (V16Int32 -> V16Int32 -> V16Int32 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int32)) -vpermt2vard512_maskz = Ext.intrinsic ExtX86.avx512 "maskz.vpermt2var.d.512" - -vpermt2varhi128_mask :: Ext.T (V8Int16 -> V8Int16 -> V8Int16 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int16)) -vpermt2varhi128_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermt2var.hi.128" - -vpermt2varhi128_maskz :: Ext.T (V8Int16 -> V8Int16 -> V8Int16 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int16)) -vpermt2varhi128_maskz = Ext.intrinsic ExtX86.avx512 "maskz.vpermt2var.hi.128" - -vpermt2varhi256_mask :: Ext.T (V16Int16 -> V16Int16 -> V16Int16 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int16)) -vpermt2varhi256_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermt2var.hi.256" - -vpermt2varhi256_maskz :: Ext.T (V16Int16 -> V16Int16 -> V16Int16 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int16)) -vpermt2varhi256_maskz = Ext.intrinsic ExtX86.avx512 "maskz.vpermt2var.hi.256" - -vpermt2varhi512_mask :: Ext.T (V32Int16 -> V32Int16 -> V32Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V32Int16)) -vpermt2varhi512_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermt2var.hi.512" - -vpermt2varhi512_maskz :: Ext.T (V32Int16 -> V32Int16 -> V32Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V32Int16)) -vpermt2varhi512_maskz = Ext.intrinsic ExtX86.avx512 "maskz.vpermt2var.hi.512" - -vpermt2varpd128_mask :: Ext.T (V2Int64 -> V2Double -> V2Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Double)) -vpermt2varpd128_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermt2var.pd.128" - -vpermt2varpd128_maskz :: Ext.T (V2Int64 -> V2Double -> V2Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Double)) -vpermt2varpd128_maskz = Ext.intrinsic ExtX86.avx512 "maskz.vpermt2var.pd.128" - -vpermt2varpd256_mask :: Ext.T (V4Int64 -> V4Double -> V4Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Double)) -vpermt2varpd256_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermt2var.pd.256" - -vpermt2varpd256_maskz :: Ext.T (V4Int64 -> V4Double -> V4Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Double)) -vpermt2varpd256_maskz = Ext.intrinsic ExtX86.avx512 "maskz.vpermt2var.pd.256" - -vpermt2varpd512_maskz :: Ext.T (V8Int64 -> V8Double -> V8Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Double)) -vpermt2varpd512_maskz = Ext.intrinsic ExtX86.avx512 "maskz.vpermt2var.pd.512" - -vpermt2varps128_mask :: Ext.T (V4Int32 -> V4Float -> V4Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Float)) -vpermt2varps128_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermt2var.ps.128" - -vpermt2varps128_maskz :: Ext.T (V4Int32 -> V4Float -> V4Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Float)) -vpermt2varps128_maskz = Ext.intrinsic ExtX86.avx512 "maskz.vpermt2var.ps.128" - -vpermt2varps256_mask :: Ext.T (V8Int32 -> V8Float -> V8Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Float)) -vpermt2varps256_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermt2var.ps.256" - -vpermt2varps256_maskz :: Ext.T (V8Int32 -> V8Float -> V8Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Float)) -vpermt2varps256_maskz = Ext.intrinsic ExtX86.avx512 "maskz.vpermt2var.ps.256" - -vpermt2varps512_maskz :: Ext.T (V16Int32 -> V16Float -> V16Float -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Float)) -vpermt2varps512_maskz = Ext.intrinsic ExtX86.avx512 "maskz.vpermt2var.ps.512" - -vpermt2varq128_mask :: Ext.T (V2Int64 -> V2Int64 -> V2Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Int64)) -vpermt2varq128_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermt2var.q.128" - -vpermt2varq128_maskz :: Ext.T (V2Int64 -> V2Int64 -> V2Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Int64)) -vpermt2varq128_maskz = Ext.intrinsic ExtX86.avx512 "maskz.vpermt2var.q.128" - -vpermt2varq256_mask :: Ext.T (V4Int64 -> V4Int64 -> V4Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int64)) -vpermt2varq256_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermt2var.q.256" - -vpermt2varq256_maskz :: Ext.T (V4Int64 -> V4Int64 -> V4Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int64)) -vpermt2varq256_maskz = Ext.intrinsic ExtX86.avx512 "maskz.vpermt2var.q.256" - -vpermt2varq512_maskz :: Ext.T (V8Int64 -> V8Int64 -> V8Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int64)) -vpermt2varq512_maskz = Ext.intrinsic ExtX86.avx512 "maskz.vpermt2var.q.512" - -vpermilpd_mask :: Ext.T (V2Double -> LLVM.Value I.Int32 -> V2Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Double)) -vpermilpd_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermil.pd.128" - -vpermilpd256_mask :: Ext.T (V4Double -> LLVM.Value I.Int32 -> V4Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Double)) -vpermilpd256_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermil.pd.256" - -vpermilpd512_mask :: Ext.T (V8Double -> LLVM.Value I.Int32 -> V8Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Double)) -vpermilpd512_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermil.pd.512" - -vpermilps_mask :: Ext.T (V4Float -> LLVM.Value I.Int32 -> V4Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Float)) -vpermilps_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermil.ps.128" - -vpermilps256_mask :: Ext.T (V8Float -> LLVM.Value I.Int32 -> V8Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Float)) -vpermilps256_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermil.ps.256" - -vpermilps512_mask :: Ext.T (V16Float -> LLVM.Value I.Int32 -> V16Float -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Float)) -vpermilps512_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermil.ps.512" - -vpermilvarpd256_mask :: Ext.T (V4Double -> V4Int64 -> V4Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Double)) -vpermilvarpd256_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermilvar.pd.256" - -vpermilvarpd512_mask :: Ext.T (V8Double -> V8Int64 -> V8Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Double)) -vpermilvarpd512_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermilvar.pd.512" - -vpermilvarpd_mask :: Ext.T (V2Double -> V2Int64 -> V2Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Double)) -vpermilvarpd_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermilvar.pd.128" - -vpermilvarps256_mask :: Ext.T (V8Float -> V8Int32 -> V8Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Float)) -vpermilvarps256_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermilvar.ps.256" - -vpermilvarps512_mask :: Ext.T (V16Float -> V16Int32 -> V16Float -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Float)) -vpermilvarps512_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermilvar.ps.512" - -vpermilvarps_mask :: Ext.T (V4Float -> V4Int32 -> V4Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Float)) -vpermilvarps_mask = Ext.intrinsic ExtX86.avx512 "mask.vpermilvar.ps.128" - -pshufb128_mask :: Ext.T (V16Int8 -> V16Int8 -> V16Int8 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int8)) -pshufb128_mask = Ext.intrinsic ExtX86.avx512 "mask.pshuf.b.128" - -pshufb256_mask :: Ext.T (V32Int8 -> V32Int8 -> V32Int8 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V32Int8)) -pshufb256_mask = Ext.intrinsic ExtX86.avx512 "mask.pshuf.b.256" - -pshufb512_mask :: Ext.T (V64Int8 -> V64Int8 -> V64Int8 -> LLVM.Value I.Int64 -> LLVM.CodeGenFunction r (V64Int8)) -pshufb512_mask = Ext.intrinsic ExtX86.avx512 "mask.pshuf.b.512" - -pshufd128_mask :: Ext.T (V4Int32 -> LLVM.Value I.Int16 -> V4Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int32)) -pshufd128_mask = Ext.intrinsic ExtX86.avx512 "mask.pshuf.d.128" - -pshufd256_mask :: Ext.T (V8Int32 -> LLVM.Value I.Int16 -> V8Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int32)) -pshufd256_mask = Ext.intrinsic ExtX86.avx512 "mask.pshuf.d.256" - -pshufd512_mask :: Ext.T (V16Int32 -> LLVM.Value I.Int16 -> V16Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V16Int32)) -pshufd512_mask = Ext.intrinsic ExtX86.avx512 "mask.pshuf.d.512" - -pshufhw128_mask :: Ext.T (V8Int16 -> LLVM.Value I.Int8 -> V8Int16 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int16)) -pshufhw128_mask = Ext.intrinsic ExtX86.avx512 "mask.pshufh.w.128" - -pshufhw256_mask :: Ext.T (V16Int16 -> LLVM.Value I.Int8 -> V16Int16 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int16)) -pshufhw256_mask = Ext.intrinsic ExtX86.avx512 "mask.pshufh.w.256" - -pshufhw512_mask :: Ext.T (V32Int16 -> LLVM.Value I.Int8 -> V32Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V32Int16)) -pshufhw512_mask = Ext.intrinsic ExtX86.avx512 "mask.pshufh.w.512" - -pshuflw128_mask :: Ext.T (V8Int16 -> LLVM.Value I.Int8 -> V8Int16 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int16)) -pshuflw128_mask = Ext.intrinsic ExtX86.avx512 "mask.pshufl.w.128" - -pshuflw256_mask :: Ext.T (V16Int16 -> LLVM.Value I.Int8 -> V16Int16 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int16)) -pshuflw256_mask = Ext.intrinsic ExtX86.avx512 "mask.pshufl.w.256" - -pshuflw512_mask :: Ext.T (V32Int16 -> LLVM.Value I.Int8 -> V32Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V32Int16)) -pshuflw512_mask = Ext.intrinsic ExtX86.avx512 "mask.pshufl.w.512" - -shuf_f32x4_256_mask :: Ext.T (V8Float -> V8Float -> LLVM.Value I.Int32 -> V8Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Float)) -shuf_f32x4_256_mask = Ext.intrinsic ExtX86.avx512 "mask.shuf.f32x4.256" - -shuf_f32x4_mask :: Ext.T (V16Float -> V16Float -> LLVM.Value I.Int32 -> V16Float -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Float)) -shuf_f32x4_mask = Ext.intrinsic ExtX86.avx512 "mask.shuf.f32x4" - -shuf_f64x2_256_mask :: Ext.T (V4Double -> V4Double -> LLVM.Value I.Int32 -> V4Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Double)) -shuf_f64x2_256_mask = Ext.intrinsic ExtX86.avx512 "mask.shuf.f64x2.256" - -shuf_f64x2_mask :: Ext.T (V8Double -> V8Double -> LLVM.Value I.Int32 -> V8Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Double)) -shuf_f64x2_mask = Ext.intrinsic ExtX86.avx512 "mask.shuf.f64x2" - -shuf_i32x4_256_mask :: Ext.T (V8Int32 -> V8Int32 -> LLVM.Value I.Int32 -> V8Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int32)) -shuf_i32x4_256_mask = Ext.intrinsic ExtX86.avx512 "mask.shuf.i32x4.256" - -shuf_i32x4_mask :: Ext.T (V16Int32 -> V16Int32 -> LLVM.Value I.Int32 -> V16Int32 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int32)) -shuf_i32x4_mask = Ext.intrinsic ExtX86.avx512 "mask.shuf.i32x4" - -shuf_i64x2_256_mask :: Ext.T (V4Int64 -> V4Int64 -> LLVM.Value I.Int32 -> V4Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int64)) -shuf_i64x2_256_mask = Ext.intrinsic ExtX86.avx512 "mask.shuf.i64x2.256" - -shuf_i64x2_mask :: Ext.T (V8Int64 -> V8Int64 -> LLVM.Value I.Int32 -> V8Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int64)) -shuf_i64x2_mask = Ext.intrinsic ExtX86.avx512 "mask.shuf.i64x2" - -shufpd128_mask :: Ext.T (V2Double -> V2Double -> LLVM.Value I.Int32 -> V2Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Double)) -shufpd128_mask = Ext.intrinsic ExtX86.avx512 "mask.shuf.pd.128" - -shufpd256_mask :: Ext.T (V4Double -> V4Double -> LLVM.Value I.Int32 -> V4Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Double)) -shufpd256_mask = Ext.intrinsic ExtX86.avx512 "mask.shuf.pd.256" - -shufpd512_mask :: Ext.T (V8Double -> V8Double -> LLVM.Value I.Int32 -> V8Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Double)) -shufpd512_mask = Ext.intrinsic ExtX86.avx512 "mask.shuf.pd.512" - -shufps128_mask :: Ext.T (V4Float -> V4Float -> LLVM.Value I.Int32 -> V4Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Float)) -shufps128_mask = Ext.intrinsic ExtX86.avx512 "mask.shuf.ps.128" - -shufps256_mask :: Ext.T (V8Float -> V8Float -> LLVM.Value I.Int32 -> V8Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Float)) -shufps256_mask = Ext.intrinsic ExtX86.avx512 "mask.shuf.ps.256" - -shufps512_mask :: Ext.T (V16Float -> V16Float -> LLVM.Value I.Int32 -> V16Float -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Float)) -shufps512_mask = Ext.intrinsic ExtX86.avx512 "mask.shuf.ps.512" - -movshdup128_mask :: Ext.T (V4Float -> V4Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Float)) -movshdup128_mask = Ext.intrinsic ExtX86.avx512 "mask.movshdup.128" - -movshdup256_mask :: Ext.T (V8Float -> V8Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Float)) -movshdup256_mask = Ext.intrinsic ExtX86.avx512 "mask.movshdup.256" - -movshdup512_mask :: Ext.T (V16Float -> V16Float -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Float)) -movshdup512_mask = Ext.intrinsic ExtX86.avx512 "mask.movshdup.512" - -movsldup128_mask :: Ext.T (V4Float -> V4Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Float)) -movsldup128_mask = Ext.intrinsic ExtX86.avx512 "mask.movsldup.128" - -movsldup256_mask :: Ext.T (V8Float -> V8Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Float)) -movsldup256_mask = Ext.intrinsic ExtX86.avx512 "mask.movsldup.256" - -movsldup512_mask :: Ext.T (V16Float -> V16Float -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Float)) -movsldup512_mask = Ext.intrinsic ExtX86.avx512 "mask.movsldup.512" - -movddup128_mask :: Ext.T (V2Double -> V2Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Double)) -movddup128_mask = Ext.intrinsic ExtX86.avx512 "mask.movddup.128" - -movddup256_mask :: Ext.T (V4Double -> V4Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Double)) -movddup256_mask = Ext.intrinsic ExtX86.avx512 "mask.movddup.256" - -movddup512_mask :: Ext.T (V8Double -> V8Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Double)) -movddup512_mask = Ext.intrinsic ExtX86.avx512 "mask.movddup.512" - -blendvpd256 :: Ext.T (V4Double -> V4Double -> V4Double -> LLVM.CodeGenFunction r (V4Double)) -blendvpd256 = Ext.intrinsic ExtX86.avx "blendv.pd.256" - -blendvps256 :: Ext.T (V8Float -> V8Float -> V8Float -> LLVM.CodeGenFunction r (V8Float)) -blendvps256 = Ext.intrinsic ExtX86.avx "blendv.ps.256" - -dpps256 :: Ext.T (V8Float -> V8Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Float)) -dpps256 = Ext.intrinsic ExtX86.avx "dp.ps.256" - -cmppd256 :: Ext.T (V4Double -> V4Double -> LLVM.Value W.Word8 -> LLVM.CodeGenFunction r (V4Double)) -cmppd256 = Ext.intrinsic ExtX86.avx "cmp.pd.256" - -cmpps256 :: Ext.T (V8Float -> V8Float -> LLVM.Value W.Word8 -> LLVM.CodeGenFunction r (V8Float)) -cmpps256 = Ext.intrinsic ExtX86.avx "cmp.ps.256" - -cvtdq2pd256 :: Ext.T (V4Int32 -> LLVM.CodeGenFunction r (V4Double)) -cvtdq2pd256 = Ext.intrinsic ExtX86.avx "cvtdq2.pd.256" - -cvtdq2ps256 :: Ext.T (V8Int32 -> LLVM.CodeGenFunction r (V8Float)) -cvtdq2ps256 = Ext.intrinsic ExtX86.avx "cvtdq2.ps.256" - -cvtpd2ps256 :: Ext.T (V4Double -> LLVM.CodeGenFunction r (V4Float)) -cvtpd2ps256 = Ext.intrinsic ExtX86.avx "cvt.pd2.ps.256" - -cvtps2dq256 :: Ext.T (V8Float -> LLVM.CodeGenFunction r (V8Int32)) -cvtps2dq256 = Ext.intrinsic ExtX86.avx "cvt.ps2dq.256" - -cvtps2pd256 :: Ext.T (V4Float -> LLVM.CodeGenFunction r (V4Double)) -cvtps2pd256 = Ext.intrinsic ExtX86.avx "cvt.ps2.pd.256" - -cvttpd2dq256 :: Ext.T (V4Double -> LLVM.CodeGenFunction r (V4Int32)) -cvttpd2dq256 = Ext.intrinsic ExtX86.avx "cvtt.pd2dq.256" - -cvtpd2dq256 :: Ext.T (V4Double -> LLVM.CodeGenFunction r (V4Int32)) -cvtpd2dq256 = Ext.intrinsic ExtX86.avx "cvt.pd2dq.256" - -cvttps2dq256 :: Ext.T (V8Float -> LLVM.CodeGenFunction r (V8Int32)) -cvttps2dq256 = Ext.intrinsic ExtX86.avx "cvtt.ps2dq.256" - -vtestzpd :: Ext.T (V2Double -> V2Double -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -vtestzpd = Ext.intrinsic ExtX86.avx "vtestz.pd" - -vtestcpd :: Ext.T (V2Double -> V2Double -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -vtestcpd = Ext.intrinsic ExtX86.avx "vtestc.pd" - -vtestnzcpd :: Ext.T (V2Double -> V2Double -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -vtestnzcpd = Ext.intrinsic ExtX86.avx "vtestnzc.pd" - -vtestzps :: Ext.T (V4Float -> V4Float -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -vtestzps = Ext.intrinsic ExtX86.avx "vtestz.ps" - -vtestcps :: Ext.T (V4Float -> V4Float -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -vtestcps = Ext.intrinsic ExtX86.avx "vtestc.ps" - -vtestnzcps :: Ext.T (V4Float -> V4Float -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -vtestnzcps = Ext.intrinsic ExtX86.avx "vtestnzc.ps" - -vtestzpd256 :: Ext.T (V4Double -> V4Double -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -vtestzpd256 = Ext.intrinsic ExtX86.avx "vtestz.pd.256" - -vtestcpd256 :: Ext.T (V4Double -> V4Double -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -vtestcpd256 = Ext.intrinsic ExtX86.avx "vtestc.pd.256" - -vtestnzcpd256 :: Ext.T (V4Double -> V4Double -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -vtestnzcpd256 = Ext.intrinsic ExtX86.avx "vtestnzc.pd.256" - -vtestzps256 :: Ext.T (V8Float -> V8Float -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -vtestzps256 = Ext.intrinsic ExtX86.avx "vtestz.ps.256" - -vtestcps256 :: Ext.T (V8Float -> V8Float -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -vtestcps256 = Ext.intrinsic ExtX86.avx "vtestc.ps.256" - -vtestnzcps256 :: Ext.T (V8Float -> V8Float -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -vtestnzcps256 = Ext.intrinsic ExtX86.avx "vtestnzc.ps.256" - -ptestz256 :: Ext.T (V4Int64 -> V4Int64 -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -ptestz256 = Ext.intrinsic ExtX86.avx "ptestz.256" - -ptestc256 :: Ext.T (V4Int64 -> V4Int64 -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -ptestc256 = Ext.intrinsic ExtX86.avx "ptestc.256" - -ptestnzc256 :: Ext.T (V4Int64 -> V4Int64 -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -ptestnzc256 = Ext.intrinsic ExtX86.avx "ptestnzc.256" - -ptestmd512 :: Ext.T (V16Int32 -> V16Int32 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (LLVM.Value I.Int16)) -ptestmd512 = Ext.intrinsic ExtX86.avx512 "mask.ptestm.d.512" - -ptestmq512 :: Ext.T (V8Int64 -> V8Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (LLVM.Value I.Int8)) -ptestmq512 = Ext.intrinsic ExtX86.avx512 "mask.ptestm.q.512" - -fpclasspd128_mask :: Ext.T (V2Double -> LLVM.Value I.Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (LLVM.Value I.Int8)) -fpclasspd128_mask = Ext.intrinsic ExtX86.avx512 "mask.fpclass.pd.128" - -fpclasspd256_mask :: Ext.T (V4Double -> LLVM.Value I.Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (LLVM.Value I.Int8)) -fpclasspd256_mask = Ext.intrinsic ExtX86.avx512 "mask.fpclass.pd.256" - -fpclasspd512_mask :: Ext.T (V8Double -> LLVM.Value I.Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (LLVM.Value I.Int8)) -fpclasspd512_mask = Ext.intrinsic ExtX86.avx512 "mask.fpclass.pd.512" - -fpclassps128_mask :: Ext.T (V4Float -> LLVM.Value I.Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (LLVM.Value I.Int8)) -fpclassps128_mask = Ext.intrinsic ExtX86.avx512 "mask.fpclass.ps.128" - -fpclassps256_mask :: Ext.T (V8Float -> LLVM.Value I.Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (LLVM.Value I.Int8)) -fpclassps256_mask = Ext.intrinsic ExtX86.avx512 "mask.fpclass.ps.256" - -fpclassps512_mask :: Ext.T (V16Float -> LLVM.Value I.Int32 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (LLVM.Value I.Int16)) -fpclassps512_mask = Ext.intrinsic ExtX86.avx512 "mask.fpclass.ps.512" - -fpclasssd :: Ext.T (V2Double -> LLVM.Value I.Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (LLVM.Value I.Int8)) -fpclasssd = Ext.intrinsic ExtX86.avx512 "mask.fpclass.sd" - -fpclassss :: Ext.T (V4Float -> LLVM.Value I.Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (LLVM.Value I.Int8)) -fpclassss = Ext.intrinsic ExtX86.avx512 "mask.fpclass.ss" - -movmskpd256 :: Ext.T (V4Double -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -movmskpd256 = Ext.intrinsic ExtX86.avx "movmsk.pd.256" - -movmskps256 :: Ext.T (V8Float -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -movmskps256 = Ext.intrinsic ExtX86.avx "movmsk.ps.256" - -vzeroall :: Ext.T (LLVM.CodeGenFunction r (LLVM.Value ())) -vzeroall = Ext.intrinsic ExtX86.avx "vzeroall" - -vzeroupper :: Ext.T (LLVM.CodeGenFunction r (LLVM.Value ())) -vzeroupper = Ext.intrinsic ExtX86.avx "vzeroupper" - -vbroadcastf128_pd256 :: Ext.T (LLVM.Value (Ptr ()) -> LLVM.CodeGenFunction r (V4Double)) -vbroadcastf128_pd256 = Ext.intrinsic ExtX86.avx "vbroadcastf128.pd.256" - -vbroadcastf128_ps256 :: Ext.T (LLVM.Value (Ptr ()) -> LLVM.CodeGenFunction r (V8Float)) -vbroadcastf128_ps256 = Ext.intrinsic ExtX86.avx "vbroadcastf128.ps.256" - -lddqu256 :: Ext.T (LLVM.Value (Ptr ()) -> LLVM.CodeGenFunction r (V32Int8)) -lddqu256 = Ext.intrinsic ExtX86.avx "ldu.dq.256" - -storeupd256 :: Ext.T (LLVM.Value (Ptr ()) -> V4Double -> LLVM.CodeGenFunction r (LLVM.Value ())) -storeupd256 = Ext.intrinsic ExtX86.avx "storeu.pd.256" - -storeups256 :: Ext.T (LLVM.Value (Ptr ()) -> V8Float -> LLVM.CodeGenFunction r (LLVM.Value ())) -storeups256 = Ext.intrinsic ExtX86.avx "storeu.ps.256" - -storedqu256 :: Ext.T (LLVM.Value (Ptr ()) -> V32Int8 -> LLVM.CodeGenFunction r (LLVM.Value ())) -storedqu256 = Ext.intrinsic ExtX86.avx "storeu.dq.256" - -maskloadpd :: Ext.T (LLVM.Value (Ptr ()) -> V2Int64 -> LLVM.CodeGenFunction r (V2Double)) -maskloadpd = Ext.intrinsic ExtX86.avx "maskload.pd" - -maskloadps :: Ext.T (LLVM.Value (Ptr ()) -> V4Int32 -> LLVM.CodeGenFunction r (V4Float)) -maskloadps = Ext.intrinsic ExtX86.avx "maskload.ps" - -maskloadpd256 :: Ext.T (LLVM.Value (Ptr ()) -> V4Int64 -> LLVM.CodeGenFunction r (V4Double)) -maskloadpd256 = Ext.intrinsic ExtX86.avx "maskload.pd.256" - -maskloadps256 :: Ext.T (LLVM.Value (Ptr ()) -> V8Int32 -> LLVM.CodeGenFunction r (V8Float)) -maskloadps256 = Ext.intrinsic ExtX86.avx "maskload.ps.256" - -loadups128_mask :: Ext.T (LLVM.Value (Ptr ()) -> V4Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Float)) -loadups128_mask = Ext.intrinsic ExtX86.avx512 "mask.loadu.ps.128" - -loadups256_mask :: Ext.T (LLVM.Value (Ptr ()) -> V8Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Float)) -loadups256_mask = Ext.intrinsic ExtX86.avx512 "mask.loadu.ps.256" - -loadups512_mask :: Ext.T (LLVM.Value (Ptr ()) -> V16Float -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Float)) -loadups512_mask = Ext.intrinsic ExtX86.avx512 "mask.loadu.ps.512" - -loadupd128_mask :: Ext.T (LLVM.Value (Ptr ()) -> V2Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Double)) -loadupd128_mask = Ext.intrinsic ExtX86.avx512 "mask.loadu.pd.128" - -loadupd256_mask :: Ext.T (LLVM.Value (Ptr ()) -> V4Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Double)) -loadupd256_mask = Ext.intrinsic ExtX86.avx512 "mask.loadu.pd.256" - -loadupd512_mask :: Ext.T (LLVM.Value (Ptr ()) -> V8Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Double)) -loadupd512_mask = Ext.intrinsic ExtX86.avx512 "mask.loadu.pd.512" - -loadaps128_mask :: Ext.T (LLVM.Value (Ptr ()) -> V4Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Float)) -loadaps128_mask = Ext.intrinsic ExtX86.avx512 "mask.load.ps.128" - -loadaps256_mask :: Ext.T (LLVM.Value (Ptr ()) -> V8Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Float)) -loadaps256_mask = Ext.intrinsic ExtX86.avx512 "mask.load.ps.256" - -loadaps512_mask :: Ext.T (LLVM.Value (Ptr ()) -> V16Float -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Float)) -loadaps512_mask = Ext.intrinsic ExtX86.avx512 "mask.load.ps.512" - -loadapd128_mask :: Ext.T (LLVM.Value (Ptr ()) -> V2Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Double)) -loadapd128_mask = Ext.intrinsic ExtX86.avx512 "mask.load.pd.128" - -loadapd256_mask :: Ext.T (LLVM.Value (Ptr ()) -> V4Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Double)) -loadapd256_mask = Ext.intrinsic ExtX86.avx512 "mask.load.pd.256" - -loadapd512_mask :: Ext.T (LLVM.Value (Ptr ()) -> V8Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Double)) -loadapd512_mask = Ext.intrinsic ExtX86.avx512 "mask.load.pd.512" - -movss_mask :: Ext.T (V4Float -> V4Float -> V4Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Float)) -movss_mask = Ext.intrinsic ExtX86.avx512 "mask.move.ss" - -movsd_mask :: Ext.T (V2Double -> V2Double -> V2Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Double)) -movsd_mask = Ext.intrinsic ExtX86.avx512 "mask.move.sd" - -maskstorepd :: Ext.T (LLVM.Value (Ptr ()) -> V2Int64 -> V2Double -> LLVM.CodeGenFunction r (LLVM.Value ())) -maskstorepd = Ext.intrinsic ExtX86.avx "maskstore.pd" - -maskstoreps :: Ext.T (LLVM.Value (Ptr ()) -> V4Int32 -> V4Float -> LLVM.CodeGenFunction r (LLVM.Value ())) -maskstoreps = Ext.intrinsic ExtX86.avx "maskstore.ps" - -maskstorepd256 :: Ext.T (LLVM.Value (Ptr ()) -> V4Int64 -> V4Double -> LLVM.CodeGenFunction r (LLVM.Value ())) -maskstorepd256 = Ext.intrinsic ExtX86.avx "maskstore.pd.256" - -maskstoreps256 :: Ext.T (LLVM.Value (Ptr ()) -> V8Int32 -> V8Float -> LLVM.CodeGenFunction r (LLVM.Value ())) -maskstoreps256 = Ext.intrinsic ExtX86.avx "maskstore.ps.256" - -storeups512_mask :: Ext.T (LLVM.Value (Ptr ()) -> V16Float -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (LLVM.Value ())) -storeups512_mask = Ext.intrinsic ExtX86.avx512 "mask.storeu.ps.512" - -storeupd512_mask :: Ext.T (LLVM.Value (Ptr ()) -> V8Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (LLVM.Value ())) -storeupd512_mask = Ext.intrinsic ExtX86.avx512 "mask.storeu.pd.512" - -storeaps512_mask :: Ext.T (LLVM.Value (Ptr ()) -> V16Float -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (LLVM.Value ())) -storeaps512_mask = Ext.intrinsic ExtX86.avx512 "mask.store.ps.512" - -storeapd512_mask :: Ext.T (LLVM.Value (Ptr ()) -> V8Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (LLVM.Value ())) -storeapd512_mask = Ext.intrinsic ExtX86.avx512 "mask.store.pd.512" - -storess_mask :: Ext.T (LLVM.Value (Ptr ()) -> V4Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (LLVM.Value ())) -storess_mask = Ext.intrinsic ExtX86.avx512 "mask.store.ss" - -paddsb256 :: Ext.T (V32Int8 -> V32Int8 -> LLVM.CodeGenFunction r (V32Int8)) -paddsb256 = Ext.intrinsic ExtX86.avx2 "padds.b" - -paddsw256 :: Ext.T (V16Int16 -> V16Int16 -> LLVM.CodeGenFunction r (V16Int16)) -paddsw256 = Ext.intrinsic ExtX86.avx2 "padds.w" - -paddusb256 :: Ext.T (V32Word8 -> V32Word8 -> LLVM.CodeGenFunction r (V32Word8)) -paddusb256 = Ext.intrinsic ExtX86.avx2 "paddus.b" - -paddusw256 :: Ext.T (V16Word16 -> V16Word16 -> LLVM.CodeGenFunction r (V16Word16)) -paddusw256 = Ext.intrinsic ExtX86.avx2 "paddus.w" - -psubsb256 :: Ext.T (V32Int8 -> V32Int8 -> LLVM.CodeGenFunction r (V32Int8)) -psubsb256 = Ext.intrinsic ExtX86.avx2 "psubs.b" - -psubsw256 :: Ext.T (V16Int16 -> V16Int16 -> LLVM.CodeGenFunction r (V16Int16)) -psubsw256 = Ext.intrinsic ExtX86.avx2 "psubs.w" - -psubusb256 :: Ext.T (V32Word8 -> V32Word8 -> LLVM.CodeGenFunction r (V32Word8)) -psubusb256 = Ext.intrinsic ExtX86.avx2 "psubus.b" - -psubusw256 :: Ext.T (V16Word16 -> V16Word16 -> LLVM.CodeGenFunction r (V16Word16)) -psubusw256 = Ext.intrinsic ExtX86.avx2 "psubus.w" - -pmulhuw256 :: Ext.T (V16Word16 -> V16Word16 -> LLVM.CodeGenFunction r (V16Word16)) -pmulhuw256 = Ext.intrinsic ExtX86.avx2 "pmulhu.w" - -pmulhw256 :: Ext.T (V16Int16 -> V16Int16 -> LLVM.CodeGenFunction r (V16Int16)) -pmulhw256 = Ext.intrinsic ExtX86.avx2 "pmulh.w" - -pmuludq256 :: Ext.T (V8Word32 -> V8Word32 -> LLVM.CodeGenFunction r (V4Word64)) -pmuludq256 = Ext.intrinsic ExtX86.avx2 "pmulu.dq" - -pmuldq256 :: Ext.T (V8Int32 -> V8Int32 -> LLVM.CodeGenFunction r (V4Int64)) -pmuldq256 = Ext.intrinsic ExtX86.avx2 "pmul.dq" - -pmaddwd256 :: Ext.T (V16Int16 -> V16Int16 -> LLVM.CodeGenFunction r (V8Int32)) -pmaddwd256 = Ext.intrinsic ExtX86.avx2 "pmadd.wd" - -pavgb256 :: Ext.T (V32Int8 -> V32Int8 -> LLVM.CodeGenFunction r (V32Int8)) -pavgb256 = Ext.intrinsic ExtX86.avx2 "pavg.b" - -pavgw256 :: Ext.T (V16Int16 -> V16Int16 -> LLVM.CodeGenFunction r (V16Int16)) -pavgw256 = Ext.intrinsic ExtX86.avx2 "pavg.w" - -psadbw256 :: Ext.T (V32Int8 -> V32Int8 -> LLVM.CodeGenFunction r (V4Int64)) -psadbw256 = Ext.intrinsic ExtX86.avx2 "psad.bw" - -pmaxub256 :: Ext.T (V32Word8 -> V32Word8 -> LLVM.CodeGenFunction r (V32Word8)) -pmaxub256 = Ext.intrinsic ExtX86.avx2 "pmaxu.b" - -pmaxuw256 :: Ext.T (V16Word16 -> V16Word16 -> LLVM.CodeGenFunction r (V16Word16)) -pmaxuw256 = Ext.intrinsic ExtX86.avx2 "pmaxu.w" - -pmaxud256 :: Ext.T (V8Word32 -> V8Word32 -> LLVM.CodeGenFunction r (V8Word32)) -pmaxud256 = Ext.intrinsic ExtX86.avx2 "pmaxu.d" - -pmaxsb256 :: Ext.T (V32Int8 -> V32Int8 -> LLVM.CodeGenFunction r (V32Int8)) -pmaxsb256 = Ext.intrinsic ExtX86.avx2 "pmaxs.b" - -pmaxsw256 :: Ext.T (V16Int16 -> V16Int16 -> LLVM.CodeGenFunction r (V16Int16)) -pmaxsw256 = Ext.intrinsic ExtX86.avx2 "pmaxs.w" - -pmaxsd256 :: Ext.T (V8Int32 -> V8Int32 -> LLVM.CodeGenFunction r (V8Int32)) -pmaxsd256 = Ext.intrinsic ExtX86.avx2 "pmaxs.d" - -pminub256 :: Ext.T (V32Word8 -> V32Word8 -> LLVM.CodeGenFunction r (V32Word8)) -pminub256 = Ext.intrinsic ExtX86.avx2 "pminu.b" - -pminuw256 :: Ext.T (V16Word16 -> V16Word16 -> LLVM.CodeGenFunction r (V16Word16)) -pminuw256 = Ext.intrinsic ExtX86.avx2 "pminu.w" - -pminud256 :: Ext.T (V8Word32 -> V8Word32 -> LLVM.CodeGenFunction r (V8Word32)) -pminud256 = Ext.intrinsic ExtX86.avx2 "pminu.d" - -pminsb256 :: Ext.T (V32Int8 -> V32Int8 -> LLVM.CodeGenFunction r (V32Int8)) -pminsb256 = Ext.intrinsic ExtX86.avx2 "pmins.b" - -pminsw256 :: Ext.T (V16Int16 -> V16Int16 -> LLVM.CodeGenFunction r (V16Int16)) -pminsw256 = Ext.intrinsic ExtX86.avx2 "pmins.w" - -pminsd256 :: Ext.T (V8Int32 -> V8Int32 -> LLVM.CodeGenFunction r (V8Int32)) -pminsd256 = Ext.intrinsic ExtX86.avx2 "pmins.d" - -pmaxsb128_mask :: Ext.T (V16Int8 -> V16Int8 -> V16Int8 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int8)) -pmaxsb128_mask = Ext.intrinsic ExtX86.avx512 "mask.pmaxs.b.128" - -pmaxsb256_mask :: Ext.T (V32Int8 -> V32Int8 -> V32Int8 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V32Int8)) -pmaxsb256_mask = Ext.intrinsic ExtX86.avx512 "mask.pmaxs.b.256" - -pmaxsb512_mask :: Ext.T (V64Int8 -> V64Int8 -> V64Int8 -> LLVM.Value I.Int64 -> LLVM.CodeGenFunction r (V64Int8)) -pmaxsb512_mask = Ext.intrinsic ExtX86.avx512 "mask.pmaxs.b.512" - -pmaxub128_mask :: Ext.T (V16Int8 -> V16Int8 -> V16Int8 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int8)) -pmaxub128_mask = Ext.intrinsic ExtX86.avx512 "mask.pmaxu.b.128" - -pmaxub256_mask :: Ext.T (V32Int8 -> V32Int8 -> V32Int8 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V32Int8)) -pmaxub256_mask = Ext.intrinsic ExtX86.avx512 "mask.pmaxu.b.256" - -pmaxub512_mask :: Ext.T (V64Int8 -> V64Int8 -> V64Int8 -> LLVM.Value I.Int64 -> LLVM.CodeGenFunction r (V64Int8)) -pmaxub512_mask = Ext.intrinsic ExtX86.avx512 "mask.pmaxu.b.512" - -pmaxsw128_mask :: Ext.T (V8Int16 -> V8Int16 -> V8Int16 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int16)) -pmaxsw128_mask = Ext.intrinsic ExtX86.avx512 "mask.pmaxs.w.128" - -pmaxsw256_mask :: Ext.T (V16Int16 -> V16Int16 -> V16Int16 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int16)) -pmaxsw256_mask = Ext.intrinsic ExtX86.avx512 "mask.pmaxs.w.256" - -pmaxsw512_mask :: Ext.T (V32Int16 -> V32Int16 -> V32Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V32Int16)) -pmaxsw512_mask = Ext.intrinsic ExtX86.avx512 "mask.pmaxs.w.512" - -pmaxuw128_mask :: Ext.T (V8Int16 -> V8Int16 -> V8Int16 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int16)) -pmaxuw128_mask = Ext.intrinsic ExtX86.avx512 "mask.pmaxu.w.128" - -pmaxuw256_mask :: Ext.T (V16Int16 -> V16Int16 -> V16Int16 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int16)) -pmaxuw256_mask = Ext.intrinsic ExtX86.avx512 "mask.pmaxu.w.256" - -pmaxuw512_mask :: Ext.T (V32Int16 -> V32Int16 -> V32Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V32Int16)) -pmaxuw512_mask = Ext.intrinsic ExtX86.avx512 "mask.pmaxu.w.512" - -pminsb128_mask :: Ext.T (V16Int8 -> V16Int8 -> V16Int8 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int8)) -pminsb128_mask = Ext.intrinsic ExtX86.avx512 "mask.pmins.b.128" - -pminsb256_mask :: Ext.T (V32Int8 -> V32Int8 -> V32Int8 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V32Int8)) -pminsb256_mask = Ext.intrinsic ExtX86.avx512 "mask.pmins.b.256" - -pminsb512_mask :: Ext.T (V64Int8 -> V64Int8 -> V64Int8 -> LLVM.Value I.Int64 -> LLVM.CodeGenFunction r (V64Int8)) -pminsb512_mask = Ext.intrinsic ExtX86.avx512 "mask.pmins.b.512" - -pminub128_mask :: Ext.T (V16Int8 -> V16Int8 -> V16Int8 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int8)) -pminub128_mask = Ext.intrinsic ExtX86.avx512 "mask.pminu.b.128" - -pminub256_mask :: Ext.T (V32Int8 -> V32Int8 -> V32Int8 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V32Int8)) -pminub256_mask = Ext.intrinsic ExtX86.avx512 "mask.pminu.b.256" - -pminub512_mask :: Ext.T (V64Int8 -> V64Int8 -> V64Int8 -> LLVM.Value I.Int64 -> LLVM.CodeGenFunction r (V64Int8)) -pminub512_mask = Ext.intrinsic ExtX86.avx512 "mask.pminu.b.512" - -pminsw128_mask :: Ext.T (V8Int16 -> V8Int16 -> V8Int16 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int16)) -pminsw128_mask = Ext.intrinsic ExtX86.avx512 "mask.pmins.w.128" - -pminsw256_mask :: Ext.T (V16Int16 -> V16Int16 -> V16Int16 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int16)) -pminsw256_mask = Ext.intrinsic ExtX86.avx512 "mask.pmins.w.256" - -pminsw512_mask :: Ext.T (V32Int16 -> V32Int16 -> V32Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V32Int16)) -pminsw512_mask = Ext.intrinsic ExtX86.avx512 "mask.pmins.w.512" - -pminuw128_mask :: Ext.T (V8Int16 -> V8Int16 -> V8Int16 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int16)) -pminuw128_mask = Ext.intrinsic ExtX86.avx512 "mask.pminu.w.128" - -pminuw256_mask :: Ext.T (V16Int16 -> V16Int16 -> V16Int16 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int16)) -pminuw256_mask = Ext.intrinsic ExtX86.avx512 "mask.pminu.w.256" - -pminuw512_mask :: Ext.T (V32Int16 -> V32Int16 -> V32Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V32Int16)) -pminuw512_mask = Ext.intrinsic ExtX86.avx512 "mask.pminu.w.512" - -pmaxud512_mask :: Ext.T (V16Int32 -> V16Int32 -> V16Int32 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int32)) -pmaxud512_mask = Ext.intrinsic ExtX86.avx512 "mask.pmaxu.d.512" - -pmaxud256_mask :: Ext.T (V8Int32 -> V8Int32 -> V8Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int32)) -pmaxud256_mask = Ext.intrinsic ExtX86.avx512 "mask.pmaxu.d.256" - -pmaxud128_mask :: Ext.T (V4Int32 -> V4Int32 -> V4Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int32)) -pmaxud128_mask = Ext.intrinsic ExtX86.avx512 "mask.pmaxu.d.128" - -pmaxsd512_mask :: Ext.T (V16Int32 -> V16Int32 -> V16Int32 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int32)) -pmaxsd512_mask = Ext.intrinsic ExtX86.avx512 "mask.pmaxs.d.512" - -pmaxsd256_mask :: Ext.T (V8Int32 -> V8Int32 -> V8Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int32)) -pmaxsd256_mask = Ext.intrinsic ExtX86.avx512 "mask.pmaxs.d.256" - -pmaxsd128_mask :: Ext.T (V4Int32 -> V4Int32 -> V4Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int32)) -pmaxsd128_mask = Ext.intrinsic ExtX86.avx512 "mask.pmaxs.d.128" - -pmaxuq512_mask :: Ext.T (V8Int64 -> V8Int64 -> V8Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int64)) -pmaxuq512_mask = Ext.intrinsic ExtX86.avx512 "mask.pmaxu.q.512" - -pmaxuq256_mask :: Ext.T (V4Int64 -> V4Int64 -> V4Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int64)) -pmaxuq256_mask = Ext.intrinsic ExtX86.avx512 "mask.pmaxu.q.256" - -pmaxuq128_mask :: Ext.T (V2Int64 -> V2Int64 -> V2Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Int64)) -pmaxuq128_mask = Ext.intrinsic ExtX86.avx512 "mask.pmaxu.q.128" - -pmaxsq512_mask :: Ext.T (V8Int64 -> V8Int64 -> V8Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int64)) -pmaxsq512_mask = Ext.intrinsic ExtX86.avx512 "mask.pmaxs.q.512" - -pmaxsq256_mask :: Ext.T (V4Int64 -> V4Int64 -> V4Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int64)) -pmaxsq256_mask = Ext.intrinsic ExtX86.avx512 "mask.pmaxs.q.256" - -pmaxsq128_mask :: Ext.T (V2Int64 -> V2Int64 -> V2Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Int64)) -pmaxsq128_mask = Ext.intrinsic ExtX86.avx512 "mask.pmaxs.q.128" - -pminud512_mask :: Ext.T (V16Int32 -> V16Int32 -> V16Int32 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int32)) -pminud512_mask = Ext.intrinsic ExtX86.avx512 "mask.pminu.d.512" - -pminud256_mask :: Ext.T (V8Int32 -> V8Int32 -> V8Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int32)) -pminud256_mask = Ext.intrinsic ExtX86.avx512 "mask.pminu.d.256" - -pminud128_mask :: Ext.T (V4Int32 -> V4Int32 -> V4Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int32)) -pminud128_mask = Ext.intrinsic ExtX86.avx512 "mask.pminu.d.128" - -pminsd512_mask :: Ext.T (V16Int32 -> V16Int32 -> V16Int32 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int32)) -pminsd512_mask = Ext.intrinsic ExtX86.avx512 "mask.pmins.d.512" - -pminsd256_mask :: Ext.T (V8Int32 -> V8Int32 -> V8Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int32)) -pminsd256_mask = Ext.intrinsic ExtX86.avx512 "mask.pmins.d.256" - -pminsd128_mask :: Ext.T (V4Int32 -> V4Int32 -> V4Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int32)) -pminsd128_mask = Ext.intrinsic ExtX86.avx512 "mask.pmins.d.128" - -pminuq512_mask :: Ext.T (V8Int64 -> V8Int64 -> V8Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int64)) -pminuq512_mask = Ext.intrinsic ExtX86.avx512 "mask.pminu.q.512" - -pminuq256_mask :: Ext.T (V4Int64 -> V4Int64 -> V4Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int64)) -pminuq256_mask = Ext.intrinsic ExtX86.avx512 "mask.pminu.q.256" - -pminuq128_mask :: Ext.T (V2Int64 -> V2Int64 -> V2Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Int64)) -pminuq128_mask = Ext.intrinsic ExtX86.avx512 "mask.pminu.q.128" - -pminsq512_mask :: Ext.T (V8Int64 -> V8Int64 -> V8Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int64)) -pminsq512_mask = Ext.intrinsic ExtX86.avx512 "mask.pmins.q.512" - -pminsq256_mask :: Ext.T (V4Int64 -> V4Int64 -> V4Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int64)) -pminsq256_mask = Ext.intrinsic ExtX86.avx512 "mask.pmins.q.256" - -pminsq128_mask :: Ext.T (V2Int64 -> V2Int64 -> V2Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Int64)) -pminsq128_mask = Ext.intrinsic ExtX86.avx512 "mask.pmins.q.128" - -psllw256 :: Ext.T (V16Int16 -> V8Int16 -> LLVM.CodeGenFunction r (V16Int16)) -psllw256 = Ext.intrinsic ExtX86.avx2 "psll.w" - -pslld256 :: Ext.T (V8Int32 -> V4Int32 -> LLVM.CodeGenFunction r (V8Int32)) -pslld256 = Ext.intrinsic ExtX86.avx2 "psll.d" - -psllq256 :: Ext.T (V4Int64 -> V2Int64 -> LLVM.CodeGenFunction r (V4Int64)) -psllq256 = Ext.intrinsic ExtX86.avx2 "psll.q" - -psrlw256 :: Ext.T (V16Int16 -> V8Int16 -> LLVM.CodeGenFunction r (V16Int16)) -psrlw256 = Ext.intrinsic ExtX86.avx2 "psrl.w" - -psrld256 :: Ext.T (V8Int32 -> V4Int32 -> LLVM.CodeGenFunction r (V8Int32)) -psrld256 = Ext.intrinsic ExtX86.avx2 "psrl.d" - -psrlq256 :: Ext.T (V4Int64 -> V2Int64 -> LLVM.CodeGenFunction r (V4Int64)) -psrlq256 = Ext.intrinsic ExtX86.avx2 "psrl.q" - -psraw256 :: Ext.T (V16Int16 -> V8Int16 -> LLVM.CodeGenFunction r (V16Int16)) -psraw256 = Ext.intrinsic ExtX86.avx2 "psra.w" - -psrad256 :: Ext.T (V8Int32 -> V4Int32 -> LLVM.CodeGenFunction r (V8Int32)) -psrad256 = Ext.intrinsic ExtX86.avx2 "psra.d" - -psllwi256 :: Ext.T (V16Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V16Int16)) -psllwi256 = Ext.intrinsic ExtX86.avx2 "pslli.w" - -pslldi256 :: Ext.T (V8Int32 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V8Int32)) -pslldi256 = Ext.intrinsic ExtX86.avx2 "pslli.d" - -psllqi256 :: Ext.T (V4Int64 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V4Int64)) -psllqi256 = Ext.intrinsic ExtX86.avx2 "pslli.q" - -psrlwi256 :: Ext.T (V16Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V16Int16)) -psrlwi256 = Ext.intrinsic ExtX86.avx2 "psrli.w" - -psrldi256 :: Ext.T (V8Int32 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V8Int32)) -psrldi256 = Ext.intrinsic ExtX86.avx2 "psrli.d" - -psrlqi256 :: Ext.T (V4Int64 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V4Int64)) -psrlqi256 = Ext.intrinsic ExtX86.avx2 "psrli.q" - -psrawi256 :: Ext.T (V16Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V16Int16)) -psrawi256 = Ext.intrinsic ExtX86.avx2 "psrai.w" - -psradi256 :: Ext.T (V8Int32 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V8Int32)) -psradi256 = Ext.intrinsic ExtX86.avx2 "psrai.d" - -pslldi512 :: Ext.T (V16Int32 -> LLVM.Value I.Int32 -> V16Int32 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int32)) -pslldi512 = Ext.intrinsic ExtX86.avx512 "mask.pslli.d" - -psllqi512 :: Ext.T (V8Int64 -> LLVM.Value I.Int32 -> V8Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int64)) -psllqi512 = Ext.intrinsic ExtX86.avx512 "mask.pslli.q" - -psrldi512 :: Ext.T (V16Int32 -> LLVM.Value I.Int32 -> V16Int32 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int32)) -psrldi512 = Ext.intrinsic ExtX86.avx512 "mask.psrli.d" - -psrlqi512 :: Ext.T (V8Int64 -> LLVM.Value I.Int32 -> V8Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int64)) -psrlqi512 = Ext.intrinsic ExtX86.avx512 "mask.psrli.q" - -psradi512 :: Ext.T (V16Int32 -> LLVM.Value I.Int32 -> V16Int32 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int32)) -psradi512 = Ext.intrinsic ExtX86.avx512 "mask.psrai.d" - -psraqi512 :: Ext.T (V8Int64 -> LLVM.Value I.Int32 -> V8Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int64)) -psraqi512 = Ext.intrinsic ExtX86.avx512 "mask.psrai.q" - -psrlw128_mask :: Ext.T (V8Int16 -> V8Int16 -> V8Int16 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int16)) -psrlw128_mask = Ext.intrinsic ExtX86.avx512 "mask.psrl.w.128" - -psrlw256_mask :: Ext.T (V16Int16 -> V8Int16 -> V16Int16 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int16)) -psrlw256_mask = Ext.intrinsic ExtX86.avx512 "mask.psrl.w.256" - -psrlw512_mask :: Ext.T (V32Int16 -> V8Int16 -> V32Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V32Int16)) -psrlw512_mask = Ext.intrinsic ExtX86.avx512 "mask.psrl.w.512" - -psrlwi128_mask :: Ext.T (V8Int16 -> LLVM.Value I.Int8 -> V8Int16 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int16)) -psrlwi128_mask = Ext.intrinsic ExtX86.avx512 "mask.psrl.wi.128" - -psrlwi256_mask :: Ext.T (V16Int16 -> LLVM.Value I.Int8 -> V16Int16 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int16)) -psrlwi256_mask = Ext.intrinsic ExtX86.avx512 "mask.psrl.wi.256" - -psrlwi512_mask :: Ext.T (V32Int16 -> LLVM.Value I.Int8 -> V32Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V32Int16)) -psrlwi512_mask = Ext.intrinsic ExtX86.avx512 "mask.psrl.wi.512" - -psraw128_mask :: Ext.T (V8Int16 -> V8Int16 -> V8Int16 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int16)) -psraw128_mask = Ext.intrinsic ExtX86.avx512 "mask.psra.w.128" - -psraw256_mask :: Ext.T (V16Int16 -> V8Int16 -> V16Int16 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int16)) -psraw256_mask = Ext.intrinsic ExtX86.avx512 "mask.psra.w.256" - -psraw512_mask :: Ext.T (V32Int16 -> V8Int16 -> V32Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V32Int16)) -psraw512_mask = Ext.intrinsic ExtX86.avx512 "mask.psra.w.512" - -psrawi128_mask :: Ext.T (V8Int16 -> LLVM.Value I.Int8 -> V8Int16 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int16)) -psrawi128_mask = Ext.intrinsic ExtX86.avx512 "mask.psra.wi.128" - -psrawi256_mask :: Ext.T (V16Int16 -> LLVM.Value I.Int8 -> V16Int16 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int16)) -psrawi256_mask = Ext.intrinsic ExtX86.avx512 "mask.psra.wi.256" - -psrawi512_mask :: Ext.T (V32Int16 -> LLVM.Value I.Int8 -> V32Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V32Int16)) -psrawi512_mask = Ext.intrinsic ExtX86.avx512 "mask.psra.wi.512" - -pslld512_mask :: Ext.T (V16Int32 -> V4Int32 -> V16Int32 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int32)) -pslld512_mask = Ext.intrinsic ExtX86.avx512 "mask.psll.d" - -psllq512_mask :: Ext.T (V8Int64 -> V2Int64 -> V8Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int64)) -psllq512_mask = Ext.intrinsic ExtX86.avx512 "mask.psll.q" - -psrld512_mask :: Ext.T (V16Int32 -> V4Int32 -> V16Int32 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int32)) -psrld512_mask = Ext.intrinsic ExtX86.avx512 "mask.psrl.d" - -psrlq512_mask :: Ext.T (V8Int64 -> V2Int64 -> V8Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int64)) -psrlq512_mask = Ext.intrinsic ExtX86.avx512 "mask.psrl.q" - -psrad512_mask :: Ext.T (V16Int32 -> V4Int32 -> V16Int32 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int32)) -psrad512_mask = Ext.intrinsic ExtX86.avx512 "mask.psra.d" - -psraq512_mask :: Ext.T (V8Int64 -> V2Int64 -> V8Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int64)) -psraq512_mask = Ext.intrinsic ExtX86.avx512 "mask.psra.q" - -psllw128_mask :: Ext.T (V8Int16 -> V8Int16 -> V8Int16 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int16)) -psllw128_mask = Ext.intrinsic ExtX86.avx512 "mask.psll.w.128" - -psllw256_mask :: Ext.T (V16Int16 -> V8Int16 -> V16Int16 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int16)) -psllw256_mask = Ext.intrinsic ExtX86.avx512 "mask.psll.w.256" - -psllw512_mask :: Ext.T (V32Int16 -> V8Int16 -> V32Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V32Int16)) -psllw512_mask = Ext.intrinsic ExtX86.avx512 "mask.psll.w.512" - -psllwi128_mask :: Ext.T (V8Int16 -> LLVM.Value I.Int8 -> V8Int16 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int16)) -psllwi128_mask = Ext.intrinsic ExtX86.avx512 "mask.psll.wi.128" - -psllwi256_mask :: Ext.T (V16Int16 -> LLVM.Value I.Int8 -> V16Int16 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int16)) -psllwi256_mask = Ext.intrinsic ExtX86.avx512 "mask.psll.wi.256" - -psllwi512_mask :: Ext.T (V32Int16 -> LLVM.Value I.Int8 -> V32Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V32Int16)) -psllwi512_mask = Ext.intrinsic ExtX86.avx512 "mask.psll.wi.512" - -psllv16hi_mask :: Ext.T (V16Int16 -> V16Int16 -> V16Int16 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int16)) -psllv16hi_mask = Ext.intrinsic ExtX86.avx512 "mask.psllv16.hi" - -psllv2di_mask :: Ext.T (V2Int64 -> V2Int64 -> V2Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Int64)) -psllv2di_mask = Ext.intrinsic ExtX86.avx512 "mask.psllv2.di" - -psllv32hi_mask :: Ext.T (V32Int16 -> V32Int16 -> V32Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V32Int16)) -psllv32hi_mask = Ext.intrinsic ExtX86.avx512 "mask.psllv32hi" - -psllv4di_mask :: Ext.T (V4Int64 -> V4Int64 -> V4Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int64)) -psllv4di_mask = Ext.intrinsic ExtX86.avx512 "mask.psllv4.di" - -psllv4si_mask :: Ext.T (V4Int32 -> V4Int32 -> V4Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int32)) -psllv4si_mask = Ext.intrinsic ExtX86.avx512 "mask.psllv4.si" - -psllv8hi_mask :: Ext.T (V8Int16 -> V8Int16 -> V8Int16 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int16)) -psllv8hi_mask = Ext.intrinsic ExtX86.avx512 "mask.psllv8.hi" - -psllv8si_mask :: Ext.T (V8Int32 -> V8Int32 -> V8Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int32)) -psllv8si_mask = Ext.intrinsic ExtX86.avx512 "mask.psllv8.si" - -psrad128_mask :: Ext.T (V4Int32 -> V4Int32 -> V4Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int32)) -psrad128_mask = Ext.intrinsic ExtX86.avx512 "mask.psra.d.128" - -psrad256_mask :: Ext.T (V8Int32 -> V4Int32 -> V8Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int32)) -psrad256_mask = Ext.intrinsic ExtX86.avx512 "mask.psra.d.256" - -psradi128_mask :: Ext.T (V4Int32 -> LLVM.Value I.Int8 -> V4Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int32)) -psradi128_mask = Ext.intrinsic ExtX86.avx512 "mask.psra.di.128" - -psradi256_mask :: Ext.T (V8Int32 -> LLVM.Value I.Int8 -> V8Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int32)) -psradi256_mask = Ext.intrinsic ExtX86.avx512 "mask.psra.di.256" - -psradi512_mask :: Ext.T (V16Int32 -> LLVM.Value I.Int8 -> V16Int32 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int32)) -psradi512_mask = Ext.intrinsic ExtX86.avx512 "mask.psra.di.512" - -psraq128_mask :: Ext.T (V2Int64 -> V2Int64 -> V2Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Int64)) -psraq128_mask = Ext.intrinsic ExtX86.avx512 "mask.psra.q.128" - -psraq256_mask :: Ext.T (V4Int64 -> V2Int64 -> V4Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int64)) -psraq256_mask = Ext.intrinsic ExtX86.avx512 "mask.psra.q.256" - -psraqi128_mask :: Ext.T (V2Int64 -> LLVM.Value I.Int8 -> V2Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Int64)) -psraqi128_mask = Ext.intrinsic ExtX86.avx512 "mask.psra.qi.128" - -psraqi256_mask :: Ext.T (V4Int64 -> LLVM.Value I.Int8 -> V4Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int64)) -psraqi256_mask = Ext.intrinsic ExtX86.avx512 "mask.psra.qi.256" - -psraqi512_mask :: Ext.T (V8Int64 -> LLVM.Value I.Int8 -> V8Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int64)) -psraqi512_mask = Ext.intrinsic ExtX86.avx512 "mask.psra.qi.512" - -psrld128_mask :: Ext.T (V4Int32 -> V4Int32 -> V4Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int32)) -psrld128_mask = Ext.intrinsic ExtX86.avx512 "mask.psrl.d.128" - -psrld256_mask :: Ext.T (V8Int32 -> V4Int32 -> V8Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int32)) -psrld256_mask = Ext.intrinsic ExtX86.avx512 "mask.psrl.d.256" - -psrldi128_mask :: Ext.T (V4Int32 -> LLVM.Value I.Int8 -> V4Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int32)) -psrldi128_mask = Ext.intrinsic ExtX86.avx512 "mask.psrl.di.128" - -psrldi256_mask :: Ext.T (V8Int32 -> LLVM.Value I.Int8 -> V8Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int32)) -psrldi256_mask = Ext.intrinsic ExtX86.avx512 "mask.psrl.di.256" - -psrldi512_mask :: Ext.T (V16Int32 -> LLVM.Value I.Int8 -> V16Int32 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int32)) -psrldi512_mask = Ext.intrinsic ExtX86.avx512 "mask.psrl.di.512" - -psrlq128_mask :: Ext.T (V2Int64 -> V2Int64 -> V2Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Int64)) -psrlq128_mask = Ext.intrinsic ExtX86.avx512 "mask.psrl.q.128" - -psrlq256_mask :: Ext.T (V4Int64 -> V2Int64 -> V4Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int64)) -psrlq256_mask = Ext.intrinsic ExtX86.avx512 "mask.psrl.q.256" - -psrlqi128_mask :: Ext.T (V2Int64 -> LLVM.Value I.Int8 -> V2Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Int64)) -psrlqi128_mask = Ext.intrinsic ExtX86.avx512 "mask.psrl.qi.128" - -psrlqi256_mask :: Ext.T (V4Int64 -> LLVM.Value I.Int8 -> V4Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int64)) -psrlqi256_mask = Ext.intrinsic ExtX86.avx512 "mask.psrl.qi.256" - -psrlqi512_mask :: Ext.T (V8Int64 -> LLVM.Value I.Int8 -> V8Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int64)) -psrlqi512_mask = Ext.intrinsic ExtX86.avx512 "mask.psrl.qi.512" - -packsswb256 :: Ext.T (V16Int16 -> V16Int16 -> LLVM.CodeGenFunction r (V32Int8)) -packsswb256 = Ext.intrinsic ExtX86.avx2 "packsswb" - -packssdw256 :: Ext.T (V8Int32 -> V8Int32 -> LLVM.CodeGenFunction r (V16Int16)) -packssdw256 = Ext.intrinsic ExtX86.avx2 "packssdw" - -packuswb256 :: Ext.T (V16Word16 -> V16Word16 -> LLVM.CodeGenFunction r (V32Word8)) -packuswb256 = Ext.intrinsic ExtX86.avx2 "packuswb" - -packusdw256 :: Ext.T (V8Word32 -> V8Word32 -> LLVM.CodeGenFunction r (V16Word16)) -packusdw256 = Ext.intrinsic ExtX86.avx2 "packusdw" - -pabsb256 :: Ext.T (V32Int8 -> LLVM.CodeGenFunction r (V32Int8)) -pabsb256 = Ext.intrinsic ExtX86.avx2 "pabs.b" - -pabsw256 :: Ext.T (V16Int16 -> LLVM.CodeGenFunction r (V16Int16)) -pabsw256 = Ext.intrinsic ExtX86.avx2 "pabs.w" - -pabsd256 :: Ext.T (V8Int32 -> LLVM.CodeGenFunction r (V8Int32)) -pabsd256 = Ext.intrinsic ExtX86.avx2 "pabs.d" - -pabsb128_mask :: Ext.T (V16Int8 -> V16Int8 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int8)) -pabsb128_mask = Ext.intrinsic ExtX86.avx512 "mask.pabs.b.128" - -pabsb256_mask :: Ext.T (V32Int8 -> V32Int8 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V32Int8)) -pabsb256_mask = Ext.intrinsic ExtX86.avx512 "mask.pabs.b.256" - -pabsb512_mask :: Ext.T (V64Int8 -> V64Int8 -> LLVM.Value I.Int64 -> LLVM.CodeGenFunction r (V64Int8)) -pabsb512_mask = Ext.intrinsic ExtX86.avx512 "mask.pabs.b.512" - -pabsd128_mask :: Ext.T (V4Int32 -> V4Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int32)) -pabsd128_mask = Ext.intrinsic ExtX86.avx512 "mask.pabs.d.128" - -pabsd256_mask :: Ext.T (V8Int32 -> V8Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int32)) -pabsd256_mask = Ext.intrinsic ExtX86.avx512 "mask.pabs.d.256" - -pabsd512_mask :: Ext.T (V16Int32 -> V16Int32 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int32)) -pabsd512_mask = Ext.intrinsic ExtX86.avx512 "mask.pabs.d.512" - -pabsq128_mask :: Ext.T (V2Int64 -> V2Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Int64)) -pabsq128_mask = Ext.intrinsic ExtX86.avx512 "mask.pabs.q.128" - -pabsq256_mask :: Ext.T (V4Int64 -> V4Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int64)) -pabsq256_mask = Ext.intrinsic ExtX86.avx512 "mask.pabs.q.256" - -pabsq512_mask :: Ext.T (V8Int64 -> V8Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int64)) -pabsq512_mask = Ext.intrinsic ExtX86.avx512 "mask.pabs.q.512" - -pabsw128_mask :: Ext.T (V8Int16 -> V8Int16 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int16)) -pabsw128_mask = Ext.intrinsic ExtX86.avx512 "mask.pabs.w.128" - -pabsw256_mask :: Ext.T (V16Int16 -> V16Int16 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int16)) -pabsw256_mask = Ext.intrinsic ExtX86.avx512 "mask.pabs.w.256" - -pabsw512_mask :: Ext.T (V32Int16 -> V32Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V32Int16)) -pabsw512_mask = Ext.intrinsic ExtX86.avx512 "mask.pabs.w.512" - -phaddw256 :: Ext.T (V16Int16 -> V16Int16 -> LLVM.CodeGenFunction r (V16Int16)) -phaddw256 = Ext.intrinsic ExtX86.avx2 "phadd.w" - -phaddd256 :: Ext.T (V8Int32 -> V8Int32 -> LLVM.CodeGenFunction r (V8Int32)) -phaddd256 = Ext.intrinsic ExtX86.avx2 "phadd.d" - -phaddsw256 :: Ext.T (V16Int16 -> V16Int16 -> LLVM.CodeGenFunction r (V16Int16)) -phaddsw256 = Ext.intrinsic ExtX86.avx2 "phadd.sw" - -phsubw256 :: Ext.T (V16Int16 -> V16Int16 -> LLVM.CodeGenFunction r (V16Int16)) -phsubw256 = Ext.intrinsic ExtX86.avx2 "phsub.w" - -phsubd256 :: Ext.T (V8Int32 -> V8Int32 -> LLVM.CodeGenFunction r (V8Int32)) -phsubd256 = Ext.intrinsic ExtX86.avx2 "phsub.d" - -phsubsw256 :: Ext.T (V16Int16 -> V16Int16 -> LLVM.CodeGenFunction r (V16Int16)) -phsubsw256 = Ext.intrinsic ExtX86.avx2 "phsub.sw" - -pmaddubsw256 :: Ext.T (V32Word8 -> V32Word8 -> LLVM.CodeGenFunction r (V16Word16)) -pmaddubsw256 = Ext.intrinsic ExtX86.avx2 "pmadd.ub.sw" - -psignb256 :: Ext.T (V32Int8 -> V32Int8 -> LLVM.CodeGenFunction r (V32Int8)) -psignb256 = Ext.intrinsic ExtX86.avx2 "psign.b" - -psignw256 :: Ext.T (V16Int16 -> V16Int16 -> LLVM.CodeGenFunction r (V16Int16)) -psignw256 = Ext.intrinsic ExtX86.avx2 "psign.w" - -psignd256 :: Ext.T (V8Int32 -> V8Int32 -> LLVM.CodeGenFunction r (V8Int32)) -psignd256 = Ext.intrinsic ExtX86.avx2 "psign.d" - -pmulhrsw256 :: Ext.T (V16Int16 -> V16Int16 -> LLVM.CodeGenFunction r (V16Int16)) -pmulhrsw256 = Ext.intrinsic ExtX86.avx2 "pmul.hr.sw" - -pmulhrsw128_mask :: Ext.T (V8Int16 -> V8Int16 -> V8Int16 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int16)) -pmulhrsw128_mask = Ext.intrinsic ExtX86.avx512 "mask.pmul.hr.sw.128" - -pmulhrsw256_mask :: Ext.T (V16Int16 -> V16Int16 -> V16Int16 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int16)) -pmulhrsw256_mask = Ext.intrinsic ExtX86.avx512 "mask.pmul.hr.sw.256" - -pmulhrsw512_mask :: Ext.T (V32Int16 -> V32Int16 -> V32Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V32Int16)) -pmulhrsw512_mask = Ext.intrinsic ExtX86.avx512 "mask.pmul.hr.sw.512" - -pmovsxbd256 :: Ext.T (V16Int8 -> LLVM.CodeGenFunction r (V8Int32)) -pmovsxbd256 = Ext.intrinsic ExtX86.avx2 "pmovsxbd" - -pmovsxbq256 :: Ext.T (V16Int8 -> LLVM.CodeGenFunction r (V4Int64)) -pmovsxbq256 = Ext.intrinsic ExtX86.avx2 "pmovsxbq" - -pmovsxbw256 :: Ext.T (V16Int8 -> LLVM.CodeGenFunction r (V16Int16)) -pmovsxbw256 = Ext.intrinsic ExtX86.avx2 "pmovsxbw" - -pmovsxdq256 :: Ext.T (V4Int32 -> LLVM.CodeGenFunction r (V4Int64)) -pmovsxdq256 = Ext.intrinsic ExtX86.avx2 "pmovsxdq" - -pmovsxwd256 :: Ext.T (V8Int16 -> LLVM.CodeGenFunction r (V8Int32)) -pmovsxwd256 = Ext.intrinsic ExtX86.avx2 "pmovsxwd" - -pmovsxwq256 :: Ext.T (V8Int16 -> LLVM.CodeGenFunction r (V4Int64)) -pmovsxwq256 = Ext.intrinsic ExtX86.avx2 "pmovsxwq" - -pmovzxbd256 :: Ext.T (V16Int8 -> LLVM.CodeGenFunction r (V8Int32)) -pmovzxbd256 = Ext.intrinsic ExtX86.avx2 "pmovzxbd" - -pmovzxbq256 :: Ext.T (V16Int8 -> LLVM.CodeGenFunction r (V4Int64)) -pmovzxbq256 = Ext.intrinsic ExtX86.avx2 "pmovzxbq" - -pmovzxbw256 :: Ext.T (V16Int8 -> LLVM.CodeGenFunction r (V16Int16)) -pmovzxbw256 = Ext.intrinsic ExtX86.avx2 "pmovzxbw" - -pmovzxdq256 :: Ext.T (V4Int32 -> LLVM.CodeGenFunction r (V4Int64)) -pmovzxdq256 = Ext.intrinsic ExtX86.avx2 "pmovzxdq" - -pmovzxwd256 :: Ext.T (V8Int16 -> LLVM.CodeGenFunction r (V8Int32)) -pmovzxwd256 = Ext.intrinsic ExtX86.avx2 "pmovzxwd" - -pmovzxwq256 :: Ext.T (V8Int16 -> LLVM.CodeGenFunction r (V4Int64)) -pmovzxwq256 = Ext.intrinsic ExtX86.avx2 "pmovzxwq" - -pblendvb256 :: Ext.T (V32Int8 -> V32Int8 -> V32Int8 -> LLVM.CodeGenFunction r (V32Int8)) -pblendvb256 = Ext.intrinsic ExtX86.avx2 "pblendvb" - -pbroadcastd512_gpr_mask :: Ext.T (LLVM.Value I.Int32 -> V16Int32 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int32)) -pbroadcastd512_gpr_mask = Ext.intrinsic ExtX86.avx512 "mask.pbroadcast.d.gpr.512" - -pbroadcastq512_gpr_mask :: Ext.T (LLVM.Value I.Int64 -> V8Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int64)) -pbroadcastq512_gpr_mask = Ext.intrinsic ExtX86.avx512 "mask.pbroadcast.q.gpr.512" - -pbroadcastq512_mem_mask :: Ext.T (LLVM.Value I.Int64 -> V8Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int64)) -pbroadcastq512_mem_mask = Ext.intrinsic ExtX86.avx512 "mask.pbroadcast.q.mem.512" - -permvarsi256 :: Ext.T (V8Int32 -> V8Int32 -> LLVM.CodeGenFunction r (V8Int32)) -permvarsi256 = Ext.intrinsic ExtX86.avx2 "permd" - -permvarsf256 :: Ext.T (V8Float -> V8Int32 -> LLVM.CodeGenFunction r (V8Float)) -permvarsf256 = Ext.intrinsic ExtX86.avx2 "permps" - -permti256 :: Ext.T (V4Int64 -> V4Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int64)) -permti256 = Ext.intrinsic ExtX86.avx2 "vperm2i128" - -extractf32x4_mask :: Ext.T (V16Float -> LLVM.Value I.Int32 -> V4Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Float)) -extractf32x4_mask = Ext.intrinsic ExtX86.avx512 "mask.vextractf32x4.512" - -extracti32x4_mask :: Ext.T (V16Int32 -> LLVM.Value I.Int32 -> V4Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int32)) -extracti32x4_mask = Ext.intrinsic ExtX86.avx512 "mask.vextracti32x4.512" - -extractf32x4_256_mask :: Ext.T (V8Float -> LLVM.Value I.Int32 -> V4Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Float)) -extractf32x4_256_mask = Ext.intrinsic ExtX86.avx512 "mask.vextractf32x4.256" - -extracti32x4_256_mask :: Ext.T (V8Int32 -> LLVM.Value I.Int32 -> V4Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int32)) -extracti32x4_256_mask = Ext.intrinsic ExtX86.avx512 "mask.vextracti32x4.256" - -extractf64x2_256_mask :: Ext.T (V4Double -> LLVM.Value I.Int32 -> V2Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Double)) -extractf64x2_256_mask = Ext.intrinsic ExtX86.avx512 "mask.vextractf64x2.256" - -extracti64x2_256_mask :: Ext.T (V4Int64 -> LLVM.Value I.Int32 -> V2Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Int64)) -extracti64x2_256_mask = Ext.intrinsic ExtX86.avx512 "mask.vextracti64x2.256" - -extractf64x2_512_mask :: Ext.T (V8Double -> LLVM.Value I.Int32 -> V2Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Double)) -extractf64x2_512_mask = Ext.intrinsic ExtX86.avx512 "mask.vextractf64x2.512" - -extracti64x2_512_mask :: Ext.T (V8Int64 -> LLVM.Value I.Int32 -> V2Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Int64)) -extracti64x2_512_mask = Ext.intrinsic ExtX86.avx512 "mask.vextracti64x2.512" - -extractf32x8_mask :: Ext.T (V16Float -> LLVM.Value I.Int32 -> V8Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Float)) -extractf32x8_mask = Ext.intrinsic ExtX86.avx512 "mask.vextractf32x8.512" - -extracti32x8_mask :: Ext.T (V16Int32 -> LLVM.Value I.Int32 -> V8Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int32)) -extracti32x8_mask = Ext.intrinsic ExtX86.avx512 "mask.vextracti32x8.512" - -extractf64x4_mask :: Ext.T (V8Double -> LLVM.Value I.Int32 -> V4Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Double)) -extractf64x4_mask = Ext.intrinsic ExtX86.avx512 "mask.vextractf64x4.512" - -extracti64x4_mask :: Ext.T (V8Int64 -> LLVM.Value I.Int32 -> V4Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int64)) -extracti64x4_mask = Ext.intrinsic ExtX86.avx512 "mask.vextracti64x4.512" - -insertf32x4_256_mask :: Ext.T (V8Float -> V4Float -> LLVM.Value I.Int32 -> V8Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Float)) -insertf32x4_256_mask = Ext.intrinsic ExtX86.avx512 "mask.insertf32x4.256" - -insertf32x4_512_mask :: Ext.T (V16Float -> V4Float -> LLVM.Value I.Int32 -> V16Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V16Float)) -insertf32x4_512_mask = Ext.intrinsic ExtX86.avx512 "mask.insertf32x4.512" - -insertf32x8_mask :: Ext.T (V16Float -> V8Float -> LLVM.Value I.Int32 -> V16Float -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Float)) -insertf32x8_mask = Ext.intrinsic ExtX86.avx512 "mask.insertf32x8.512" - -insertf64x2_256_mask :: Ext.T (V4Double -> V2Double -> LLVM.Value I.Int32 -> V4Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Double)) -insertf64x2_256_mask = Ext.intrinsic ExtX86.avx512 "mask.insertf64x2.256" - -insertf64x2_512_mask :: Ext.T (V8Double -> V2Double -> LLVM.Value I.Int32 -> V8Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Double)) -insertf64x2_512_mask = Ext.intrinsic ExtX86.avx512 "mask.insertf64x2.512" - -insertf64x4_mask :: Ext.T (V8Double -> V4Double -> LLVM.Value I.Int32 -> V8Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Double)) -insertf64x4_mask = Ext.intrinsic ExtX86.avx512 "mask.insertf64x4.512" - -inserti32x4_256_mask :: Ext.T (V8Int32 -> V4Int32 -> LLVM.Value I.Int32 -> V8Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int32)) -inserti32x4_256_mask = Ext.intrinsic ExtX86.avx512 "mask.inserti32x4.256" - -inserti32x4_512_mask :: Ext.T (V16Int32 -> V4Int32 -> LLVM.Value I.Int32 -> V16Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V16Int32)) -inserti32x4_512_mask = Ext.intrinsic ExtX86.avx512 "mask.inserti32x4.512" - -inserti32x8_mask :: Ext.T (V16Int32 -> V8Int32 -> LLVM.Value I.Int32 -> V16Int32 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int32)) -inserti32x8_mask = Ext.intrinsic ExtX86.avx512 "mask.inserti32x8.512" - -inserti64x2_256_mask :: Ext.T (V4Int64 -> V2Int64 -> LLVM.Value I.Int32 -> V4Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int64)) -inserti64x2_256_mask = Ext.intrinsic ExtX86.avx512 "mask.inserti64x2.256" - -inserti64x2_512_mask :: Ext.T (V8Int64 -> V2Int64 -> LLVM.Value I.Int32 -> V8Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int64)) -inserti64x2_512_mask = Ext.intrinsic ExtX86.avx512 "mask.inserti64x2.512" - -inserti64x4_mask :: Ext.T (V8Int64 -> V4Int64 -> LLVM.Value I.Int32 -> V8Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int64)) -inserti64x4_mask = Ext.intrinsic ExtX86.avx512 "mask.inserti64x4.512" - -maskloadd :: Ext.T (LLVM.Value (Ptr ()) -> V4Int32 -> LLVM.CodeGenFunction r (V4Int32)) -maskloadd = Ext.intrinsic ExtX86.avx2 "maskload.d" - -maskloadq :: Ext.T (LLVM.Value (Ptr ()) -> V2Int64 -> LLVM.CodeGenFunction r (V2Int64)) -maskloadq = Ext.intrinsic ExtX86.avx2 "maskload.q" - -maskloadd256 :: Ext.T (LLVM.Value (Ptr ()) -> V8Int32 -> LLVM.CodeGenFunction r (V8Int32)) -maskloadd256 = Ext.intrinsic ExtX86.avx2 "maskload.d.256" - -maskloadq256 :: Ext.T (LLVM.Value (Ptr ()) -> V4Int64 -> LLVM.CodeGenFunction r (V4Int64)) -maskloadq256 = Ext.intrinsic ExtX86.avx2 "maskload.q.256" - -loaddqusi512_mask :: Ext.T (LLVM.Value (Ptr ()) -> V16Int32 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int32)) -loaddqusi512_mask = Ext.intrinsic ExtX86.avx512 "mask.loadu.d.512" - -loaddqudi512_mask :: Ext.T (LLVM.Value (Ptr ()) -> V8Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int64)) -loaddqudi512_mask = Ext.intrinsic ExtX86.avx512 "mask.loadu.q.512" - -maskstored :: Ext.T (LLVM.Value (Ptr ()) -> V4Int32 -> V4Int32 -> LLVM.CodeGenFunction r (LLVM.Value ())) -maskstored = Ext.intrinsic ExtX86.avx2 "maskstore.d" - -maskstoreq :: Ext.T (LLVM.Value (Ptr ()) -> V2Int64 -> V2Int64 -> LLVM.CodeGenFunction r (LLVM.Value ())) -maskstoreq = Ext.intrinsic ExtX86.avx2 "maskstore.q" - -maskstored256 :: Ext.T (LLVM.Value (Ptr ()) -> V8Int32 -> V8Int32 -> LLVM.CodeGenFunction r (LLVM.Value ())) -maskstored256 = Ext.intrinsic ExtX86.avx2 "maskstore.d.256" - -maskstoreq256 :: Ext.T (LLVM.Value (Ptr ()) -> V4Int64 -> V4Int64 -> LLVM.CodeGenFunction r (LLVM.Value ())) -maskstoreq256 = Ext.intrinsic ExtX86.avx2 "maskstore.q.256" - -storedqusi512_mask :: Ext.T (LLVM.Value (Ptr ()) -> V16Int32 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (LLVM.Value ())) -storedqusi512_mask = Ext.intrinsic ExtX86.avx512 "mask.storeu.d.512" - -storedqudi512_mask :: Ext.T (LLVM.Value (Ptr ()) -> V8Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (LLVM.Value ())) -storedqudi512_mask = Ext.intrinsic ExtX86.avx512 "mask.storeu.q.512" - -psllv4si :: Ext.T (V4Int32 -> V4Int32 -> LLVM.CodeGenFunction r (V4Int32)) -psllv4si = Ext.intrinsic ExtX86.avx2 "psllv.d" - -psllv8si :: Ext.T (V8Int32 -> V8Int32 -> LLVM.CodeGenFunction r (V8Int32)) -psllv8si = Ext.intrinsic ExtX86.avx2 "psllv.d.256" - -psllv2di :: Ext.T (V2Int64 -> V2Int64 -> LLVM.CodeGenFunction r (V2Int64)) -psllv2di = Ext.intrinsic ExtX86.avx2 "psllv.q" - -psllv4di :: Ext.T (V4Int64 -> V4Int64 -> LLVM.CodeGenFunction r (V4Int64)) -psllv4di = Ext.intrinsic ExtX86.avx2 "psllv.q.256" - -psrlv4si :: Ext.T (V4Int32 -> V4Int32 -> LLVM.CodeGenFunction r (V4Int32)) -psrlv4si = Ext.intrinsic ExtX86.avx2 "psrlv.d" - -psrlv8si :: Ext.T (V8Int32 -> V8Int32 -> LLVM.CodeGenFunction r (V8Int32)) -psrlv8si = Ext.intrinsic ExtX86.avx2 "psrlv.d.256" - -psrlv2di :: Ext.T (V2Int64 -> V2Int64 -> LLVM.CodeGenFunction r (V2Int64)) -psrlv2di = Ext.intrinsic ExtX86.avx2 "psrlv.q" - -psrlv4di :: Ext.T (V4Int64 -> V4Int64 -> LLVM.CodeGenFunction r (V4Int64)) -psrlv4di = Ext.intrinsic ExtX86.avx2 "psrlv.q.256" - -psrav4si :: Ext.T (V4Int32 -> V4Int32 -> LLVM.CodeGenFunction r (V4Int32)) -psrav4si = Ext.intrinsic ExtX86.avx2 "psrav.d" - -psrav8si :: Ext.T (V8Int32 -> V8Int32 -> LLVM.CodeGenFunction r (V8Int32)) -psrav8si = Ext.intrinsic ExtX86.avx2 "psrav.d.256" - -psllv16si_mask :: Ext.T (V16Int32 -> V16Int32 -> V16Int32 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int32)) -psllv16si_mask = Ext.intrinsic ExtX86.avx512 "mask.psllv.d" - -psllv8di_mask :: Ext.T (V8Int64 -> V8Int64 -> V8Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int64)) -psllv8di_mask = Ext.intrinsic ExtX86.avx512 "mask.psllv.q" - -psrav16si_mask :: Ext.T (V16Int32 -> V16Int32 -> V16Int32 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int32)) -psrav16si_mask = Ext.intrinsic ExtX86.avx512 "mask.psrav.d" - -psrav8di_mask :: Ext.T (V8Int64 -> V8Int64 -> V8Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int64)) -psrav8di_mask = Ext.intrinsic ExtX86.avx512 "mask.psrav.q" - -psrlv16si_mask :: Ext.T (V16Int32 -> V16Int32 -> V16Int32 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int32)) -psrlv16si_mask = Ext.intrinsic ExtX86.avx512 "mask.psrlv.d" - -psrlv8di_mask :: Ext.T (V8Int64 -> V8Int64 -> V8Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int64)) -psrlv8di_mask = Ext.intrinsic ExtX86.avx512 "mask.psrlv.q" - -pslldq512 :: Ext.T (V8Int64 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V8Int64)) -pslldq512 = Ext.intrinsic ExtX86.avx512 "psll.dq.512" - -psrldq512 :: Ext.T (V8Int64 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V8Int64)) -psrldq512 = Ext.intrinsic ExtX86.avx512 "psrl.dq.512" - -pslld128_mask :: Ext.T (V4Int32 -> V4Int32 -> V4Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int32)) -pslld128_mask = Ext.intrinsic ExtX86.avx512 "mask.psll.d.128" - -pslld256_mask :: Ext.T (V8Int32 -> V4Int32 -> V8Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int32)) -pslld256_mask = Ext.intrinsic ExtX86.avx512 "mask.psll.d.256" - -pslldi128_mask :: Ext.T (V4Int32 -> LLVM.Value I.Int8 -> V4Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int32)) -pslldi128_mask = Ext.intrinsic ExtX86.avx512 "mask.psll.di.128" - -pslldi256_mask :: Ext.T (V8Int32 -> LLVM.Value I.Int8 -> V8Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int32)) -pslldi256_mask = Ext.intrinsic ExtX86.avx512 "mask.psll.di.256" - -pslldi512_mask :: Ext.T (V16Int32 -> LLVM.Value I.Int8 -> V16Int32 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int32)) -pslldi512_mask = Ext.intrinsic ExtX86.avx512 "mask.psll.di.512" - -psllq128_mask :: Ext.T (V2Int64 -> V2Int64 -> V2Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Int64)) -psllq128_mask = Ext.intrinsic ExtX86.avx512 "mask.psll.q.128" - -psllq256_mask :: Ext.T (V4Int64 -> V2Int64 -> V4Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int64)) -psllq256_mask = Ext.intrinsic ExtX86.avx512 "mask.psll.q.256" - -psllqi128_mask :: Ext.T (V2Int64 -> LLVM.Value I.Int8 -> V2Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Int64)) -psllqi128_mask = Ext.intrinsic ExtX86.avx512 "mask.psll.qi.128" - -psllqi256_mask :: Ext.T (V4Int64 -> LLVM.Value I.Int8 -> V4Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int64)) -psllqi256_mask = Ext.intrinsic ExtX86.avx512 "mask.psll.qi.256" - -psllqi512_mask :: Ext.T (V8Int64 -> LLVM.Value I.Int8 -> V8Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int64)) -psllqi512_mask = Ext.intrinsic ExtX86.avx512 "mask.psll.qi.512" - -psrav16hi_mask :: Ext.T (V16Int16 -> V16Int16 -> V16Int16 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int16)) -psrav16hi_mask = Ext.intrinsic ExtX86.avx512 "mask.psrav16.hi" - -psrav32hi_mask :: Ext.T (V32Int16 -> V32Int16 -> V32Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V32Int16)) -psrav32hi_mask = Ext.intrinsic ExtX86.avx512 "mask.psrav32.hi" - -psrav4si_mask :: Ext.T (V4Int32 -> V4Int32 -> V4Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int32)) -psrav4si_mask = Ext.intrinsic ExtX86.avx512 "mask.psrav4.si" - -psrav8hi_mask :: Ext.T (V8Int16 -> V8Int16 -> V8Int16 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int16)) -psrav8hi_mask = Ext.intrinsic ExtX86.avx512 "mask.psrav8.hi" - -psrav8si_mask :: Ext.T (V8Int32 -> V8Int32 -> V8Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int32)) -psrav8si_mask = Ext.intrinsic ExtX86.avx512 "mask.psrav8.si" - -psravq128_mask :: Ext.T (V2Int64 -> V2Int64 -> V2Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Int64)) -psravq128_mask = Ext.intrinsic ExtX86.avx512 "mask.psrav.q.128" - -psravq256_mask :: Ext.T (V4Int64 -> V4Int64 -> V4Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int64)) -psravq256_mask = Ext.intrinsic ExtX86.avx512 "mask.psrav.q.256" - -psrlv16hi_mask :: Ext.T (V16Int16 -> V16Int16 -> V16Int16 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int16)) -psrlv16hi_mask = Ext.intrinsic ExtX86.avx512 "mask.psrlv16.hi" - -psrlv2di_mask :: Ext.T (V2Int64 -> V2Int64 -> V2Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Int64)) -psrlv2di_mask = Ext.intrinsic ExtX86.avx512 "mask.psrlv2.di" - -psrlv32hi_mask :: Ext.T (V32Int16 -> V32Int16 -> V32Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V32Int16)) -psrlv32hi_mask = Ext.intrinsic ExtX86.avx512 "mask.psrlv32hi" - -psrlv4di_mask :: Ext.T (V4Int64 -> V4Int64 -> V4Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int64)) -psrlv4di_mask = Ext.intrinsic ExtX86.avx512 "mask.psrlv4.di" - -psrlv4si_mask :: Ext.T (V4Int32 -> V4Int32 -> V4Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int32)) -psrlv4si_mask = Ext.intrinsic ExtX86.avx512 "mask.psrlv4.si" - -psrlv8hi_mask :: Ext.T (V8Int16 -> V8Int16 -> V8Int16 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int16)) -psrlv8hi_mask = Ext.intrinsic ExtX86.avx512 "mask.psrlv8.hi" - -psrlv8si_mask :: Ext.T (V8Int32 -> V8Int32 -> V8Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int32)) -psrlv8si_mask = Ext.intrinsic ExtX86.avx512 "mask.psrlv8.si" - -prorvd128_mask :: Ext.T (V4Int32 -> V4Int32 -> V4Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int32)) -prorvd128_mask = Ext.intrinsic ExtX86.avx512 "mask.prorv.d.128" - -prorvd256_mask :: Ext.T (V8Int32 -> V8Int32 -> V8Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int32)) -prorvd256_mask = Ext.intrinsic ExtX86.avx512 "mask.prorv.d.256" - -prorvd512_mask :: Ext.T (V16Int32 -> V16Int32 -> V16Int32 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int32)) -prorvd512_mask = Ext.intrinsic ExtX86.avx512 "mask.prorv.d.512" - -prorvq128_mask :: Ext.T (V2Int64 -> V2Int64 -> V2Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Int64)) -prorvq128_mask = Ext.intrinsic ExtX86.avx512 "mask.prorv.q.128" - -prorvq256_mask :: Ext.T (V4Int64 -> V4Int64 -> V4Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int64)) -prorvq256_mask = Ext.intrinsic ExtX86.avx512 "mask.prorv.q.256" - -prorvq512_mask :: Ext.T (V8Int64 -> V8Int64 -> V8Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int64)) -prorvq512_mask = Ext.intrinsic ExtX86.avx512 "mask.prorv.q.512" - -prold128_mask :: Ext.T (V4Int32 -> LLVM.Value I.Int8 -> V4Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int32)) -prold128_mask = Ext.intrinsic ExtX86.avx512 "mask.prol.d.128" - -prold256_mask :: Ext.T (V8Int32 -> LLVM.Value I.Int8 -> V8Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int32)) -prold256_mask = Ext.intrinsic ExtX86.avx512 "mask.prol.d.256" - -prold512_mask :: Ext.T (V16Int32 -> LLVM.Value I.Int8 -> V16Int32 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int32)) -prold512_mask = Ext.intrinsic ExtX86.avx512 "mask.prol.d.512" - -prolq128_mask :: Ext.T (V2Int64 -> LLVM.Value I.Int8 -> V2Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Int64)) -prolq128_mask = Ext.intrinsic ExtX86.avx512 "mask.prol.q.128" - -prolq256_mask :: Ext.T (V4Int64 -> LLVM.Value I.Int8 -> V4Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int64)) -prolq256_mask = Ext.intrinsic ExtX86.avx512 "mask.prol.q.256" - -prolq512_mask :: Ext.T (V8Int64 -> LLVM.Value I.Int8 -> V8Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int64)) -prolq512_mask = Ext.intrinsic ExtX86.avx512 "mask.prol.q.512" - -prolvd128_mask :: Ext.T (V4Int32 -> V4Int32 -> V4Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int32)) -prolvd128_mask = Ext.intrinsic ExtX86.avx512 "mask.prolv.d.128" - -prolvd256_mask :: Ext.T (V8Int32 -> V8Int32 -> V8Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int32)) -prolvd256_mask = Ext.intrinsic ExtX86.avx512 "mask.prolv.d.256" - -prolvd512_mask :: Ext.T (V16Int32 -> V16Int32 -> V16Int32 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int32)) -prolvd512_mask = Ext.intrinsic ExtX86.avx512 "mask.prolv.d.512" - -prolvq128_mask :: Ext.T (V2Int64 -> V2Int64 -> V2Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Int64)) -prolvq128_mask = Ext.intrinsic ExtX86.avx512 "mask.prolv.q.128" - -prolvq256_mask :: Ext.T (V4Int64 -> V4Int64 -> V4Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int64)) -prolvq256_mask = Ext.intrinsic ExtX86.avx512 "mask.prolv.q.256" - -prolvq512_mask :: Ext.T (V8Int64 -> V8Int64 -> V8Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int64)) -prolvq512_mask = Ext.intrinsic ExtX86.avx512 "mask.prolv.q.512" - -prord128_mask :: Ext.T (V4Int32 -> LLVM.Value I.Int8 -> V4Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int32)) -prord128_mask = Ext.intrinsic ExtX86.avx512 "mask.pror.d.128" - -prord256_mask :: Ext.T (V8Int32 -> LLVM.Value I.Int8 -> V8Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int32)) -prord256_mask = Ext.intrinsic ExtX86.avx512 "mask.pror.d.256" - -prord512_mask :: Ext.T (V16Int32 -> LLVM.Value I.Int8 -> V16Int32 -> LLVM.Value I.Int16 -> LLVM.CodeGenFunction r (V16Int32)) -prord512_mask = Ext.intrinsic ExtX86.avx512 "mask.pror.d.512" - -prorq128_mask :: Ext.T (V2Int64 -> LLVM.Value I.Int8 -> V2Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Int64)) -prorq128_mask = Ext.intrinsic ExtX86.avx512 "mask.pror.q.128" - -prorq256_mask :: Ext.T (V4Int64 -> LLVM.Value I.Int8 -> V4Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int64)) -prorq256_mask = Ext.intrinsic ExtX86.avx512 "mask.pror.q.256" - -prorq512_mask :: Ext.T (V8Int64 -> LLVM.Value I.Int8 -> V8Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int64)) -prorq512_mask = Ext.intrinsic ExtX86.avx512 "mask.pror.q.512" - -gatherd_pd :: Ext.T (V2Double -> LLVM.Value (Ptr ()) -> V4Int32 -> V2Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Double)) -gatherd_pd = Ext.intrinsic ExtX86.avx2 "gather.d.pd" - -gatherd_pd256 :: Ext.T (V4Double -> LLVM.Value (Ptr ()) -> V4Int32 -> V4Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Double)) -gatherd_pd256 = Ext.intrinsic ExtX86.avx2 "gather.d.pd.256" - -gatherq_pd :: Ext.T (V2Double -> LLVM.Value (Ptr ()) -> V2Int64 -> V2Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Double)) -gatherq_pd = Ext.intrinsic ExtX86.avx2 "gather.q.pd" - -gatherq_pd256 :: Ext.T (V4Double -> LLVM.Value (Ptr ()) -> V4Int64 -> V4Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Double)) -gatherq_pd256 = Ext.intrinsic ExtX86.avx2 "gather.q.pd.256" - -gatherd_ps :: Ext.T (V4Float -> LLVM.Value (Ptr ()) -> V4Int32 -> V4Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Float)) -gatherd_ps = Ext.intrinsic ExtX86.avx2 "gather.d.ps" - -gatherd_ps256 :: Ext.T (V8Float -> LLVM.Value (Ptr ()) -> V8Int32 -> V8Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Float)) -gatherd_ps256 = Ext.intrinsic ExtX86.avx2 "gather.d.ps.256" - -gatherq_ps :: Ext.T (V4Float -> LLVM.Value (Ptr ()) -> V2Int64 -> V4Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Float)) -gatherq_ps = Ext.intrinsic ExtX86.avx2 "gather.q.ps" - -gatherq_ps256 :: Ext.T (V4Float -> LLVM.Value (Ptr ()) -> V4Int64 -> V4Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Float)) -gatherq_ps256 = Ext.intrinsic ExtX86.avx2 "gather.q.ps.256" - -gatherd_q :: Ext.T (V2Int64 -> LLVM.Value (Ptr ()) -> V4Int32 -> V2Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Int64)) -gatherd_q = Ext.intrinsic ExtX86.avx2 "gather.d.q" - -gatherd_q256 :: Ext.T (V4Int64 -> LLVM.Value (Ptr ()) -> V4Int32 -> V4Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int64)) -gatherd_q256 = Ext.intrinsic ExtX86.avx2 "gather.d.q.256" - -gatherq_q :: Ext.T (V2Int64 -> LLVM.Value (Ptr ()) -> V2Int64 -> V2Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Int64)) -gatherq_q = Ext.intrinsic ExtX86.avx2 "gather.q.q" - -gatherq_q256 :: Ext.T (V4Int64 -> LLVM.Value (Ptr ()) -> V4Int64 -> V4Int64 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int64)) -gatherq_q256 = Ext.intrinsic ExtX86.avx2 "gather.q.q.256" - -gatherd_d :: Ext.T (V4Int32 -> LLVM.Value (Ptr ()) -> V4Int32 -> V4Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int32)) -gatherd_d = Ext.intrinsic ExtX86.avx2 "gather.d.d" - -gatherd_d256 :: Ext.T (V8Int32 -> LLVM.Value (Ptr ()) -> V8Int32 -> V8Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Int32)) -gatherd_d256 = Ext.intrinsic ExtX86.avx2 "gather.d.d.256" - -gatherq_d :: Ext.T (V4Int32 -> LLVM.Value (Ptr ()) -> V2Int64 -> V4Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int32)) -gatherq_d = Ext.intrinsic ExtX86.avx2 "gather.q.d" - -gatherq_d256 :: Ext.T (V4Int32 -> LLVM.Value (Ptr ()) -> V4Int64 -> V4Int32 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Int32)) -gatherq_d256 = Ext.intrinsic ExtX86.avx2 "gather.q.d.256" - -pmovmskb256 :: Ext.T (V32Int8 -> LLVM.CodeGenFunction r (LLVM.Value I.Int32)) -pmovmskb256 = Ext.intrinsic ExtX86.avx2 "pmovmskb" - -pshufb256 :: Ext.T (V32Int8 -> V32Int8 -> LLVM.CodeGenFunction r (V32Int8)) -pshufb256 = Ext.intrinsic ExtX86.avx2 "pshuf.b" - -mpsadbw256 :: Ext.T (V32Int8 -> V32Int8 -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V16Int16)) -mpsadbw256 = Ext.intrinsic ExtX86.avx2 "mpsadbw" - -movntdqa256 :: Ext.T (LLVM.Value (Ptr ()) -> LLVM.CodeGenFunction r (V4Int64)) -movntdqa256 = Ext.intrinsic ExtX86.avx2 "movntdqa" - -vfmaddss :: Ext.T (V4Float -> V4Float -> V4Float -> LLVM.CodeGenFunction r (V4Float)) -vfmaddss = Ext.intrinsic ExtX86.fma "vfmadd.ss" - -vfmaddsd :: Ext.T (V2Double -> V2Double -> V2Double -> LLVM.CodeGenFunction r (V2Double)) -vfmaddsd = Ext.intrinsic ExtX86.fma "vfmadd.sd" - -vfmaddps :: Ext.T (V4Float -> V4Float -> V4Float -> LLVM.CodeGenFunction r (V4Float)) -vfmaddps = Ext.intrinsic ExtX86.fma "vfmadd.ps" - -vfmaddpd :: Ext.T (V2Double -> V2Double -> V2Double -> LLVM.CodeGenFunction r (V2Double)) -vfmaddpd = Ext.intrinsic ExtX86.fma "vfmadd.pd" - -vfmaddps256 :: Ext.T (V8Float -> V8Float -> V8Float -> LLVM.CodeGenFunction r (V8Float)) -vfmaddps256 = Ext.intrinsic ExtX86.fma "vfmadd.ps.256" - -vfmaddpd256 :: Ext.T (V4Double -> V4Double -> V4Double -> LLVM.CodeGenFunction r (V4Double)) -vfmaddpd256 = Ext.intrinsic ExtX86.fma "vfmadd.pd.256" - -vfmsubss :: Ext.T (V4Float -> V4Float -> V4Float -> LLVM.CodeGenFunction r (V4Float)) -vfmsubss = Ext.intrinsic ExtX86.fma "vfmsub.ss" - -vfmsubsd :: Ext.T (V2Double -> V2Double -> V2Double -> LLVM.CodeGenFunction r (V2Double)) -vfmsubsd = Ext.intrinsic ExtX86.fma "vfmsub.sd" - -vfmsubps :: Ext.T (V4Float -> V4Float -> V4Float -> LLVM.CodeGenFunction r (V4Float)) -vfmsubps = Ext.intrinsic ExtX86.fma "vfmsub.ps" - -vfmsubpd :: Ext.T (V2Double -> V2Double -> V2Double -> LLVM.CodeGenFunction r (V2Double)) -vfmsubpd = Ext.intrinsic ExtX86.fma "vfmsub.pd" - -vfmsubps256 :: Ext.T (V8Float -> V8Float -> V8Float -> LLVM.CodeGenFunction r (V8Float)) -vfmsubps256 = Ext.intrinsic ExtX86.fma "vfmsub.ps.256" - -vfmsubpd256 :: Ext.T (V4Double -> V4Double -> V4Double -> LLVM.CodeGenFunction r (V4Double)) -vfmsubpd256 = Ext.intrinsic ExtX86.fma "vfmsub.pd.256" - -vfnmaddss :: Ext.T (V4Float -> V4Float -> V4Float -> LLVM.CodeGenFunction r (V4Float)) -vfnmaddss = Ext.intrinsic ExtX86.fma "vfnmadd.ss" - -vfnmaddsd :: Ext.T (V2Double -> V2Double -> V2Double -> LLVM.CodeGenFunction r (V2Double)) -vfnmaddsd = Ext.intrinsic ExtX86.fma "vfnmadd.sd" - -vfnmaddps :: Ext.T (V4Float -> V4Float -> V4Float -> LLVM.CodeGenFunction r (V4Float)) -vfnmaddps = Ext.intrinsic ExtX86.fma "vfnmadd.ps" - -vfnmaddpd :: Ext.T (V2Double -> V2Double -> V2Double -> LLVM.CodeGenFunction r (V2Double)) -vfnmaddpd = Ext.intrinsic ExtX86.fma "vfnmadd.pd" - -vfnmaddps256 :: Ext.T (V8Float -> V8Float -> V8Float -> LLVM.CodeGenFunction r (V8Float)) -vfnmaddps256 = Ext.intrinsic ExtX86.fma "vfnmadd.ps.256" - -vfnmaddpd256 :: Ext.T (V4Double -> V4Double -> V4Double -> LLVM.CodeGenFunction r (V4Double)) -vfnmaddpd256 = Ext.intrinsic ExtX86.fma "vfnmadd.pd.256" - -vfnmsubss :: Ext.T (V4Float -> V4Float -> V4Float -> LLVM.CodeGenFunction r (V4Float)) -vfnmsubss = Ext.intrinsic ExtX86.fma "vfnmsub.ss" - -vfnmsubsd :: Ext.T (V2Double -> V2Double -> V2Double -> LLVM.CodeGenFunction r (V2Double)) -vfnmsubsd = Ext.intrinsic ExtX86.fma "vfnmsub.sd" - -vfnmsubps :: Ext.T (V4Float -> V4Float -> V4Float -> LLVM.CodeGenFunction r (V4Float)) -vfnmsubps = Ext.intrinsic ExtX86.fma "vfnmsub.ps" - -vfnmsubpd :: Ext.T (V2Double -> V2Double -> V2Double -> LLVM.CodeGenFunction r (V2Double)) -vfnmsubpd = Ext.intrinsic ExtX86.fma "vfnmsub.pd" - -vfnmsubps256 :: Ext.T (V8Float -> V8Float -> V8Float -> LLVM.CodeGenFunction r (V8Float)) -vfnmsubps256 = Ext.intrinsic ExtX86.fma "vfnmsub.ps.256" - -vfnmsubpd256 :: Ext.T (V4Double -> V4Double -> V4Double -> LLVM.CodeGenFunction r (V4Double)) -vfnmsubpd256 = Ext.intrinsic ExtX86.fma "vfnmsub.pd.256" - -vfmaddsubps :: Ext.T (V4Float -> V4Float -> V4Float -> LLVM.CodeGenFunction r (V4Float)) -vfmaddsubps = Ext.intrinsic ExtX86.fma "vfmaddsub.ps" - -vfmaddsubpd :: Ext.T (V2Double -> V2Double -> V2Double -> LLVM.CodeGenFunction r (V2Double)) -vfmaddsubpd = Ext.intrinsic ExtX86.fma "vfmaddsub.pd" - -vfmaddsubps256 :: Ext.T (V8Float -> V8Float -> V8Float -> LLVM.CodeGenFunction r (V8Float)) -vfmaddsubps256 = Ext.intrinsic ExtX86.fma "vfmaddsub.ps.256" - -vfmaddsubpd256 :: Ext.T (V4Double -> V4Double -> V4Double -> LLVM.CodeGenFunction r (V4Double)) -vfmaddsubpd256 = Ext.intrinsic ExtX86.fma "vfmaddsub.pd.256" - -vfmsubaddps :: Ext.T (V4Float -> V4Float -> V4Float -> LLVM.CodeGenFunction r (V4Float)) -vfmsubaddps = Ext.intrinsic ExtX86.fma "vfmsubadd.ps" - -vfmsubaddpd :: Ext.T (V2Double -> V2Double -> V2Double -> LLVM.CodeGenFunction r (V2Double)) -vfmsubaddpd = Ext.intrinsic ExtX86.fma "vfmsubadd.pd" - -vfmsubaddps256 :: Ext.T (V8Float -> V8Float -> V8Float -> LLVM.CodeGenFunction r (V8Float)) -vfmsubaddps256 = Ext.intrinsic ExtX86.fma "vfmsubadd.ps.256" - -vfmsubaddpd256 :: Ext.T (V4Double -> V4Double -> V4Double -> LLVM.CodeGenFunction r (V4Double)) -vfmsubaddpd256 = Ext.intrinsic ExtX86.fma "vfmsubadd.pd.256" - -vfmaddpd128_mask :: Ext.T (V2Double -> V2Double -> V2Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Double)) -vfmaddpd128_mask = Ext.intrinsic ExtX86.avx512 "mask.vfmadd.pd.128" - -vfmaddpd128_mask3 :: Ext.T (V2Double -> V2Double -> V2Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Double)) -vfmaddpd128_mask3 = Ext.intrinsic ExtX86.avx512 "mask3.vfmadd.pd.128" - -vfmaddpd128_maskz :: Ext.T (V2Double -> V2Double -> V2Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Double)) -vfmaddpd128_maskz = Ext.intrinsic ExtX86.avx512 "maskz.vfmadd.pd.128" - -vfmaddpd256_mask :: Ext.T (V4Double -> V4Double -> V4Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Double)) -vfmaddpd256_mask = Ext.intrinsic ExtX86.avx512 "mask.vfmadd.pd.256" - -vfmaddpd256_mask3 :: Ext.T (V4Double -> V4Double -> V4Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Double)) -vfmaddpd256_mask3 = Ext.intrinsic ExtX86.avx512 "mask3.vfmadd.pd.256" - -vfmaddpd256_maskz :: Ext.T (V4Double -> V4Double -> V4Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Double)) -vfmaddpd256_maskz = Ext.intrinsic ExtX86.avx512 "maskz.vfmadd.pd.256" - -vfmaddpd512_mask :: Ext.T (V8Double -> V8Double -> V8Double -> LLVM.Value I.Int8 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V8Double)) -vfmaddpd512_mask = Ext.intrinsic ExtX86.avx512 "mask.vfmadd.pd.512" - -vfmaddpd512_mask3 :: Ext.T (V8Double -> V8Double -> V8Double -> LLVM.Value I.Int8 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V8Double)) -vfmaddpd512_mask3 = Ext.intrinsic ExtX86.avx512 "mask3.vfmadd.pd.512" - -vfmaddpd512_maskz :: Ext.T (V8Double -> V8Double -> V8Double -> LLVM.Value I.Int8 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V8Double)) -vfmaddpd512_maskz = Ext.intrinsic ExtX86.avx512 "maskz.vfmadd.pd.512" - -vfmaddps128_mask :: Ext.T (V4Float -> V4Float -> V4Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Float)) -vfmaddps128_mask = Ext.intrinsic ExtX86.avx512 "mask.vfmadd.ps.128" - -vfmaddps128_mask3 :: Ext.T (V4Float -> V4Float -> V4Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Float)) -vfmaddps128_mask3 = Ext.intrinsic ExtX86.avx512 "mask3.vfmadd.ps.128" - -vfmaddps128_maskz :: Ext.T (V4Float -> V4Float -> V4Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Float)) -vfmaddps128_maskz = Ext.intrinsic ExtX86.avx512 "maskz.vfmadd.ps.128" - -vfmaddps256_mask :: Ext.T (V8Float -> V8Float -> V8Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Float)) -vfmaddps256_mask = Ext.intrinsic ExtX86.avx512 "mask.vfmadd.ps.256" - -vfmaddps256_mask3 :: Ext.T (V8Float -> V8Float -> V8Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Float)) -vfmaddps256_mask3 = Ext.intrinsic ExtX86.avx512 "mask3.vfmadd.ps.256" - -vfmaddps256_maskz :: Ext.T (V8Float -> V8Float -> V8Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Float)) -vfmaddps256_maskz = Ext.intrinsic ExtX86.avx512 "maskz.vfmadd.ps.256" - -vfmaddps512_mask :: Ext.T (V16Float -> V16Float -> V16Float -> LLVM.Value I.Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V16Float)) -vfmaddps512_mask = Ext.intrinsic ExtX86.avx512 "mask.vfmadd.ps.512" - -vfmaddps512_mask3 :: Ext.T (V16Float -> V16Float -> V16Float -> LLVM.Value I.Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V16Float)) -vfmaddps512_mask3 = Ext.intrinsic ExtX86.avx512 "mask3.vfmadd.ps.512" - -vfmaddps512_maskz :: Ext.T (V16Float -> V16Float -> V16Float -> LLVM.Value I.Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V16Float)) -vfmaddps512_maskz = Ext.intrinsic ExtX86.avx512 "maskz.vfmadd.ps.512" - -vfmaddsubpd128_mask :: Ext.T (V2Double -> V2Double -> V2Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Double)) -vfmaddsubpd128_mask = Ext.intrinsic ExtX86.avx512 "mask.vfmaddsub.pd.128" - -vfmaddsubpd128_mask3 :: Ext.T (V2Double -> V2Double -> V2Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Double)) -vfmaddsubpd128_mask3 = Ext.intrinsic ExtX86.avx512 "mask3.vfmaddsub.pd.128" - -vfmaddsubpd128_maskz :: Ext.T (V2Double -> V2Double -> V2Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Double)) -vfmaddsubpd128_maskz = Ext.intrinsic ExtX86.avx512 "maskz.vfmaddsub.pd.128" - -vfmaddsubpd256_mask :: Ext.T (V4Double -> V4Double -> V4Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Double)) -vfmaddsubpd256_mask = Ext.intrinsic ExtX86.avx512 "mask.vfmaddsub.pd.256" - -vfmaddsubpd256_mask3 :: Ext.T (V4Double -> V4Double -> V4Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Double)) -vfmaddsubpd256_mask3 = Ext.intrinsic ExtX86.avx512 "mask3.vfmaddsub.pd.256" - -vfmaddsubpd256_maskz :: Ext.T (V4Double -> V4Double -> V4Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Double)) -vfmaddsubpd256_maskz = Ext.intrinsic ExtX86.avx512 "maskz.vfmaddsub.pd.256" - -vfmaddsubpd512_mask :: Ext.T (V8Double -> V8Double -> V8Double -> LLVM.Value I.Int8 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V8Double)) -vfmaddsubpd512_mask = Ext.intrinsic ExtX86.avx512 "mask.vfmaddsub.pd.512" - -vfmaddsubpd512_mask3 :: Ext.T (V8Double -> V8Double -> V8Double -> LLVM.Value I.Int8 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V8Double)) -vfmaddsubpd512_mask3 = Ext.intrinsic ExtX86.avx512 "mask3.vfmaddsub.pd.512" - -vfmaddsubpd512_maskz :: Ext.T (V8Double -> V8Double -> V8Double -> LLVM.Value I.Int8 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V8Double)) -vfmaddsubpd512_maskz = Ext.intrinsic ExtX86.avx512 "maskz.vfmaddsub.pd.512" - -vfmaddsubps128_mask :: Ext.T (V4Float -> V4Float -> V4Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Float)) -vfmaddsubps128_mask = Ext.intrinsic ExtX86.avx512 "mask.vfmaddsub.ps.128" - -vfmaddsubps128_mask3 :: Ext.T (V4Float -> V4Float -> V4Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Float)) -vfmaddsubps128_mask3 = Ext.intrinsic ExtX86.avx512 "mask3.vfmaddsub.ps.128" - -vfmaddsubps128_maskz :: Ext.T (V4Float -> V4Float -> V4Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Float)) -vfmaddsubps128_maskz = Ext.intrinsic ExtX86.avx512 "maskz.vfmaddsub.ps.128" - -vfmaddsubps256_mask :: Ext.T (V8Float -> V8Float -> V8Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Float)) -vfmaddsubps256_mask = Ext.intrinsic ExtX86.avx512 "mask.vfmaddsub.ps.256" - -vfmaddsubps256_mask3 :: Ext.T (V8Float -> V8Float -> V8Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Float)) -vfmaddsubps256_mask3 = Ext.intrinsic ExtX86.avx512 "mask3.vfmaddsub.ps.256" - -vfmaddsubps256_maskz :: Ext.T (V8Float -> V8Float -> V8Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Float)) -vfmaddsubps256_maskz = Ext.intrinsic ExtX86.avx512 "maskz.vfmaddsub.ps.256" - -vfmaddsubps512_mask :: Ext.T (V16Float -> V16Float -> V16Float -> LLVM.Value I.Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V16Float)) -vfmaddsubps512_mask = Ext.intrinsic ExtX86.avx512 "mask.vfmaddsub.ps.512" - -vfmaddsubps512_mask3 :: Ext.T (V16Float -> V16Float -> V16Float -> LLVM.Value I.Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V16Float)) -vfmaddsubps512_mask3 = Ext.intrinsic ExtX86.avx512 "mask3.vfmaddsub.ps.512" - -vfmaddsubps512_maskz :: Ext.T (V16Float -> V16Float -> V16Float -> LLVM.Value I.Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V16Float)) -vfmaddsubps512_maskz = Ext.intrinsic ExtX86.avx512 "maskz.vfmaddsub.ps.512" - -vfmsubpd128_mask3 :: Ext.T (V2Double -> V2Double -> V2Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Double)) -vfmsubpd128_mask3 = Ext.intrinsic ExtX86.avx512 "mask3.vfmsub.pd.128" - -vfmsubpd256_mask3 :: Ext.T (V4Double -> V4Double -> V4Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Double)) -vfmsubpd256_mask3 = Ext.intrinsic ExtX86.avx512 "mask3.vfmsub.pd.256" - -vfmsubpd512_mask3 :: Ext.T (V8Double -> V8Double -> V8Double -> LLVM.Value I.Int8 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V8Double)) -vfmsubpd512_mask3 = Ext.intrinsic ExtX86.avx512 "mask3.vfmsub.pd.512" - -vfmsubps128_mask3 :: Ext.T (V4Float -> V4Float -> V4Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Float)) -vfmsubps128_mask3 = Ext.intrinsic ExtX86.avx512 "mask3.vfmsub.ps.128" - -vfmsubps256_mask3 :: Ext.T (V8Float -> V8Float -> V8Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Float)) -vfmsubps256_mask3 = Ext.intrinsic ExtX86.avx512 "mask3.vfmsub.ps.256" - -vfmsubps512_mask3 :: Ext.T (V16Float -> V16Float -> V16Float -> LLVM.Value I.Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V16Float)) -vfmsubps512_mask3 = Ext.intrinsic ExtX86.avx512 "mask3.vfmsub.ps.512" - -vfmsubaddpd128_mask3 :: Ext.T (V2Double -> V2Double -> V2Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Double)) -vfmsubaddpd128_mask3 = Ext.intrinsic ExtX86.avx512 "mask3.vfmsubadd.pd.128" - -vfmsubaddpd256_mask3 :: Ext.T (V4Double -> V4Double -> V4Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Double)) -vfmsubaddpd256_mask3 = Ext.intrinsic ExtX86.avx512 "mask3.vfmsubadd.pd.256" - -vfmsubaddpd512_mask3 :: Ext.T (V8Double -> V8Double -> V8Double -> LLVM.Value I.Int8 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V8Double)) -vfmsubaddpd512_mask3 = Ext.intrinsic ExtX86.avx512 "mask3.vfmsubadd.pd.512" - -vfmsubaddps128_mask3 :: Ext.T (V4Float -> V4Float -> V4Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Float)) -vfmsubaddps128_mask3 = Ext.intrinsic ExtX86.avx512 "mask3.vfmsubadd.ps.128" - -vfmsubaddps256_mask3 :: Ext.T (V8Float -> V8Float -> V8Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Float)) -vfmsubaddps256_mask3 = Ext.intrinsic ExtX86.avx512 "mask3.vfmsubadd.ps.256" - -vfmsubaddps512_mask3 :: Ext.T (V16Float -> V16Float -> V16Float -> LLVM.Value I.Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V16Float)) -vfmsubaddps512_mask3 = Ext.intrinsic ExtX86.avx512 "mask3.vfmsubadd.ps.512" - -vfnmaddpd128_mask :: Ext.T (V2Double -> V2Double -> V2Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Double)) -vfnmaddpd128_mask = Ext.intrinsic ExtX86.avx512 "mask.vfnmadd.pd.128" - -vfnmaddpd256_mask :: Ext.T (V4Double -> V4Double -> V4Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Double)) -vfnmaddpd256_mask = Ext.intrinsic ExtX86.avx512 "mask.vfnmadd.pd.256" - -vfnmaddpd512_mask :: Ext.T (V8Double -> V8Double -> V8Double -> LLVM.Value I.Int8 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V8Double)) -vfnmaddpd512_mask = Ext.intrinsic ExtX86.avx512 "mask.vfnmadd.pd.512" - -vfnmaddps128_mask :: Ext.T (V4Float -> V4Float -> V4Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Float)) -vfnmaddps128_mask = Ext.intrinsic ExtX86.avx512 "mask.vfnmadd.ps.128" - -vfnmaddps256_mask :: Ext.T (V8Float -> V8Float -> V8Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Float)) -vfnmaddps256_mask = Ext.intrinsic ExtX86.avx512 "mask.vfnmadd.ps.256" - -vfnmaddps512_mask :: Ext.T (V16Float -> V16Float -> V16Float -> LLVM.Value I.Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V16Float)) -vfnmaddps512_mask = Ext.intrinsic ExtX86.avx512 "mask.vfnmadd.ps.512" - -vfnmsubpd128_mask :: Ext.T (V2Double -> V2Double -> V2Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Double)) -vfnmsubpd128_mask = Ext.intrinsic ExtX86.avx512 "mask.vfnmsub.pd.128" - -vfnmsubpd128_mask3 :: Ext.T (V2Double -> V2Double -> V2Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V2Double)) -vfnmsubpd128_mask3 = Ext.intrinsic ExtX86.avx512 "mask3.vfnmsub.pd.128" - -vfnmsubpd256_mask :: Ext.T (V4Double -> V4Double -> V4Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Double)) -vfnmsubpd256_mask = Ext.intrinsic ExtX86.avx512 "mask.vfnmsub.pd.256" - -vfnmsubpd256_mask3 :: Ext.T (V4Double -> V4Double -> V4Double -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Double)) -vfnmsubpd256_mask3 = Ext.intrinsic ExtX86.avx512 "mask3.vfnmsub.pd.256" - -vfnmsubpd512_mask :: Ext.T (V8Double -> V8Double -> V8Double -> LLVM.Value I.Int8 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V8Double)) -vfnmsubpd512_mask = Ext.intrinsic ExtX86.avx512 "mask.vfnmsub.pd.512" - -vfnmsubpd512_mask3 :: Ext.T (V8Double -> V8Double -> V8Double -> LLVM.Value I.Int8 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V8Double)) -vfnmsubpd512_mask3 = Ext.intrinsic ExtX86.avx512 "mask3.vfnmsub.pd.512" - -vfnmsubps128_mask :: Ext.T (V4Float -> V4Float -> V4Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Float)) -vfnmsubps128_mask = Ext.intrinsic ExtX86.avx512 "mask.vfnmsub.ps.128" - -vfnmsubps128_mask3 :: Ext.T (V4Float -> V4Float -> V4Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V4Float)) -vfnmsubps128_mask3 = Ext.intrinsic ExtX86.avx512 "mask3.vfnmsub.ps.128" - -vfnmsubps256_mask :: Ext.T (V8Float -> V8Float -> V8Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Float)) -vfnmsubps256_mask = Ext.intrinsic ExtX86.avx512 "mask.vfnmsub.ps.256" - -vfnmsubps256_mask3 :: Ext.T (V8Float -> V8Float -> V8Float -> LLVM.Value I.Int8 -> LLVM.CodeGenFunction r (V8Float)) -vfnmsubps256_mask3 = Ext.intrinsic ExtX86.avx512 "mask3.vfnmsub.ps.256" - -vfnmsubps512_mask :: Ext.T (V16Float -> V16Float -> V16Float -> LLVM.Value I.Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V16Float)) -vfnmsubps512_mask = Ext.intrinsic ExtX86.avx512 "mask.vfnmsub.ps.512" - -vfnmsubps512_mask3 :: Ext.T (V16Float -> V16Float -> V16Float -> LLVM.Value I.Int16 -> LLVM.Value I.Int32 -> LLVM.CodeGenFunction r (V16Float)) -vfnmsubps512_mask3 = Ext.intrinsic ExtX86.avx512 "mask3.vfnmsub.ps.512" - - diff --git a/src/LLVM/Extra/FastMath.hs b/src/LLVM/Extra/FastMath.hs --- a/src/LLVM/Extra/FastMath.hs +++ b/src/LLVM/Extra/FastMath.hs @@ -1,8 +1,38 @@ {-# LANGUAGE TypeFamilies #-} {-# LANGUAGE UndecidableInstances #-} {-# LANGUAGE GeneralizedNewtypeDeriving #-} -module LLVM.Extra.FastMath where +module LLVM.Extra.FastMath ( + NoNaNs(NoNaNs), + NoInfs(NoInfs), + NoSignedZeros(NoSignedZeros), + AllowReciprocal(AllowReciprocal), + Fast(Fast), + Flags(setFlags), + Number(Number, deconsNumber), + getNumber, + mvNumber, + mvDenumber, + + MultiValue(setMultiValueFlags), + attachMultiValueFlags, + liftNumberM, + liftNumberM2, + mvecNumber, + mvecDenumber, + + MultiVector(setMultiVectorFlags), + attachMultiVectorFlags, + liftMultiVectorM, + liftMultiVectorM2, + + Tuple(setTupleFlags), + Context(Context), + attachTupleFlags, + liftContext, + liftContext2, + ) where + import qualified LLVM.Extra.Multi.Vector as MultiVector import qualified LLVM.Extra.Multi.Value.Private as MV import qualified LLVM.Extra.Arithmetic as A @@ -264,14 +294,12 @@ fmap mvecNumber . MultiVector.insert k (mvDenumber x) . mvecDenumber instance - (Flags flags, - MultiVector a, MV.IntegerConstant a, MultiVector.IntegerConstant a) => + (Flags flags, MultiVector a, MultiVector.IntegerConstant a) => MultiVector.IntegerConstant (Number flags a) where fromInteger' = mvecNumber . MultiVector.fromInteger' instance - (Flags flags, - MultiVector a, MV.RationalConstant a, MultiVector.RationalConstant a) => + (Flags flags, MultiVector a, MultiVector.RationalConstant a) => MultiVector.RationalConstant (Number flags a) where fromRational' = mvecNumber . MultiVector.fromRational' diff --git a/src/LLVM/Extra/Iterator.hs b/src/LLVM/Extra/Iterator.hs --- a/src/LLVM/Extra/Iterator.hs +++ b/src/LLVM/Extra/Iterator.hs @@ -1,7 +1,34 @@ {-# LANGUAGE ExistentialQuantification #-} {-# LANGUAGE Rank2Types #-} {-# LANGUAGE TypeFamilies #-} -module LLVM.Extra.Iterator where +module LLVM.Extra.Iterator ( + T, + -- * consumers + mapM_, + mapState_, + mapStateM_, + mapWhileState_, + -- * producers + empty, + singleton, + cons, + iterate, + countDown, + arrayPtrs, + -- * modifiers + mapM, + mapMaybe, + catMaybes, + takeWhileJust, + takeWhile, + cartesian, + take, + -- * application examples + fixedLengthLoop, + arrayLoop, + arrayLoopWithExit, + arrayLoop2, + ) where import qualified LLVM.Extra.MaybeContinuation as MaybeCont import qualified LLVM.Extra.Maybe as Maybe @@ -13,7 +40,7 @@ import LLVM.Util.Loop (Phi, ) import LLVM.Core (CodeGenFunction, Value, value, valueOf, - CmpRet, CmpResult, IsInteger, IsType, IsConst, ) + CmpRet, IsInteger, IsType, IsConst, IsPrimitive) import Foreign.Ptr (Ptr, ) @@ -25,7 +52,7 @@ import Data.Tuple.HT (mapFst, mapSnd, ) -import Prelude2010 hiding (iterate, takeWhile, take, mapM) +import Prelude2010 hiding (iterate, takeWhile, take, mapM, mapM_) import Prelude () @@ -164,6 +191,7 @@ (Phi a, Class.Undefined a) => (a -> CodeGenFunction r a) -> a -> T r a iterate f a = Cons a (\a0 -> MaybeCont.lift $ fmap ((,) a0) $ f a0) + cartesianAux :: (Phi a, Phi b, Class.Undefined a, Class.Undefined b) => T r a -> T r b -> T r (Maybe.T (a,b)) @@ -180,22 +208,19 @@ (\(b1,sb1) -> return (Maybe.just (a1,b1), (Maybe.just a1, sa1, sb1)))) - --- * helper functions - cartesian :: (Phi a, Phi b, Class.Undefined a, Class.Undefined b) => T r a -> T r b -> T r (a,b) cartesian as bs = catMaybes $ cartesianAux as bs countDown :: - (Num i, IsConst i, IsInteger i, CmpRet i, CmpResult i ~ Bool) => + (Num i, IsConst i, IsInteger i, CmpRet i, IsPrimitive i) => Value i -> T r (Value i) countDown len = takeWhile (A.cmp LLVM.CmpLT (value LLVM.zero)) $ iterate A.dec len take :: - (Num i, IsConst i, IsInteger i, CmpRet i, CmpResult i ~ Bool) => + (Num i, IsConst i, IsInteger i, CmpRet i, IsPrimitive i) => Value i -> T r a -> T r a take len xs = liftA2 const xs (countDown len) @@ -206,8 +231,7 @@ -- * examples fixedLengthLoop :: - (Phi s, - Num i, IsConst i, IsInteger i, CmpRet i, CmpResult i ~ Bool) => + (Phi s, Num i, IsConst i, IsInteger i, CmpRet i, IsPrimitive i) => Value i -> s -> (s -> CodeGenFunction r s) -> CodeGenFunction r s @@ -215,8 +239,7 @@ mapState_ (const loopBody) (countDown len) start arrayLoop :: - (Phi a, IsType b, - Num i, IsConst i, IsInteger i, CmpRet i, CmpResult i ~ Bool) => + (Phi a, IsType b, Num i, IsConst i, IsInteger i, CmpRet i, IsPrimitive i) => Value i -> Value (Ptr b) -> a -> (Value (Ptr b) -> a -> CodeGenFunction r a) -> CodeGenFunction r a @@ -224,8 +247,7 @@ mapState_ loopBody (take len $ arrayPtrs ptr) start arrayLoopWithExit :: - (Phi s, IsType a, - Num i, IsConst i, IsInteger i, CmpRet i, CmpResult i ~ Bool) => + (Phi s, IsType a, Num i, IsConst i, IsInteger i, CmpRet i, IsPrimitive i) => Value i -> Value (Ptr a) -> s -> (Value (Ptr a) -> s -> CodeGenFunction r (Value Bool, s)) -> CodeGenFunction r (Value i, s) @@ -239,8 +261,7 @@ return (pos, end) arrayLoop2 :: - (Phi s, IsType a, IsType b, - Num i, IsConst i, IsInteger i, CmpRet i, CmpResult i ~ Bool) => + (Phi s, IsType a, IsType b, Num i, IsConst i, IsInteger i, CmpRet i, IsPrimitive i) => Value i -> Value (Ptr a) -> Value (Ptr b) -> s -> (Value (Ptr a) -> Value (Ptr b) -> s -> CodeGenFunction r s) -> CodeGenFunction r s diff --git a/src/LLVM/Extra/MaybeContinuation.hs b/src/LLVM/Extra/MaybeContinuation.hs --- a/src/LLVM/Extra/MaybeContinuation.hs +++ b/src/LLVM/Extra/MaybeContinuation.hs @@ -12,10 +12,8 @@ import qualified LLVM.Core as LLVM import LLVM.Core - (Value, value, valueOf, - CodeGenFunction, - IsConst, IsType, IsFirstClass, IsInteger, - CmpRet, CmpResult, ) + (CodeGenFunction, Value, value, valueOf, + IsConst, IsType, IsPrimitive, IsInteger, CmpRet) import LLVM.Util.Loop (Phi, ) -- (phis, addPhis, ) import qualified Control.Monad as M @@ -150,7 +148,7 @@ fixedLengthLoop :: (Phi s, Undefined s, - Num i, IsConst i, IsInteger i, IsFirstClass i, CmpRet i, CmpResult i ~ Bool) => + Num i, IsConst i, IsInteger i, CmpRet i, IsPrimitive i) => Value i -> s -> (s -> T r (Maybe.T s) s) -> CodeGenFunction r (Value i, Maybe.T s) @@ -175,7 +173,7 @@ -} arrayLoop :: (Phi s, Undefined s, IsType a, - Num i, IsConst i, IsInteger i, IsFirstClass i, CmpRet i, CmpResult i ~ Bool) => + Num i, IsConst i, IsInteger i, CmpRet i, IsPrimitive i) => Value i -> Value (Ptr a) -> s -> (Value (Ptr a) -> s -> T r (Maybe.T (Value (Ptr a), s)) s) -> @@ -190,7 +188,7 @@ arrayLoop2 :: (Phi s, Undefined s, IsType a, IsType b, - Num i, IsConst i, IsInteger i, IsFirstClass i, CmpRet i, CmpResult i ~ Bool) => + Num i, IsConst i, IsInteger i, CmpRet i, IsPrimitive i) => Value i -> Value (Ptr a) -> Value (Ptr b) -> s -> (Value (Ptr a) -> Value (Ptr b) -> s -> diff --git a/src/LLVM/Extra/MaybePrivate.hs b/src/LLVM/Extra/MaybePrivate.hs --- a/src/LLVM/Extra/MaybePrivate.hs +++ b/src/LLVM/Extra/MaybePrivate.hs @@ -90,8 +90,7 @@ lift2 f (Cons b0 a0) (Cons b1 a1) = Monad.liftM (flip Cons (f a0 a1)) (LLVM.and b0 b1) -sequence :: - T (CodeGenFunction r a) -> CodeGenFunction r (T a) +sequence :: T (CodeGenFunction r a) -> CodeGenFunction r (T a) sequence (Cons b0 a0) = Monad.liftM (Cons b0) a0 diff --git a/src/LLVM/Extra/Multi/Iterator.hs b/src/LLVM/Extra/Multi/Iterator.hs --- a/src/LLVM/Extra/Multi/Iterator.hs +++ b/src/LLVM/Extra/Multi/Iterator.hs @@ -49,14 +49,14 @@ take len xs = liftA2 const xs (countDown len) -class Enum a where +class (MultiValue.C a) => Enum a where succ, pred :: MultiValue.T a -> LLVM.CodeGenFunction r (MultiValue.T a) enumFrom :: MultiValue.T a -> Iter.T r (MultiValue.T a) enumFromTo :: MultiValue.T a -> MultiValue.T a -> Iter.T r (MultiValue.T a) instance (LLVM.IsInteger w, SoV.IntegerConstant w, Num w, - LLVM.CmpRet w, LLVM.CmpResult w ~ Bool, P.Enum e) => + LLVM.CmpRet w, LLVM.IsPrimitive w, P.Enum e) => Enum (Enum.T w e) where succ = MultiValue.succ pred = MultiValue.pred @@ -71,7 +71,7 @@ succMax :: (LLVM.IsInteger w, SoV.IntegerConstant w, Num w, - LLVM.CmpRet w, LLVM.CmpResult w ~ Bool, P.Enum e) => + LLVM.CmpRet w, LLVM.IsPrimitive w, P.Enum e) => MultiValue.T (Enum.T w e) -> MultiValue.T (Enum.T w e) -> LLVM.CodeGenFunction r (Maybe.T (MultiValue.T (Enum.T w e))) @@ -87,7 +87,7 @@ -} _enumFromToSimple :: (LLVM.IsInteger w, SoV.IntegerConstant w, Num w, - LLVM.CmpRet w, LLVM.CmpResult w ~ Bool, P.Enum e) => + LLVM.CmpRet w, LLVM.IsPrimitive w, P.Enum e) => MultiValue.T (Enum.T w e) -> MultiValue.T (Enum.T w e) -> Iter.T r (MultiValue.T (Enum.T w e)) diff --git a/src/LLVM/Extra/Multi/Value/Private.hs b/src/LLVM/Extra/Multi/Value/Private.hs --- a/src/LLVM/Extra/Multi/Value/Private.hs +++ b/src/LLVM/Extra/Multi/Value/Private.hs @@ -1,4 +1,5 @@ {-# LANGUAGE TypeFamilies #-} +{-# LANGUAGE FlexibleInstances #-} {-# LANGUAGE FlexibleContexts #-} {-# LANGUAGE MultiParamTypeClasses #-} module LLVM.Extra.Multi.Value.Private where @@ -14,7 +15,6 @@ import LLVM.Core (WordN, IntN, ) import qualified Type.Data.Num.Decimal.Number as Dec -import Type.Data.Num.Decimal (D1) import Foreign.StablePtr (StablePtr, ) import Foreign.Ptr (Ptr, FunPtr, ) @@ -284,7 +284,7 @@ -- cannot be an instance of 'Comparison' because there is no 'Real' instance cmpEnum :: - (LLVM.CmpRet w, LLVM.CmpResult w ~ Bool) => + (LLVM.CmpRet w, LLVM.IsPrimitive w) => LLVM.CmpPredicate -> T (Enum.T w a) -> T (Enum.T w a) -> LLVM.CodeGenFunction r (T Bool) cmpEnum = liftM2 . LLVM.cmp @@ -1009,8 +1009,8 @@ class (Repr LLVM.Value i ~ LLVM.Value ir, - LLVM.IsInteger ir, SoV.IntegerConstant ir, LLVM.CmpRet ir, - LLVM.NumberOfElements ir ~ D1, LLVM.CmpResult ir ~ Bool) => + LLVM.IsInteger ir, SoV.IntegerConstant ir, + LLVM.CmpRet ir, LLVM.IsPrimitive ir) => NativeInteger i ir where instance NativeInteger Word8 Word8 where @@ -1023,11 +1023,13 @@ instance NativeInteger Int32 Int32 where instance NativeInteger Int64 Int64 where +instance NativeInteger a a => NativeInteger (Tagged tag a) a where + class (Repr LLVM.Value a ~ LLVM.Value ar, - LLVM.IsFloating ar, SoV.RationalConstant ar, LLVM.CmpRet ar, - LLVM.NumberOfElements ar ~ D1, LLVM.CmpResult ar ~ Bool) => + LLVM.IsFloating ar, SoV.RationalConstant ar, + LLVM.CmpRet ar, LLVM.IsPrimitive ar) => NativeFloating a ar where instance NativeFloating Float Float where diff --git a/src/LLVM/Extra/Multi/Vector.hs b/src/LLVM/Extra/Multi/Vector.hs --- a/src/LLVM/Extra/Multi/Vector.hs +++ b/src/LLVM/Extra/Multi/Vector.hs @@ -424,10 +424,12 @@ (insert k a2 v2) -class (C a) => IntegerConstant a where +class (MultiValue.IntegerConstant a, C a) => IntegerConstant a where fromInteger' :: (TypeNum.Positive n) => Integer -> T n a -class (IntegerConstant a) => RationalConstant a where +class + (MultiValue.RationalConstant a, IntegerConstant a) => + RationalConstant a where fromRational' :: (TypeNum.Positive n) => Rational -> T n a instance IntegerConstant Float where fromInteger' = fromIntegerPrimitive @@ -922,7 +924,7 @@ -class (C a) => Select a where +class (MultiValue.Select a, C a) => Select a where select :: (TypeNum.Positive n) => T n Bool -> T n a -> T n a -> @@ -1035,7 +1037,7 @@ -class BitShift a where +class (MultiValue.BitShift a, C a) => BitShift a where shl :: (TypeNum.Positive n) => T n a -> T n a -> LLVM.CodeGenFunction r (T n a) shr :: (TypeNum.Positive n) => T n a -> T n a -> LLVM.CodeGenFunction r (T n a) diff --git a/src/LLVM/Extra/ScalarOrVector.hs b/src/LLVM/Extra/ScalarOrVector.hs --- a/src/LLVM/Extra/ScalarOrVector.hs +++ b/src/LLVM/Extra/ScalarOrVector.hs @@ -24,39 +24,36 @@ Replicate (replicate, replicateConst), replicateOf, Real (min, max, abs, signum), + Saturated(addSat, subSat), PseudoModule (scale, scaleConst), IntegerConstant(constFromInteger), RationalConstant(constFromRational), TranscendentalConstant(constPi), ) where -import LLVM.Extra.Vector (Element, Size, ) - +import qualified LLVM.Extra.ScalarOrVectorPrivate as Priv import qualified LLVM.Extra.Vector as Vector -import qualified LLVM.Extra.Extension.X86 as X86 -import qualified LLVM.Extra.Extension as Ext - -import qualified LLVM.Extra.Class as Class import qualified LLVM.Extra.ArithmeticPrivate as A - -import qualified Type.Data.Num.Decimal as TypeNum -import Type.Data.Num.Decimal (D1, ) +import LLVM.Extra.ScalarOrVectorPrivate + (Scalar, Replicate(replicate, replicateConst)) +import qualified LLVM.Util.Intrinsic as Intrinsic +import qualified LLVM.Util.Proxy as LP import qualified LLVM.Core as LLVM import LLVM.Core - (Value, ConstValue, valueOf, constOf, - CmpRet, CmpResult, NumberOfElements, + (Value, ConstValue, constOf, + CmpRet, CmpResult, ShapeOf, Vector, WordN(WordN), IntN(IntN), FP128, IsConst, IsInteger, IsFloating, CodeGenFunction, ) -import Control.Monad.HT ((<=<), ) +import qualified Type.Data.Num.Decimal as TypeNum -import qualified Data.NonEmpty as NonEmpty import Data.Word (Word8, Word16, Word32, Word64, ) import Data.Int (Int8, Int16, Int32, Int64, ) +import Data.Maybe (fromMaybe) -import Prelude hiding (Real, replicate, min, max, abs, truncate, floor, round, ) +import Prelude hiding (Real, replicate, min, max, abs, truncate) @@ -65,42 +62,12 @@ fraction :: Value a -> CodeGenFunction r (Value a) instance Fraction Float where - truncate = - mapAuto - (LLVM.inttofp . flip asTypeOf (undefined :: Value Int32) <=< LLVM.fptoint) - (Ext.with X86.roundss $ \round x -> round x (valueOf 3)) - fraction = - (\x -> - fractionGen x - `Ext.run` - (Ext.with X86.cmpss $ \cmp -> - fractionLogical (\modus -> curry (runScalar (uncurry (cmp modus)))) x)) - `mapAuto` - (Ext.with X86.roundss $ \round x -> - A.sub x =<< round x (valueOf 1)) + truncate = Intrinsic.truncate + fraction = A.fraction instance Fraction Double where - truncate = - mapAuto - -- X86 only converts Double to Int32, it cannot target Int64 - (LLVM.inttofp . flip asTypeOf (undefined :: Value Int32) <=< LLVM.fptoint) - (Ext.with X86.roundsd $ \round x -> round x (valueOf 3)) - fraction = - (\x -> - fractionGen x - `Ext.run` - (Ext.with X86.cmpsd $ \cmp -> - fractionLogical (\modus -> curry (runScalar (uncurry (cmp modus)))) x)) -{- -For Doubles it would be more efficient to convert the lower 32 bit -instead of the lower 64 bit, -since x86 supports only conversion from 32 bit natively. - (Ext.with X86.cmpsd $ \cmp -> fractionLogical - (\x y -> cmp x y >>= LLVM.bitcast ) --} - `mapAuto` - (Ext.with X86.roundsd $ \round x -> - A.sub x =<< round x (valueOf 1)) + truncate = Intrinsic.truncate + fraction = A.fraction instance (TypeNum.Positive n, Vector.Real a, IsFloating a, IsConst a) => Fraction (Vector n a) where @@ -118,21 +85,21 @@ signedFraction x = A.sub x =<< truncate x -fractionGen :: +_fractionGen :: (IntegerConstant v, Fraction v, CmpRet v) => Value v -> CodeGenFunction r (Value v) -fractionGen x = +_fractionGen x = do xf <- signedFraction x b <- A.fcmp LLVM.FPOGE xf zero LLVM.select b xf =<< A.add xf (LLVM.value $ constFromInteger 1) -fractionLogical :: - (Fraction a, LLVM.IsScalarOrVector a, NumberOfElements a ~ D1, - IsInteger b, LLVM.IsScalarOrVector b, NumberOfElements b ~ D1) => +_fractionLogical :: + (Fraction a, LLVM.IsPrimitive a, + IsInteger b, LLVM.IsPrimitive b) => (LLVM.FPPredicate -> Value a -> Value a -> CodeGenFunction r (Value b)) -> Value a -> CodeGenFunction r (Value a) -fractionLogical cmp x = +_fractionLogical cmp x = do xf <- signedFraction x b <- cmp LLVM.FPOLT xf zero A.sub xf =<< LLVM.inttofp b @@ -159,8 +126,7 @@ truncateToInt :: - (IsFloating a, IsInteger i, - NumberOfElements a ~ NumberOfElements i) => + (IsFloating a, IsInteger i, ShapeOf a ~ ShapeOf i) => Value a -> CodeGenFunction r (Value i) truncateToInt = LLVM.fptoint @@ -174,7 +140,7 @@ (IsFloating a, RationalConstant a, CmpRet a, IsInteger i, IntegerConstant i, CmpRet i, CmpResult a ~ CmpResult i, - NumberOfElements a ~ NumberOfElements i) => + ShapeOf a ~ ShapeOf i) => Value a -> CodeGenFunction r (Value i) roundToIntFast x = do pos <- A.cmp LLVM.CmpGT x zero @@ -185,7 +151,7 @@ (IsFloating a, CmpRet a, IsInteger i, IntegerConstant i, CmpRet i, CmpResult a ~ CmpResult i, - NumberOfElements a ~ NumberOfElements i) => + ShapeOf a ~ ShapeOf i) => Value a -> CodeGenFunction r (Value i) floorToInt x = do i <- truncateToInt x @@ -196,7 +162,7 @@ (IsFloating a, CmpRet a, IsInteger i, IntegerConstant i, CmpRet i, CmpResult a ~ CmpResult i, - NumberOfElements a ~ NumberOfElements i) => + ShapeOf a ~ ShapeOf i) => Value a -> CodeGenFunction r (Value i, Value a) splitFractionToInt x = do i <- floorToInt x @@ -207,7 +173,7 @@ (IsFloating a, CmpRet a, IsInteger i, IntegerConstant i, CmpRet i, CmpResult a ~ CmpResult i, - NumberOfElements a ~ NumberOfElements i) => + ShapeOf a ~ ShapeOf i) => Value a -> CodeGenFunction r (Value i) ceilingToInt x = do i <- truncateToInt x @@ -225,66 +191,7 @@ ratio = LLVM.value . constFromRational -type family Scalar vector :: * -type instance Scalar Float = Float -type instance Scalar Double = Double -type instance Scalar FP128 = FP128 -type instance Scalar Bool = Bool -type instance Scalar Int8 = Int8 -type instance Scalar Int16 = Int16 -type instance Scalar Int32 = Int32 -type instance Scalar Int64 = Int64 -type instance Scalar Word8 = Word8 -type instance Scalar Word16 = Word16 -type instance Scalar Word32 = Word32 -type instance Scalar Word64 = Word64 -type instance Scalar (Vector n a) = a - - - -class Replicate vector where - -- | an alternative is using the 'Vector.Constant' vector type - replicate :: Value (Scalar vector) -> CodeGenFunction r (Value vector) - replicateConst :: ConstValue (Scalar vector) -> ConstValue vector - -instance Replicate Float where replicate = return; replicateConst = id; -instance Replicate Double where replicate = return; replicateConst = id; -instance Replicate FP128 where replicate = return; replicateConst = id; -instance Replicate Bool where replicate = return; replicateConst = id; -instance Replicate Int8 where replicate = return; replicateConst = id; -instance Replicate Int16 where replicate = return; replicateConst = id; -instance Replicate Int32 where replicate = return; replicateConst = id; -instance Replicate Int64 where replicate = return; replicateConst = id; -instance Replicate Word8 where replicate = return; replicateConst = id; -instance Replicate Word16 where replicate = return; replicateConst = id; -instance Replicate Word32 where replicate = return; replicateConst = id; -instance Replicate Word64 where replicate = return; replicateConst = id; -instance (TypeNum.Positive n, LLVM.IsPrimitive a) => Replicate (Vector n a) where -{- crashes LLVM-2.5, seems to be fixed in LLVM-2.6 -} - replicate x = do - v <- singleton x - LLVM.shufflevector v (LLVM.value LLVM.undef) LLVM.zero -{- crashes LLVM-2.5 - replicate x = do - v <- LLVM.insertelement (LLVM.value LLVM.undef) x (valueOf 1) - LLVM.shufflevector v (LLVM.value LLVM.undef) (constVector $ repeat $ LLVM.constOf 1) --} -{- the (repeat zero) is also converted to 'zeroinitializer' and crashes LLVM compiler - - (constVector $ repeat LLVM.zero) --} -{- - replicate = Vector.replicate --} - replicateConst x = LLVM.constCyclicVector $ NonEmpty.Cons x [] - -singleton :: - (LLVM.IsPrimitive a) => - Value a -> CodeGenFunction r (Value (Vector D1 a)) -singleton x = - LLVM.insertelement (LLVM.value LLVM.undef) x (valueOf 0) - replicateOf :: (IsConst (Scalar v), Replicate v) => Scalar v -> Value v @@ -300,65 +207,27 @@ instance Real Float where - min = zipAutoWith A.min X86.minss - max = zipAutoWith A.max X86.maxss - abs = mapAuto A.abs X86.absss - -- abs x = max x =<< LLVM.neg x - -- abs x = A.abs + min = Intrinsic.min + max = Intrinsic.max + abs = Intrinsic.abs signum = A.signum instance Real Double where - min = zipAutoWith A.min X86.minsd - max = zipAutoWith A.max X86.maxsd - abs = mapAuto A.abs X86.abssd + min = Intrinsic.min + max = Intrinsic.max + abs = Intrinsic.abs signum = A.signum instance Real FP128 where - min = A.min - max = A.max - abs = A.abs + min = Intrinsic.min + max = Intrinsic.max + abs = Intrinsic.abs signum x = do minusOne <- LLVM.inttofp $ LLVM.valueOf (-1 :: Int8) one <- LLVM.inttofp $ LLVM.valueOf ( 1 :: Int8) A.signumGen minusOne one x -infixl 1 `mapAuto` - -{- | -There are functions that are intended for processing scalars -but have formally vector input and output. -This function breaks vector function down to a scalar function -by accessing the lowest vector element. --} -runScalar :: - (Vector.C v, Vector.C w, Size v ~ Size w) => - (v -> CodeGenFunction r w) -> - (Element v -> CodeGenFunction r (Element w)) -runScalar op a = - Vector.extract (valueOf 0) - =<< op - =<< Vector.insert (valueOf 0) a Class.undefTuple - -mapAuto :: - (Vector.C v, Vector.C w, Size v ~ Size w) => - (Element v -> CodeGenFunction r (Element w)) -> - Ext.T (v -> CodeGenFunction r w) -> - (Element v -> CodeGenFunction r (Element w)) -mapAuto f g a = - Ext.run (f a) $ - Ext.with g $ \op -> runScalar op a - -zipAutoWith :: - (Vector.C u, Vector.C v, Vector.C w, - Size u ~ Size v, Size v ~ Size w) => - (Element u -> Element v -> CodeGenFunction r (Element w)) -> - Ext.T (u -> v -> CodeGenFunction r w) -> - (Element u -> Element v -> CodeGenFunction r (Element w)) -zipAutoWith f g = - curry $ mapAuto (uncurry f) (fmap uncurry g) - - instance Real Int8 where min = A.min; max = A.max; signum = A.signum; abs = A.abs; instance Real Int16 where min = A.min; max = A.max; signum = A.signum; abs = A.abs; instance Real Int32 where min = A.min; max = A.max; signum = A.signum; abs = A.abs; @@ -380,6 +249,50 @@ max = Vector.max abs = Vector.abs signum = Vector.signum + + +class (IsInteger a) => Saturated a where + addSat, subSat :: Value a -> Value a -> CodeGenFunction r (Value a) + +instance Saturated Int8 where + addSat = addSatProxy LP.Proxy; subSat = subSatProxy LP.Proxy; +instance Saturated Int16 where + addSat = addSatProxy LP.Proxy; subSat = subSatProxy LP.Proxy; +instance Saturated Int32 where + addSat = addSatProxy LP.Proxy; subSat = subSatProxy LP.Proxy; +instance Saturated Int64 where + addSat = addSatProxy LP.Proxy; subSat = subSatProxy LP.Proxy; +instance Saturated Word8 where + addSat = addSatProxy LP.Proxy; subSat = subSatProxy LP.Proxy; +instance Saturated Word16 where + addSat = addSatProxy LP.Proxy; subSat = subSatProxy LP.Proxy; +instance Saturated Word32 where + addSat = addSatProxy LP.Proxy; subSat = subSatProxy LP.Proxy; +instance Saturated Word64 where + addSat = addSatProxy LP.Proxy; subSat = subSatProxy LP.Proxy; +instance (TypeNum.Positive d) => Saturated (IntN d) where + addSat = addSatProxy LP.Proxy; subSat = subSatProxy LP.Proxy; +instance (TypeNum.Positive d) => Saturated (WordN d) where + addSat = addSatProxy LP.Proxy; subSat = subSatProxy LP.Proxy; +instance + (TypeNum.Positive n, LLVM.IsPrimitive a, + Saturated a, Bounded a, CmpRet a, IsConst a) => + Saturated (Vector n a) where + addSat = addSatProxy LP.Proxy; subSat = subSatProxy LP.Proxy; + +addSatProxy, subSatProxy :: + (IsInteger v, CmpRet v, Replicate v, ShapeOf v ~ shape, + LLVM.ShapedType shape Bool ~ bv, ShapeOf bv ~ shape, CmpRet bv, + Scalar v ~ a, IsConst a, Bounded a) => + LP.Proxy v -> Value v -> Value v -> CodeGenFunction r (Value v) +addSatProxy proxy = + if LLVM.isSigned proxy + then fromMaybe Priv.saddSat Intrinsic.maybeSAddSat + else fromMaybe Priv.uaddSat Intrinsic.maybeUAddSat +subSatProxy proxy = + if LLVM.isSigned proxy + then fromMaybe Priv.ssubSat Intrinsic.maybeSSubSat + else fromMaybe Priv.usubSat Intrinsic.maybeUSubSat diff --git a/src/LLVM/Extra/Vector.hs b/src/LLVM/Extra/Vector.hs --- a/src/LLVM/Extra/Vector.hs +++ b/src/LLVM/Extra/Vector.hs @@ -29,9 +29,9 @@ insertChunk, modify, map, mapChunks, zipChunksWith, - chop, concat, select, + chop, concat, signedFraction, - cumulate1, umul32to64, + cumulate1, Arithmetic (sum, sumToPair, sumInterleavedToPair, cumulate, dotProduct, mul), @@ -40,13 +40,9 @@ truncate, floor, fraction), ) where -import qualified LLVM.Extra.Extension.X86Auto as X86A -import qualified LLVM.Extra.ExtensionCheck.X86 as X86C -import qualified LLVM.Extra.Extension.X86 as X86 -import qualified LLVM.Extra.Extension as Ext - import qualified LLVM.Extra.Class as Class import qualified LLVM.Extra.ArithmeticPrivate as A +import qualified LLVM.Util.Intrinsic as Intrinsic import qualified LLVM.Core as LLVM import LLVM.Util.Loop (Phi(phis, addPhis), ) @@ -58,7 +54,7 @@ CodeGenFunction, ) import qualified Type.Data.Num.Decimal as TypeNum -import Type.Data.Num.Decimal (D4, (:+:), ) +import Type.Data.Num.Decimal ((:+:)) import qualified Control.Applicative as App import qualified Control.Monad.HT as M @@ -70,11 +66,9 @@ import qualified Data.Foldable as Fold import qualified Data.NonEmpty.Class as NonEmptyC import qualified Data.NonEmpty as NonEmpty -import qualified Data.Empty as Empty import qualified Data.List.HT as ListHT import qualified Data.List as List import Data.NonEmpty ((!:), ) -import Data.Tuple.HT (uncurry3, ) -- import qualified Data.Bits as Bit import Data.Int (Int8, Int16, Int32, Int64, ) @@ -582,7 +576,7 @@ else extractAll =<< g =<< assemble c assemble $ List.concat ds ++ d -zipChunks2With :: +_zipChunks2With :: (C ca, C cb, C cc, Size ca ~ Size cb, Size cb ~ Size cc, C la, C lb, C lc, Size la ~ Size lb, Size lb ~ Size lc, C va, C vb, C vc, Size va ~ Size vb, Size vb ~ Size vc, @@ -592,37 +586,11 @@ (ca -> cb -> CodeGenFunction r cc) -> (la -> lb -> CodeGenFunction r lc) -> (va -> vb -> CodeGenFunction r vc) -zipChunks2With f g a b = +_zipChunks2With f g a b = mapChunks2 (uncurry f) (uncurry g) (a,b) -infixl 1 `withRound` -withRound :: - (IsPrimitive a, IsPrimitive b, - TypeNum.Positive k, TypeNum.Positive m, TypeNum.Positive n) => - CodeGenFunction r x -> - Ext.T (Value (Vector m a) -> Value Word32 -> CodeGenFunction r (Value (Vector m b))) -> - Ext.T (Value (Vector k a) -> Value Word32 -> CodeGenFunction r (Value (Vector k b))) -> - (Value (Vector n b) -> CodeGenFunction r x) -> - Word32 -> - Value (Vector n a) -> CodeGenFunction r x -withRound generic roundSmallExt _roundLargeExt post mode x = - generic - `Ext.run` - (Ext.with roundSmallExt $ \round -> - post =<< mapChunks (flip round (valueOf mode)) x) -{- crashes LLVM-3.1 in JIT mode -Stack dump: -0. Running pass 'X86 DAG->DAG Instruction Selection' on function '@_fun1' -segmentation fault - - `Ext.run` - (Ext.with2 roundSmallExt roundLargeExt $ \round roundLarge -> - post =<< mapChunks2 (flip round (valueOf mode)) (flip roundLarge (valueOf mode)) x) --} - - {- | Ideally on ix86 with SSE41 this would be translated to 'dpps'. -} @@ -692,7 +660,7 @@ getLowestPair :: - (TypeNum.Positive n) => + (TypeNum.Positive n, IsPrimitive a) => Value (Vector n a) -> CodeGenFunction r (Value a, Value a) getLowestPair x = @@ -819,35 +787,7 @@ (takeWhile ( - Value (Vector n a) -> CodeGenFunction r (Value (Vector n b)) -inttofp = LLVM.inttofp - {- -Can be used for both integer and float types, -but we need it only for Float types, -because LLVM produces ugly code for Float and even more ugly code for Double. --} -signumLogical :: - (TypeNum.Positive n, - IsPrimitive a, IsPrimitive b, IsArithmetic b) => - (Value (Vector n a) -> - Value (Vector n a) -> - CodeGenFunction r (Value (Vector n b))) -> - Value (Vector n a) -> - CodeGenFunction r (Value (Vector n b)) -signumLogical gt x = do - let zero = LLVM.value LLVM.zero - negative <- gt zero x - positive <- gt x zero - A.sub negative positive - - -{- {- | This one does not use vectorized select. Cf. the outcommented signumInt. @@ -895,7 +835,7 @@ let zero = LLVM.value LLVM.zero negative <- LLVM.sadapt =<< A.cmp LLVM.CmpLT x zero positive <- LLVM.sadapt =<< A.cmp LLVM.CmpGT x zero - A.sub positive negative + A.sub negative positive signumWordGeneric :: (TypeNum.Positive n, @@ -928,210 +868,7 @@ signedFraction x = A.sub x =<< truncate x -floorGeneric :: - (IsFloating a, IsConst a, Real a, TypeNum.Positive n) => - Value (Vector n a) -> - CodeGenFunction r (Value (Vector n a)) -floorGeneric = floorLogical A.fcmp -{- | -On LLVM-2.6 and X86 this produces branch-free -but even slower code than 'fractionSelect', -since the comparison to booleans and -back to a floating point number is translated literally -to elementwise comparison, conversion to a 0 or -1 byte -and then to a floating point number. --} -fractionGeneric :: - (IsFloating a, IsConst a, Real a, TypeNum.Positive n) => - Value (Vector n a) -> - CodeGenFunction r (Value (Vector n a)) -fractionGeneric = fractionLogical A.fcmp - - -{- -These should be replaced by A.min, A.max, A.abs -when they work, eventually. --} -class (LLVM.IsSized a, LLVM.IsSized (Mask a), - LLVM.SizeOf a ~ LLVM.SizeOf (Mask a), - LLVM.IsPrimitive a, LLVM.IsPrimitive (Mask a), - LLVM.IsInteger (Mask a)) => - Maskable a where - type Mask a :: * - -instance Maskable Int8 where type Mask Int8 = Int8 -instance Maskable Int16 where type Mask Int16 = Int16 -instance Maskable Int32 where type Mask Int32 = Int32 -instance Maskable Int64 where type Mask Int64 = Int64 -instance Maskable Word8 where type Mask Word8 = Int8 -instance Maskable Word16 where type Mask Word16 = Int16 -instance Maskable Word32 where type Mask Word32 = Int32 -instance Maskable Word64 where type Mask Word64 = Int64 -instance Maskable Float where type Mask Float = Int32 -instance Maskable Double where type Mask Double = Int64 - -makeMask :: - (Maskable a, TypeNum.Positive n) => - Value (Vector n a) -> - Value (Vector n Bool) -> - CodeGenFunction r (Value (Vector n (Mask a))) -makeMask _ = LLVM.sadapt - - -minGeneric, maxGeneric :: - (IsConst a, Real a, Maskable a, TypeNum.Positive n) => - Value (Vector n a) -> - Value (Vector n a) -> - CodeGenFunction r (Value (Vector n a)) - -minGeneric x y = do - b <- makeMask x =<< A.cmp LLVM.CmpLT x y - selectLogical b x y - -maxGeneric x y = do - b <- makeMask x =<< A.cmp LLVM.CmpGT x y - selectLogical b x y - -absGeneric :: - (IsConst a, Real a, Maskable a, TypeNum.Positive n) => - Value (Vector n a) -> - CodeGenFunction r (Value (Vector n a)) -absGeneric x = - maxGeneric x =<< LLVM.neg x - -absAuto :: - (TypeNum.Positive n, TypeNum.Positive m, TypeNum.Positive k, - IsConst a, Real a, Maskable a) => - Ext.T (Value (Vector m a) -> CodeGenFunction r (Value (Vector m a))) -> - Ext.T (Value (Vector k a) -> CodeGenFunction r (Value (Vector k a))) -> - Value (Vector n a) -> - CodeGenFunction r (Value (Vector n a)) -absAuto byChunk byLargeChunk x = - absGeneric x - `Ext.run` - (Ext.with byChunk $ \f -> mapChunks f x) - `Ext.run` - (Ext.with2 byChunk byLargeChunk $ - \ f g -> mapChunks2 f g x) - - -{- | -LLVM.select on boolean vectors cannot be translated to X86 code in LLVM-2.6, -thus I code my own version that calls select on all elements. -This is slow but works. -When this issue is fixed, this function will be replaced by LLVM.select. --} -select :: - (LLVM.IsFirstClass a, IsPrimitive a, TypeNum.Positive n, - LLVM.CmpRet a, LLVM.CmpResult a ~ Bool) => - Value (Vector n Bool) -> - Value (Vector n a) -> - Value (Vector n a) -> - CodeGenFunction r (Value (Vector n a)) -select b x y = - map (uncurry3 LLVM.select) (b, x, y) - -{- | -'floor' implemented using 'select'. -This will need jumps. --} -_floorSelect :: - (Num a, IsFloating a, IsConst a, Real a, TypeNum.Positive n) => - Value (Vector n a) -> - CodeGenFunction r (Value (Vector n a)) -_floorSelect x = - do xr <- truncate x - b <- A.fcmp LLVM.FPOLE xr x - select b xr =<< A.sub xr =<< replicate (valueOf 1) - -{- | -'fraction' implemented using 'select'. -This will need jumps. --} -_fractionSelect :: - (Num a, IsFloating a, IsConst a, Real a, TypeNum.Positive n) => - Value (Vector n a) -> - CodeGenFunction r (Value (Vector n a)) -_fractionSelect x = - do xf <- signedFraction x - b <- A.fcmp LLVM.FPOGE xf (value LLVM.zero) - select b xf =<< A.add xf =<< replicate (valueOf 1) - - -{- | -Another implementation of 'select', -this time in terms of binary logical operations. -The selecting integers must be -(-1) for selecting an element from the first operand -and 0 for selecting an element from the second operand. -This leads to optimal code. - -On SSE41 this could be done with blendvps or blendvpd. --} -selectLogical :: - (LLVM.IsFirstClass a, IsPrimitive a, - LLVM.IsInteger i, IsPrimitive i, - LLVM.IsSized a, LLVM.IsSized i, - LLVM.SizeOf a ~ LLVM.SizeOf i, - TypeNum.Positive n) => - Value (Vector n i) -> - Value (Vector n a) -> - Value (Vector n a) -> - CodeGenFunction r (Value (Vector n a)) -selectLogical b x y = do --- bneg <- A.xor b - bneg <- LLVM.inv b - xm <- A.and b =<< LLVM.bitcastElements x - ym <- A.and bneg =<< LLVM.bitcastElements y - LLVM.bitcastElements =<< A.or xm ym - - -floorLogical :: - (IsFloating a, IsConst a, Real a, - IsPrimitive i, LLVM.IsInteger i, TypeNum.Positive n) => - (LLVM.FPPredicate -> - Value (Vector n a) -> - Value (Vector n a) -> - CodeGenFunction r (Value (Vector n i))) -> - Value (Vector n a) -> - CodeGenFunction r (Value (Vector n a)) -floorLogical cmp x = - do xr <- truncate x - b <- cmp LLVM.FPOGT xr x - A.add xr =<< LLVM.sitofp b - -fractionLogical :: - (IsFloating a, IsConst a, Real a, - IsPrimitive i, LLVM.IsInteger i, TypeNum.Positive n) => - (LLVM.FPPredicate -> - Value (Vector n a) -> - Value (Vector n a) -> - CodeGenFunction r (Value (Vector n i))) -> - Value (Vector n a) -> - CodeGenFunction r (Value (Vector n a)) -fractionLogical cmp x = - do xf <- signedFraction x - b <- cmp LLVM.FPOLT xf (value LLVM.zero) - A.sub xf =<< LLVM.sitofp b - - -order :: - (TypeNum.Positive n, TypeNum.Positive m, TypeNum.Positive k, - LLVM.IsFirstClass a, IsPrimitive a) => - (Value (Vector n a) -> Value (Vector n a) -> CodeGenFunction r (Value (Vector n a))) -> - Ext.T (Value (Vector m a) -> Value (Vector m a) -> CodeGenFunction r (Value (Vector m a))) -> - Ext.T (Value (Vector k a) -> Value (Vector k a) -> CodeGenFunction r (Value (Vector k a))) -> - (Value (Vector n a) -> Value (Vector n a) -> CodeGenFunction r (Value (Vector n a))) -order f byChunk byLargeChunk x y = - f x y - `Ext.run` - (Ext.with byChunk $ \psel -> zipChunksWith psel x y) - `Ext.run` - (Ext.with2 byChunk byLargeChunk $ - \ psel plsel -> zipChunks2With psel plsel x y) - - -- * target independent functions with target dependent optimizations {- | @@ -1190,77 +927,6 @@ mul = A.mul instance Arithmetic Float where - sum x = - Ext.runWhen (size x >= 4) (sumGeneric x) $ - Ext.with X86A.haddps $ \haddp -> - {- - We can make use of the following facts: - SSE3 has Float vectors of size 4, - there is an instruction for horizontal add. - -} - do chunkSum <- - foldl1 (M.liftJoin2 A.add) $ chop x - y <- haddp chunkSum (value undef) - z <- haddp y (value undef) -{- - y <- haddp chunkSum chunkSum - z <- haddp y y --} - extractelement z (valueOf 0) - - sumToPair x = - Ext.runWhen (size x >= 4) (getLowestPair x) $ - Ext.with X86A.haddps $ \haddp -> - let {- - reduce :: - [CodeGenFunction r (Value (Vector D4 Float))] -> - [CodeGenFunction r (Value (Vector D4 Float))] - -} - reduce [] = [] - reduce [_] = error "vector must have size power of two" - reduce (x0:x1:xs) = - M.liftJoin2 haddp x0 x1 : reduce xs - go [] = error "vector must not be empty" - go [c] = - getLowestPair - =<< flip haddp (value undef) - =<< c - go cs = go (reduce cs) - in go $ chop x - -{- -The haddps based implementation cumulate is slower than the generic one. -However, one day the x86 processors may implement a cumulative sum -which we could employ with this frame. - - cumulate a x = - Ext.runWhen (size x >= 4) (cumulateGeneric a x) $ - Ext.with X86.cumulate1s $ \cumulate1s -> do - (b,ys) <- - foldr - (\chunk0 cont a0 -> do - (a1,chunk1) <- cumulateFrom1 cumulate1s a0 =<< chunk0 - fmap (mapSnd (chunk1:)) (cont a1)) - (\a0 -> return (a0,[])) - (chop x) - a - y <- concat ys - return (b,y) --} - - dotProduct x y = - Ext.run (sum =<< A.mul x y) $ - Ext.with X86A.dpps $ \dpp -> - foldl1 (M.liftJoin2 A.add) $ - List.zipWith - (\mx my -> do - cx <- mx - cy <- my - flip extractelement (valueOf 0) - =<< dpp cx cy (valueOf 0xF1)) - (chop x) - (chop y) - instance Arithmetic Double where instance Arithmetic Int8 where @@ -1269,105 +935,12 @@ instance Arithmetic Int64 where instance Arithmetic Word8 where instance Arithmetic Word16 where +instance Arithmetic Word32 where instance Arithmetic Word64 where -instance Arithmetic Word32 where - mul x y = - A.mul x y - `Ext.run` - (Ext.with X86A.pmuludq128 $ \pmul -> - zipChunksWith - (\cx cy -> do - evenX <- shuffleMatchPlain1 cx - (constVector4 (constOf 0, undef, constOf 2, undef)) - evenY <- shuffleMatchPlain1 cy - (constVector4 (constOf 0, undef, constOf 2, undef)) - evenZ64 <- pmul evenX evenY - evenZ <- LLVM.bitcast evenZ64 - oddX <- shuffleMatchPlain1 cx - (constVector4 (constOf 1, undef, constOf 3, undef)) - oddY <- shuffleMatchPlain1 cy - (constVector4 (constOf 1, undef, constOf 3, undef)) - oddZ64 <- pmul oddX oddY - oddZ <- LLVM.bitcast oddZ64 - shuffleMatchPlain2 evenZ oddZ - (constVector4 (constOf 0, constOf 4, constOf 2, constOf 6))) - x y) - `Ext.run` - Ext.wrap X86C.sse41 (A.mul x y) -umul32to64 :: - (TypeNum.Positive n) => - Value (Vector n Word32) -> - Value (Vector n Word32) -> - CodeGenFunction r (Value (Vector n Word64)) -umul32to64 x y = - (do x64 <- map LLVM.zext x - y64 <- map LLVM.zext y - A.mul x64 y64) - `Ext.run` - (Ext.with X86A.pmuludq128 $ \pmul -> - zipChunksWith - -- save an initial shuffle - (\cx cy -> do - evenX <- shuffleMatchPlain1 cx - (constVector4 (constOf 0, undef, constOf 2, undef)) - evenY <- shuffleMatchPlain1 cy - (constVector4 (constOf 0, undef, constOf 2, undef)) - evenZ <- pmul evenX evenY - oddX <- shuffleMatchPlain1 cx - (constVector4 (constOf 1, undef, constOf 3, undef)) - oddY <- shuffleMatchPlain1 cy - (constVector4 (constOf 1, undef, constOf 3, undef)) - oddZ <- pmul oddX oddY -{- - shuffleMatchPlain2 evenZ oddZ - (constVector4 (constOf 0, constOf 2, constOf 1, constOf 3)) --} - assemble =<< (sequence $ - extract (valueOf 0) evenZ : - extract (valueOf 0) oddZ : - extract (valueOf 1) evenZ : - extract (valueOf 1) oddZ : - []) :: CodeGenFunction r (Value (Vector D4 Word64))) -{- - -- save the final shuffle - (\cx cy -> do - lowerX <- shuffleMatchPlain1 cx - (constVector4 (constOf 0, undef, constOf 1, undef)) - lowerY <- shuffleMatchPlain1 cy - (constVector4 (constOf 0, undef, constOf 1, undef)) - lowerZ <- pmul lowerX lowerY - upperX <- shuffleMatchPlain1 cx - (constVector4 (constOf 2, undef, constOf 3, undef)) - upperY <- shuffleMatchPlain1 cy - (constVector4 (constOf 2, undef, constOf 3, undef)) - upperZ <- pmul upperX upperY -{- - shuffleMatchPlain2 lowerZ upperZ - (constVector4 (constOf 0, constOf 1, constOf 2, constOf 3)) --} - concat [lowerZ, upperZ]) --} - x y) - - -constVector4 :: - (IsConst a) => - (ConstValue a, ConstValue a, ConstValue a, ConstValue a) -> - ConstValue (Vector D4 a) -constVector4 (a,b,c,d) = - LLVM.constVector $ a!:b!:c!:d!:Empty.Cons - - -{- | -Attention: -The rounding and fraction functions only work -for floating point values with maximum magnitude of @maxBound :: Int32@. -This way we save expensive handling of possibly seldom cases. --} -class (Arithmetic a, LLVM.CmpRet a, LLVM.CmpResult a ~ Bool, IsConst a) => +class (Arithmetic a, LLVM.CmpRet a, LLVM.IsPrimitive a, IsConst a) => Real a where min, max :: (TypeNum.Positive n) => @@ -1391,165 +964,62 @@ CodeGenFunction r (Value (Vector n a)) instance Real Float where - min = order minGeneric X86A.minps X86A.minps256 - max = order maxGeneric X86A.maxps X86A.maxps256 - abs x = Ext.run (absGeneric x) (Ext.with X86.absps ($x)) - signum x = - signumFloatGeneric x - `Ext.run` - (Ext.with X86.cmpps $ \cmp -> - inttofp =<< mapChunks (signumLogical (cmp LLVM.FPOGT)) x) -{- crashes LLVM-3.1 in JIT mode -Stack dump: -0. Running pass 'X86 DAG->DAG Instruction Selection' on function '@_fun1' -segmentation fault - - `Ext.run` - (Ext.with X86.cmpps256 $ \cmp -> - mapChunks (signumLogical - (\a b -> LLVM.sitofp =<< cmp LLVM.FPOGT a b)) x) --} -{- crashes LLVM-3.1 in JIT mode - only efficient in AVX2, where large integer vector subtraction is available - `Ext.run` - (Ext.with X86.cmpps256 $ \cmp -> - inttofp =<< mapChunks (signumLogical (cmp LLVM.FPOGT)) x) --} - {- - An IEEE specific implementation could do some bit manipulation: - s eeeeeeee mmmmmmmmmmmmmmmmmmmmmmm - Generate a pure power of two by clearing mantissa: - s eeeeeeee 00000000000000000000000 - Now subtract 1 in order to get the required bit mask for the mantissa - s eeeeeeee 11111111110000000000000 - multiply with 2 in order to correct exponent - and then do bitwise AND of the mask with the original number. - This method only works for numbers from 1 to 2^23-1, - that is the range is even more smaller - than that for the rounding via Int32. - -} - truncate x = - withRound - ((LLVM.inttofp . - (id :: Value (Vector n Int32) -> Value (Vector n Int32)) - <=< LLVM.fptoint) x) - X86A.roundps X86A.roundps256 return 3 x - floor x = - withRound - (floorGeneric x - `Ext.run` - (Ext.with X86.cmpps $ \cmp -> - mapChunks (floorLogical cmp) x) -{- LLVM-2.6 rearranges the MXCSR manipulations in an invalid way - `Ext.run` - (Ext.with2 (X86.withMXCSR (Bit.shiftL 1 13)) X86.cvtps2dq $ - \ with cvtps2dq -> with $ - LLVM.inttofp =<< mapChunks cvtps2dq x) --} - ) - X86A.roundps X86A.roundps256 return 1 x - fraction x = - withRound - (fractionGeneric x - `Ext.run` - (Ext.with X86.cmpps $ \cmp -> - mapChunks (fractionLogical cmp) x) -{- - `Ext.run` - (Ext.with2 (X86.withMXCSR (Bit.shiftL 1 13)) X86.cvtps2dq $ - \ with cvtps2dq -> with $ - A.sub x =<< LLVM.inttofp =<< mapChunks cvtps2dq x) --} - ) - X86A.roundps X86A.roundps256 (A.sub x) 1 x + min = Intrinsic.min + max = Intrinsic.max + abs = Intrinsic.abs + signum = signumFloatGeneric + truncate = Intrinsic.truncate + floor = Intrinsic.floor + fraction = A.fraction instance Real Double where - min = order minGeneric X86A.minpd X86A.minpd256 - max = order maxGeneric X86A.maxpd X86A.maxpd256 - abs x = Ext.run (absGeneric x) (Ext.with X86.abspd ($x)) - signum x = - signumFloatGeneric x - `Ext.run` - (Ext.with2 X86.cmppd X86A.cvtdq2pd $ \cmp tofp -> - mapChunks (signumLogical - (\a b -> do - c <- LLVM.bitcast =<< cmp LLVM.FPOGT a b - c0 <- extract (valueOf 0) (c :: Value (Vector D4 Int32)) - c1 <- extract (valueOf 2) c - tofp =<< assemble [c0,c1])) x) -{- crashes LLVM-3.1 in JIT mode - `Ext.run` - -- we could still optimize using mapChunks2 - (Ext.with2 X86.cmppd256 X86A.cvtdq2pd256 $ \cmp tofp -> - mapChunks (signumLogical - (\a b -> do - c <- LLVM.bitcast =<< cmp LLVM.FPOGT a b - c0 <- extract (valueOf 0) (c :: Value (Vector D8 Int32)) - c1 <- extract (valueOf 2) c - c2 <- extract (valueOf 4) c - c3 <- extract (valueOf 6) c - tofp =<< assemble [c0,c1,c2,c3])) x) --} - truncate x = - withRound - ((LLVM.inttofp . - (id :: Value (Vector n Int64) -> Value (Vector n Int64)) - <=< LLVM.fptoint) x) - X86A.roundpd X86A.roundpd256 return 3 x - floor x = - withRound - (floorGeneric x - `Ext.run` - (Ext.with X86.cmppd $ \cmp -> - mapChunks (floorLogical cmp) x)) - X86A.roundpd X86A.roundpd256 return 1 x - fraction x = - withRound - (fractionGeneric x - `Ext.run` - (Ext.with X86.cmppd $ \cmp -> - mapChunks (fractionLogical cmp) x)) - X86A.roundpd X86A.roundpd256 (A.sub x) 1 x + min = Intrinsic.min + max = Intrinsic.max + abs = Intrinsic.abs + signum = signumFloatGeneric + truncate = Intrinsic.truncate + floor = Intrinsic.floor + fraction = A.fraction instance Real Int8 where - min = order minGeneric X86A.pminsb128 X86A.pminsb256 - max = order maxGeneric X86A.pmaxsb128 X86A.pmaxsb256 - abs = absAuto X86A.pabsb128 X86A.pabsb256 + min = A.min + max = A.max + abs = A.abs signum = signumIntGeneric truncate = return floor = return fraction = const $ return (value LLVM.zero) instance Real Int16 where - min = order minGeneric X86A.pminsw128 X86A.pminsw256 - max = order maxGeneric X86A.pmaxsw128 X86A.pmaxsw256 - abs = absAuto X86A.pabsw128 X86A.pabsw256 + min = A.min + max = A.max + abs = A.abs signum = signumIntGeneric truncate = return floor = return fraction = const $ return (value LLVM.zero) instance Real Int32 where - min = order minGeneric X86A.pminsd128 X86A.pminsd256 - max = order maxGeneric X86A.pmaxsd128 X86A.pmaxsd256 - abs = absAuto X86A.pabsd128 X86A.pabsd256 + min = A.min + max = A.max + abs = A.abs signum = signumIntGeneric truncate = return floor = return fraction = const $ return (value LLVM.zero) instance Real Int64 where - min = minGeneric - max = maxGeneric - abs = absGeneric + min = A.min + max = A.max + abs = A.abs signum = signumIntGeneric truncate = return floor = return fraction = const $ return (value LLVM.zero) instance Real Word8 where - min = order minGeneric X86A.pminub128 X86A.pminub256 - max = order maxGeneric X86A.pmaxub128 X86A.pmaxub256 + min = A.min + max = A.max abs = return signum = signumWordGeneric truncate = return @@ -1557,8 +1027,8 @@ fraction = const $ return (value LLVM.zero) instance Real Word16 where - min = order minGeneric X86A.pminuw128 X86A.pminuw256 - max = order maxGeneric X86A.pmaxuw128 X86A.pmaxuw256 + min = A.min + max = A.max abs = return signum = signumWordGeneric truncate = return @@ -1566,8 +1036,8 @@ fraction = const $ return (value LLVM.zero) instance Real Word32 where - min = order minGeneric X86A.pminud128 X86A.pminud256 - max = order maxGeneric X86A.pmaxud128 X86A.pmaxud256 + min = A.min + max = A.max abs = return signum = signumWordGeneric truncate = return @@ -1575,8 +1045,8 @@ fraction = const $ return (value LLVM.zero) instance Real Word64 where - min = minGeneric - max = maxGeneric + min = A.min + max = A.max abs = return signum = signumWordGeneric truncate = return diff --git a/src/PrepareIntrinsics.hs b/src/PrepareIntrinsics.hs deleted file mode 100644 --- a/src/PrepareIntrinsics.hs +++ /dev/null @@ -1,326 +0,0 @@ -{-# LANGUAGE DeriveFunctor #-} -{-# LANGUAGE DeriveFoldable #-} -{-# LANGUAGE DeriveTraversable #-} -{- | -Parse an Intrinsics file and generate a Haskell interface to every intrinsic. -This is currently only tested and used for IntrinsicsX86.td -and relies on the flat structure of IntrinsicsX86.td. -In contrast to that, IntrinsicsPowerPC.td uses custom classes -and thus cannot be processed by this program. - -A safer way would be to invoke the llvm-tblgen utility in some way. - -1. We could write some Haskell or C++ code, - that queries the intrinsics from the include/llvm/Intrinsics.h interface. - -2. We could write a custom variant of llvm-tblgen - with a back-end that creates the Haskell interface for intrinsics. - This can be written in C++ or - we have to call the TableGen library functions from Haskell somehow. - -3. We could ask llvm-tblgen for a list of all records and parse its output. - This requires no C++ coding, - but we rely on the output format of @-print-records@. - - > llvm-tblgen -I /usr/local/llvm-3.1/include /usr/local/llvm-3.1/include/llvm/Intrinsics.td -print-records --} -module Main where - -import qualified Text.ParserCombinators.Parsec.Token as T -import qualified Text.ParserCombinators.Parsec.Language as L -import qualified Text.ParserCombinators.Parsec as Parsec -import Text.ParserCombinators.Parsec (CharParser, (<|>), ) - -import qualified Control.Monad.Trans.Writer as MW -import qualified Data.Map as M -import qualified Data.Set as S - -import qualified Data.Traversable as Trav -import qualified Data.Foldable as Fold -import qualified Data.List.HT as ListHT -import qualified Data.List as List -import qualified Data.Char as Char -import Control.Monad (mzero, ) -import Control.Functor.HT (void, ) -import Data.Maybe (fromMaybe, ) - -import qualified System.IO as IO - - -data Intrinsic typ = Intrinsic Name Name (FunctionType typ) - deriving (Show, Functor, Fold.Foldable, Trav.Traversable) - -data FunctionType typ = FunctionType [typ] [typ] - deriving (Show, Functor, Fold.Foldable, Trav.Traversable) - -type Name = String - -data QualName = QualName String String String - -newtype LLVMType = LLVMType String - deriving (Show) - -newtype HaskellType = HaskellType {haskellTypeDecons :: String} - deriving (Show, Eq, Ord) - - -gccBuiltinPrefix :: String -gccBuiltinPrefix = "__builtin_ia32_" - -unsignedFunctions :: S.Set String -unsignedFunctions = S.fromList $ map (gccBuiltinPrefix++) $ - "packusdw128" : - "packusdw256" : - "packuswb128" : - "packuswb256" : - "paddusb128" : - "paddusb256" : - "paddusw128" : - "paddusw256" : - "phminposuw128" : - "pmaddubsw128" : - "pmaddubsw256" : - "pmaxub128" : - "pmaxub256" : - "pmaxud128" : - "pmaxud256" : - "pmaxuw128" : - "pmaxuw256" : - "pminub128" : - "pminub256" : - "pminud128" : - "pminud256" : - "pminuw128" : - "pminuw256" : - "pmulhuw128" : - "pmulhuw256" : - "pmuludq128" : - "pmuludq256" : - "psubusb128" : - "psubusb256" : - "psubusw128" : - "psubusw256" : - "vphaddubd" : - "vphaddubq" : - "vphaddubw" : - "vphaddudq" : - "vphadduwd" : - "vphadduwq" : - -- it's only the flag set that is unsigned - -- the floating point operands are always signed - "roundps" : - "roundpd" : - "roundps256" : - "roundpd256" : - "roundss" : - "roundsd" : - "cmpps" : - "cmppd" : - "cmpps256" : - "cmppd256" : - "cmpss" : - "cmpsd" : - [] - -translateType :: - Bool -> LLVMType -> - MW.Writer (M.Map HaskellType HaskellType) HaskellType -translateType signed (LLVMType llvmTypeStr) = - let formatQType (mqual, typ) = - maybe "" (++".") mqual ++ typ - returnType shortType longType = do - MW.tell (M.singleton shortType longType) - return shortType - composedType = do - vec <- Parsec.optionMaybe $ do - void $ Parsec.char 'v' - Parsec.many1 Parsec.digit - prim <- Parsec.choice $ - (do void $ Parsec.char 'i' - fmap - (\n -> - if signed - then (Just "I", "Int"++n) - else (Just "W", "Word"++n)) $ - Parsec.many1 Parsec.digit) : - (do void $ Parsec.char 'f' - n <- Parsec.many1 Parsec.digit - case n of - "32" -> return (Nothing, "Float") - "64" -> return (Nothing, "Double") - _ -> return $ (Just "LLVM", "FP" ++ n)) : - [] - return $ - case vec of - Nothing -> return $ HaskellType $ "LLVM.Value " ++ formatQType prim - Just d -> - returnType - (HaskellType $ "V" ++ d ++ snd prim) - (HaskellType $ - "LLVM.Value (LLVM.Vector TypeNum.D" ++ - d ++ " " ++ formatQType prim ++ ")") - p = do - void $ Parsec.string "llvm_" - haskType <- Parsec.choice $ - (Parsec.string "x86mmx" >> - return (returnType (HaskellType "MMX") (HaskellType "LLVM.Value (LLVM.Vector TypeNum.D8 W.Word8)"))) : - (Parsec.string "ptr" >> - return (return (HaskellType "LLVM.Value (Ptr ())"))) : - composedType : - [] - void $ Parsec.string "_ty" - return haskType - in case Parsec.parse p "" llvmTypeStr of - Left _msg -> - let typeSyn = HaskellType $ - case llvmTypeStr of - c:cs -> Char.toUpper c : cs - _ -> "" - in do - MW.tell (M.singleton typeSyn (HaskellType "LLVM.Value ()")) - return typeSyn - Right act -> act - -splitName :: Name -> QualName -splitName name = - let p = do - void $ Parsec.string "int_" - arch <- Parsec.many1 Parsec.alphaNum - void $ Parsec.char '_' - feature <- Parsec.many1 Parsec.alphaNum - void $ Parsec.char '_' - stem <- Parsec.many1 Parsec.anyChar - return $ QualName arch feature stem - in case Parsec.parse p "" name of - Left _msg -> QualName "" "" name - Right qname -> qname - -featureMap :: M.Map String String -featureMap = M.fromList $ - ("sse", "sse1") : - ("aesni", "aes") : - ("3dnow", "amd3dnow") : - ("3dnowa", "amd3dnowa") : - [] - -formatIntrinsicInHaskell :: Intrinsic HaskellType -> String -formatIntrinsicInHaskell - (Intrinsic name gccblt (FunctionType parameters results)) = - let (QualName _arch feature stem) = splitName name - dotStem = map (\c -> case c of '_' -> '.'; _ -> c) stem - haskName = - fromMaybe gccblt $ - ListHT.maybePrefixOf gccBuiltinPrefix gccblt - resultStr = - if null results - then "LLVM.Value ()" - else List.intercalate ", " $ map haskellTypeDecons results - in unlines $ - (haskName ++ " :: Ext.T (" ++ - concatMap (\(HaskellType typ) -> typ ++ " -> ") parameters ++ - "LLVM.CodeGenFunction r (" ++ resultStr ++ "))") : - (haskName ++ " = Ext.intrinsic ExtX86." ++ - M.findWithDefault feature feature featureMap ++ - " " ++ show dotStem) : - [] - -convertIntrinsics :: [Intrinsic LLVMType] -> String -convertIntrinsics intrinsics = - unlines $ - "{- Do not edit! This file was created with the PrepareIntrinsics tool. -}" : - "module LLVM.Extra.Extension.X86Auto where" : - "" : - "import qualified LLVM.Extra.Extension as Ext" : - "import qualified LLVM.Extra.ExtensionCheck.X86 as ExtX86" : - "import qualified LLVM.Core as LLVM" : - "import qualified Type.Data.Num.Decimal as TypeNum" : - "import qualified Data.Int as I" : - "import qualified Data.Word as W" : - "import Foreign.Ptr (Ptr, )" : - "" : - case MW.runWriter $ - mapM (\intr@(Intrinsic _ gccblt _) -> - Trav.traverse (translateType (not $ S.member gccblt unsignedFunctions)) intr) $ - filter (\(Intrinsic _ gccblt _) -> not $ null gccblt) intrinsics of - (funcs, types) -> - (map (\(HaskellType short, HaskellType long) -> - "type " ++ short ++ " = " ++ long) $ - M.toList types) ++ - "" : - (map formatIntrinsicInHaskell funcs) - -lexer :: T.TokenParser st -lexer = - T.makeTokenParser $ L.emptyDef { - L.commentStart = "/*", - L.commentEnd = "*/", - L.commentLine = "//", - L.nestedComments = False, - L.identStart = identifierStart, - L.identLetter = identifierLetter, - L.opStart = mzero, - L.opLetter = mzero, - L.caseSensitive = True, - L.reservedNames = [ "let", "def", "in" ], - L.reservedOpNames = [ "=", ":", "," ] - } - -identifierStart, identifierLetter :: CharParser st Char -identifierStart = Parsec.letter <|> Parsec.char '_' - -identifierLetter = - Parsec.alphaNum <|> Parsec.char '_' <|> Parsec.char '.' - - -gccBuiltin :: CharParser st String -gccBuiltin = do - T.reserved lexer "GCCBuiltin" - T.angles lexer $ T.stringLiteral lexer - -llvmType :: CharParser st LLVMType -llvmType = fmap LLVMType $ T.identifier lexer - -intrinsic :: CharParser st (FunctionType LLVMType) -intrinsic = - Parsec.between (T.reserved lexer "Intrinsic") (T.semi lexer) $ - T.angles lexer $ do - results <- T.brackets lexer $ T.commaSep lexer llvmType - void $ T.comma lexer - parameters <- T.brackets lexer $ T.commaSep lexer llvmType - Parsec.optional $ do - void $ T.comma lexer - _attributes <- T.brackets lexer $ T.commaSep lexer $ T.identifier lexer - return () - return $ FunctionType parameters results - -letBlock :: CharParser st [Intrinsic LLVMType] -letBlock = do - T.reserved lexer "let" - T.reserved lexer "TargetPrefix" - void $ T.symbol lexer "=" - _prefix <- T.stringLiteral lexer - T.reserved lexer "in" - T.braces lexer $ Parsec.many $ do - T.reserved lexer "def" - name <- T.identifier lexer - void $ T.colon lexer - gccblt <- Parsec.option "" $ do - gccblt <- gccBuiltin - void $ T.comma lexer - return gccblt - intr <- intrinsic - return $ Intrinsic name gccblt intr - -parser :: CharParser st [Intrinsic LLVMType] -parser = - fmap concat $ Parsec.many1 letBlock - -main :: IO () -main = do - parsed <- - Parsec.parseFromFile (T.whiteSpace lexer >> parser) - "/usr/local/llvm-3.8/include/llvm/IR/IntrinsicsX86.td" - case parsed of - Left msg -> IO.hPutStrLn IO.stderr $ show msg - Right intrinsics -> - writeFile "src/LLVM/Extra/Extension/X86Auto.hs" $ convertIntrinsics intrinsics diff --git a/test/LLVM/Extra/VectorAlt.hs b/test/LLVM/Extra/VectorAlt.hs new file mode 100644 --- /dev/null +++ b/test/LLVM/Extra/VectorAlt.hs @@ -0,0 +1,224 @@ +{-# LANGUAGE TypeFamilies #-} +{-# LANGUAGE FlexibleContexts #-} +{- | +This maintains old code for LLVM-2.6 +where vector comparison and select on X86 +did not work or generated cumbersome assembly code. +It may still be useful for testing. +-} +module LLVM.Extra.VectorAlt where + +import qualified LLVM.Extra.Vector as Vector +import qualified LLVM.Extra.Arithmetic as A + +import qualified LLVM.Util.Intrinsic as Intrinsic +import qualified LLVM.Core.Guided as Guided +import qualified LLVM.Core as LLVM +import LLVM.Core + (CodeGenFunction, Value, valueOf, value, Vector, + CmpRet, IsConst, IsArithmetic, IsFloating, IsPrimitive) + +import qualified Type.Data.Num.Decimal as TypeNum + +import Data.Tuple.HT (uncurry3, ) + +import Data.Int (Int8, Int16, Int32, Int64, ) +import Data.Word (Word8, Word16, Word32, Word64, ) + +import Prelude hiding (max, min, abs, signum, floor, truncate) + + + +{- +Can be used for both integer and float types, +but we need it only for Float types, +because LLVM produces ugly code for Float and even more ugly code for Double. +-} +signum :: + (TypeNum.Positive n, + IsPrimitive a, IsPrimitive b, IsArithmetic b) => + (Value (Vector n a) -> + Value (Vector n a) -> + CodeGenFunction r (Value (Vector n b))) -> + Value (Vector n a) -> + CodeGenFunction r (Value (Vector n b)) +signum gt x = do + let zero = LLVM.value LLVM.zero + negative <- gt zero x + positive <- gt x zero + A.sub negative positive + +ext2 :: + (TypeNum.Positive n) => + Value (Vector n Bool) -> + CodeGenFunction r (Value (Vector n (LLVM.IntN TypeNum.D2))) +ext2 = Guided.extBool Guided.vector + +{- | +This has least instruction count for Vector D4 Float on X86. +-} +signumFloat :: + (TypeNum.Positive n, + IsPrimitive a, IsArithmetic a, IsFloating a, + LLVM.CmpRet a, LLVM.CmpResult a ~ Bool) => + Value (Vector n a) -> + CodeGenFunction r (Value (Vector n a)) +signumFloat x = do + let zero = LLVM.value LLVM.zero + negative <- ext2 =<< A.cmp LLVM.CmpLT x zero + positive <- ext2 =<< A.cmp LLVM.CmpGT x zero + LLVM.sitofp =<< A.sub negative positive + + +select :: + (TypeNum.Positive n, LLVM.IsFirstClass a, IsPrimitive a, + LLVM.CmpRet a, LLVM.CmpResult a ~ Bool) => + Value (Vector n Bool) -> + Value (Vector n a) -> + Value (Vector n a) -> + CodeGenFunction r (Value (Vector n a)) +select b x y = + Vector.map (uncurry3 LLVM.select) (b, x, y) + + +floor :: + (TypeNum.Positive n, IsFloating a, Vector.Real a) => + Value (Vector n a) -> + CodeGenFunction r (Value (Vector n a)) +floor = floorLogical A.fcmp + +fraction :: + (TypeNum.Positive n, IsFloating a, Vector.Real a) => + Value (Vector n a) -> + CodeGenFunction r (Value (Vector n a)) +fraction = fractionLogical A.fcmp + + +floorLogical :: + (TypeNum.Positive n, IsFloating a, Vector.Real a, + IsPrimitive i, LLVM.IsInteger i) => + (LLVM.FPPredicate -> + Value (Vector n a) -> + Value (Vector n a) -> + CodeGenFunction r (Value (Vector n i))) -> + Value (Vector n a) -> + CodeGenFunction r (Value (Vector n a)) +floorLogical cmp x = do + xr <- Intrinsic.truncate x + b <- cmp LLVM.FPOGT xr x + A.add xr =<< LLVM.sitofp b + +fractionLogical :: + (TypeNum.Positive n, IsFloating a, Vector.Real a, + IsPrimitive i, LLVM.IsInteger i) => + (LLVM.FPPredicate -> + Value (Vector n a) -> + Value (Vector n a) -> + CodeGenFunction r (Value (Vector n i))) -> + Value (Vector n a) -> + CodeGenFunction r (Value (Vector n a)) +fractionLogical cmp x = do + xf <- Vector.signedFraction x + b <- cmp LLVM.FPOLT xf (value LLVM.zero) + A.sub xf =<< LLVM.sitofp b + + +{- | +'floor' implemented using 'select'. +This will need jumps. +-} +floorSelect :: + (TypeNum.Positive n, Num a, IsFloating a, Vector.Real a) => + Value (Vector n a) -> + CodeGenFunction r (Value (Vector n a)) +floorSelect x = do + xr <- Intrinsic.truncate x + b <- A.fcmp LLVM.FPOLE xr x + select b xr =<< A.sub xr =<< Vector.replicate (valueOf 1) + +{- | +'fraction' implemented using 'select'. +This will need jumps. +-} +fractionSelect :: + (TypeNum.Positive n, Num a, IsFloating a, Vector.Real a) => + Value (Vector n a) -> + CodeGenFunction r (Value (Vector n a)) +fractionSelect x = do + xf <- Vector.signedFraction x + b <- A.fcmp LLVM.FPOGE xf (value LLVM.zero) + select b xf =<< A.add xf =<< Vector.replicate (valueOf 1) + + +class (LLVM.IsSized a, LLVM.IsSized (Mask a), + LLVM.SizeOf a ~ LLVM.SizeOf (Mask a), + LLVM.IsPrimitive a, LLVM.IsPrimitive (Mask a), + LLVM.IsInteger (Mask a)) => + Maskable a where + type Mask a :: * + +instance Maskable Int8 where type Mask Int8 = Int8 +instance Maskable Int16 where type Mask Int16 = Int16 +instance Maskable Int32 where type Mask Int32 = Int32 +instance Maskable Int64 where type Mask Int64 = Int64 +instance Maskable Word8 where type Mask Word8 = Int8 +instance Maskable Word16 where type Mask Word16 = Int16 +instance Maskable Word32 where type Mask Word32 = Int32 +instance Maskable Word64 where type Mask Word64 = Int64 +instance Maskable Float where type Mask Float = Int32 +instance Maskable Double where type Mask Double = Int64 + +makeMask :: + (Maskable a, TypeNum.Positive n) => + Value (Vector n a) -> + Value (Vector n Bool) -> + CodeGenFunction r (Value (Vector n (Mask a))) +makeMask _ = Guided.extBool Guided.vector + + +min, max :: + (IsConst a, IsArithmetic a, CmpRet a, Maskable a, TypeNum.Positive n) => + Value (Vector n a) -> + Value (Vector n a) -> + CodeGenFunction r (Value (Vector n a)) + +min x y = do + b <- makeMask x =<< A.cmp LLVM.CmpLT x y + selectLogical b x y + +max x y = do + b <- makeMask x =<< A.cmp LLVM.CmpGT x y + selectLogical b x y + +abs :: + (IsConst a, IsArithmetic a, CmpRet a, Maskable a, TypeNum.Positive n) => + Value (Vector n a) -> + CodeGenFunction r (Value (Vector n a)) +abs x = max x =<< LLVM.neg x + + +{- | +Another implementation of 'select', +this time in terms of binary logical operations. +The selecting integers must be +(-1) for selecting an element from the first operand +and 0 for selecting an element from the second operand. +This leads to optimal code. + +On SSE41 this could be done with blendvps or blendvpd. +-} +selectLogical :: + (LLVM.IsFirstClass a, IsPrimitive a, + LLVM.IsInteger i, IsPrimitive i, + LLVM.IsSized a, LLVM.IsSized i, + LLVM.SizeOf a ~ LLVM.SizeOf i, + TypeNum.Positive n) => + Value (Vector n i) -> + Value (Vector n a) -> + Value (Vector n a) -> + CodeGenFunction r (Value (Vector n a)) +selectLogical b x y = do + bneg <- LLVM.inv b + xm <- A.and b =<< Guided.bitcast Guided.vector x + ym <- A.and bneg =<< Guided.bitcast Guided.vector y + Guided.bitcast Guided.vector =<< A.or xm ym diff --git a/test/Main.hs b/test/Main.hs new file mode 100644 --- /dev/null +++ b/test/Main.hs @@ -0,0 +1,17 @@ +module Main where + +import qualified Test.Vector as Vector + +import qualified LLVM.Core as LLVM + +import Data.Tuple.HT (mapFst) + +import qualified Test.QuickCheck as QC + + +main :: IO () +main = do + LLVM.initializeNativeTarget + + mapM_ (\(msg,prop) -> putStr (msg++": ") >> prop >>= QC.quickCheck) $ + map (mapFst ("Vector."++)) Vector.tests diff --git a/test/Test/Vector.hs b/test/Test/Vector.hs new file mode 100644 --- /dev/null +++ b/test/Test/Vector.hs @@ -0,0 +1,327 @@ +{-# LANGUAGE ForeignFunctionInterface #-} +{-# LANGUAGE TypeOperators #-} +{-# LANGUAGE TypeFamilies #-} +module Test.Vector where + +import qualified LLVM.Extra.ScalarOrVectorPrivate as SoVPriv +import qualified LLVM.Extra.ScalarOrVector as SoV +import qualified LLVM.Extra.VectorAlt as VectorAlt +import qualified LLVM.Extra.Vector as Vector +import qualified LLVM.Extra.Memory as Mem +import qualified LLVM.Extra.Class as Class +import qualified LLVM.ExecutionEngine as EE +import qualified LLVM.Core as LLVM + +import qualified Type.Data.Num.Decimal as TypeNum +import Type.Base.Proxy (Proxy(Proxy)) + +import qualified Foreign.Marshal as Marsh +import Foreign.Storable.Tuple () +import Foreign.Storable (Storable, peek) +import Foreign.Ptr (FunPtr, Ptr) + +import qualified Data.Traversable as Trav +import qualified Data.Foldable as Fold +import qualified Data.Bits as Bits +import Data.Word (Word8, Word16, Word32) +import Data.Int (Int8, Int32) + +import qualified Test.QuickCheck as QC +import qualified Test.QuickCheck.Monadic as QCMon + +import Control.Applicative (liftA2, pure) + +import qualified Prelude as P +import Prelude hiding (min, max) + + +type V4 = LLVM.Vector TypeNum.D4 +type V5 = LLVM.Vector TypeNum.D5 +type V4Word32 = V4 Word32 +type V4Int32 = V4 Int32 +type V4Float = V4 Float + +type Importer func = FunPtr func -> func + +generateFunction :: + EE.ExecutionFunction f => + Importer f -> LLVM.CodeGenModule (LLVM.Function f) -> IO f +generateFunction imprt code = do + m <- LLVM.newModule + fn <- do + func <- LLVM.defineModule m $ LLVM.setTarget LLVM.hostTriple >> code + EE.runEngineAccessWithModule m $ EE.getExecutionFunction imprt func + LLVM.writeBitcodeToFile "test.bc" m + return fn + + +foreign import ccall safe "dynamic" derefTestCasePtr :: + Importer (Ptr inp -> Ptr out -> IO ()) + +modul :: + (Mem.Struct inp ~ minp, LLVM.IsType minp, Mem.C inp, + Mem.Struct out ~ mout, LLVM.IsType mout, Mem.C out) => + (inp -> LLVM.CodeGenFunction () out) -> + LLVM.CodeGenModule (LLVM.Function (Ptr minp -> Ptr mout -> IO ())) +modul codegen = + LLVM.createFunction LLVM.ExternalLinkage $ \xPtr yPtr -> do + flip Mem.store yPtr =<< codegen =<< Mem.load xPtr + LLVM.ret () + +run :: + (Mem.Struct inp ~ minp, LLVM.IsType minp, Mem.C inp, + Mem.Struct out ~ mout, LLVM.IsType mout, Mem.C out, + Storable a, Class.MakeValueTuple a, Class.ValueTuple a ~ inp, + Storable b, Class.MakeValueTuple b, Class.ValueTuple b ~ out, + Show a, QC.Arbitrary a) => + (inp -> LLVM.CodeGenFunction () out) -> + (a -> b -> Bool) -> + IO QC.Property +run codegen predicate = do + funIO <- generateFunction derefTestCasePtr $ modul codegen + return $ QC.property $ \x -> + QCMon.monadicIO $ do + y <- + QCMon.run $ + Marsh.with x $ \xPtr -> + Marsh.alloca $ \yPtr -> + funIO (Mem.castTuplePtr xPtr) (Mem.castTuplePtr yPtr) >> + peek yPtr + QCMon.assert $ predicate x y + + +vec4 :: V4 a -> V4 a +vec4 = id + + +unop :: + (LLVM.Value V4Int32 -> LLVM.CodeGenFunction () (LLVM.Value V4Int32)) -> + (Int32 -> Int32) -> + IO QC.Property +unop codegen fun = + run codegen (\x y -> fmap fun (vec4 x) == vec4 y) + +unopFloat :: + (LLVM.Value V4Float -> LLVM.CodeGenFunction () (LLVM.Value V4Float)) -> + (Float -> Float) -> + IO QC.Property +unopFloat codegen fun = + run codegen (\x y -> fmap fun (vec4 x) == vec4 y) + + +binop :: + ((TypeNum.D4 TypeNum.:*: LLVM.SizeOf am) ~ size, TypeNum.Natural size, + Mem.FirstClass a, Mem.Stored a ~ am, LLVM.IsSized am, + QC.Arbitrary a, Show a, Eq a, Storable a, LLVM.IsConst a, + LLVM.IsPrimitive a, LLVM.IsPrimitive am) => + (LLVM.Value (V4 a) -> LLVM.Value (V4 a) -> + LLVM.CodeGenFunction () (LLVM.Value (V4 a))) -> + (a -> a -> a) -> + IO QC.Property +binop codegen fun = + run (uncurry codegen) + (\(x,y) z -> liftA2 fun (vec4 x) (vec4 y) == vec4 z) + +binopInt :: + (LLVM.Value V4Int32 -> LLVM.Value V4Int32 -> + LLVM.CodeGenFunction () (LLVM.Value V4Int32)) -> + (Int32 -> Int32 -> Int32) -> + IO QC.Property +binopInt = binop + + +type Int2 = LLVM.IntN TypeNum.D2 +type Int3 = LLVM.IntN TypeNum.D3 +type Word2 = LLVM.WordN TypeNum.D2 +type Word3 = LLVM.WordN TypeNum.D3 + +vectorise :: + (TypeNum.Positive n, Integral a) => + Integer -> a -> LLVM.Vector n Integer +vectorise modu x = + snd $ Trav.mapAccumL (\xi f -> f xi) (toInteger x) $ + pure (\xi -> divMod xi modu) + +unpackInts :: + (TypeNum.Positive n, TypeNum.Positive d, Integral a) => + Integer -> a -> LLVM.Vector n (LLVM.IntN d) +unpackInts modu = + fmap + (\x -> + LLVM.IntN $ + if Bits.shiftR modu 1 Bits..&. x /= 0 + then toInteger x - modu + else toInteger x) . + vectorise modu + +unpackWords :: + (TypeNum.Positive n, TypeNum.Positive d, Integral a) => + Integer -> a -> LLVM.Vector n (LLVM.WordN d) +unpackWords modu = fmap LLVM.WordN . vectorise modu + +unpackInt2 :: Word8 -> V4 Int2 +unpackInt2 = unpackInts 4 + +unpackWord2 :: Word8 -> V4 Word2 +unpackWord2 = unpackWords 4 + +unpackInt3 :: Word16 -> V5 Int3 +unpackInt3 = unpackInts 8 + +unpackWord3 :: Word16 -> V5 Word3 +unpackWord3 = unpackWords 8 + +binopV4I2 :: + (Eq a, LLVM.IsPrimitive a, LLVM.IsSized a, LLVM.SizeOf a ~ TypeNum.D2) => + (Word8 -> V4 a) -> + (LLVM.Value (V4 a) -> LLVM.Value (V4 a) -> + LLVM.CodeGenFunction () (LLVM.Value (V4 a))) -> + (a -> a -> a) -> + IO QC.Property +binopV4I2 unpackBits codegen fun = + run + (\(x,y) -> do + vx <- LLVM.bitcast x + vy <- LLVM.bitcast y + vz <- codegen vx vy + LLVM.bitcast vz) + (\(x,y) z -> + liftA2 fun (unpackBits x) (unpackBits y) == unpackBits z) + +type Code15 r = LLVM.CodeGenFunction r (LLVM.Value (LLVM.WordN TypeNum.D15)) + +binopV5I3 :: + (Eq a, LLVM.IsPrimitive a, LLVM.IsSized a, LLVM.SizeOf a ~ TypeNum.D3) => + (Word16 -> V5 a) -> + (LLVM.Value (V5 a) -> LLVM.Value (V5 a) -> + LLVM.CodeGenFunction () (LLVM.Value (V5 a))) -> + (a -> a -> a) -> + IO QC.Property +binopV5I3 unpackBits codegen fun = + run + (\(x,y) -> do + vx <- LLVM.bitcast =<< (LLVM.trunc x :: Code15 r) + vy <- LLVM.bitcast =<< (LLVM.trunc y :: Code15 r) + vz <- codegen vx vy + LLVM.zext =<< (LLVM.bitcast vz :: Code15 r)) + (\(x,y) z -> + liftA2 fun (unpackBits x) (unpackBits y) == unpackBits z) + +binopInt8 :: + (LLVM.Value (V4 Int8) -> LLVM.Value (V4 Int8) -> + LLVM.CodeGenFunction () (LLVM.Value (V4 Int8))) -> + (Int8 -> Int8 -> Int8) -> + IO QC.Property +binopInt8 = binop + +binopWord8 :: + (LLVM.Value (V4 Word8) -> LLVM.Value (V4 Word8) -> + LLVM.CodeGenFunction () (LLVM.Value (V4 Word8))) -> + (Word8 -> Word8 -> Word8) -> + IO QC.Property +binopWord8 = binop + + +addSat, subSat :: (Bounded a, Integral a) => a -> a -> a +addSat = addSatMan (toInteger, fromInteger) +subSat = subSatMan (toInteger, fromInteger) + +addSatMan, subSatMan :: + (Bounded a) => (a -> Integer, Integer -> a) -> a -> a -> a +addSatMan = opSat (+) +subSatMan = opSat (-) + +convertIntN :: Proxy d -> (LLVM.IntN d -> Integer, Integer -> LLVM.IntN d) +convertIntN Proxy = (\(LLVM.IntN n) -> n, LLVM.IntN) + +convertWordN :: Proxy d -> (LLVM.WordN d -> Integer, Integer -> LLVM.WordN d) +convertWordN Proxy = (\(LLVM.WordN n) -> n, LLVM.WordN) + +opSat :: + (Bounded a) => + (Integer -> Integer -> Integer) -> + (a -> Integer, Integer -> a) -> + a -> a -> a +opSat op (toIntg, fromIntg) x y = + fromIntg $ + P.max (toIntg $ minBound `asTypeOf` x) $ + P.min (toIntg $ maxBound `asTypeOf` x) $ + op (toIntg x) (toIntg y) + + +fraction :: RealFrac a => a -> a +fraction x = x - fromInteger (floor x) + + +split :: String -> (a -> b -> c) -> (a,a) -> b -> [(String, c)] +split name driver (intrinsic, fallback) f = + (name ++ ".intrinsic", driver intrinsic f) : + (name ++ ".fallback", driver fallback f) : + [] + +tests :: [(String, IO QC.Property)] +tests = + ("abs", unop Vector.abs P.abs) : + ("signum", unop Vector.signum P.signum) : + ("Alt.abs", unop VectorAlt.abs P.abs) : + + ("min", binopInt Vector.min P.min) : + ("max", binopInt Vector.max P.max) : + ("Alt.min", binopInt VectorAlt.min P.min) : + ("Alt.max", binopInt VectorAlt.max P.max) : + + split "addSat.Word8" binopWord8 (SoV.addSat, SoVPriv.uaddSat) addSat ++ + split "subSat.Word8" binopWord8 (SoV.subSat, SoVPriv.usubSat) subSat ++ + split "addSat.Int8" binopInt8 (SoV.addSat, SoVPriv.saddSat) addSat ++ + split "subSat.Int8" binopInt8 (SoV.subSat, SoVPriv.ssubSat) subSat ++ + + split "addSat.Word3" + (binopV5I3 unpackWord3) (SoV.addSat, SoVPriv.uaddSat) + (addSatMan $ convertWordN TypeNum.d3) ++ + split "subSat.Word3" + (binopV5I3 unpackWord3) (SoV.subSat, SoVPriv.usubSat) + (subSatMan $ convertWordN TypeNum.d3) ++ + split "addSat.Int3" + (binopV5I3 unpackInt3) (SoV.addSat, SoVPriv.saddSat) + (addSatMan $ convertIntN TypeNum.d3) ++ + split "subSat.Int3" + (binopV5I3 unpackInt3) (SoV.subSat, SoVPriv.ssubSat) + (subSatMan $ convertIntN TypeNum.d3) ++ + + split "addSat.Word2" + (binopV4I2 unpackWord2) (SoV.addSat, SoVPriv.uaddSat) + (addSatMan $ convertWordN TypeNum.d2) ++ + split "subSat.Word2" + (binopV4I2 unpackWord2) (SoV.subSat, SoVPriv.usubSat) + (subSatMan $ convertWordN TypeNum.d2) ++ + split "addSat.Int2" + (binopV4I2 unpackInt2) (SoV.addSat, SoVPriv.saddSat) + (addSatMan $ convertIntN TypeNum.d2) ++ + split "subSat.Int2" + (binopV4I2 unpackInt2) (SoV.subSat, SoVPriv.ssubSat) + (subSatMan $ convertIntN TypeNum.d2) ++ + + ("sum", + run Vector.sum (\x y -> Fold.sum (vec4 x) == (y::Int32))) : + ("cumulate", + run + (uncurry Vector.cumulate) + (\(x0,xv) (y0,yv) -> + scanl (+) x0 (Fold.toList (vec4 xv)) + == + Fold.toList (vec4 yv) ++ [y0::Int32])) : + ("dot", + run + (uncurry Vector.dotProduct) + (\(x,y) z -> + Fold.sum (liftA2 (*) (vec4 x) (vec4 y)) == (z::Int32))) : + + ("truncate", unopFloat Vector.truncate (fromInteger . P.truncate)) : + ("floor", unopFloat Vector.floor (fromInteger . P.floor)) : + ("fraction", unopFloat Vector.fraction fraction) : + + ("floorLogical", unopFloat VectorAlt.floor (fromInteger . P.floor)) : + ("fractionLogical", unopFloat VectorAlt.fraction fraction) : + ("floorSelect", unopFloat VectorAlt.floorSelect (fromInteger . P.floor)) : + ("fractionSelect", unopFloat VectorAlt.fractionSelect fraction) : + [] diff --git a/x86/cpuid/LLVM/Extra/ExtensionCheck/X86.hs b/x86/cpuid/LLVM/Extra/ExtensionCheck/X86.hs deleted file mode 100644 --- a/x86/cpuid/LLVM/Extra/ExtensionCheck/X86.hs +++ /dev/null @@ -1,71 +0,0 @@ -module LLVM.Extra.ExtensionCheck.X86 ( - sse1, sse2, sse3, ssse3, sse41, sse42, avx, avx2, avx512, - fma, amd3dnow, amd3dnowa, aes, sse4a, - ) where - -import qualified LLVM.Extra.Extension as Ext -import qualified System.Cpuid as CPUID -import qualified System.Unsafe as Unsafe - -{- -I expect that the cpuid does not suddenly change -and thus calling Unsafe.performIO is safe. --} -subtarget :: - String -> - (CPUID.FlagSet CPUID.Feature1C -> CPUID.FlagSet CPUID.Feature1D -> Bool) -> - Ext.Subtarget -subtarget name q = - Ext.Subtarget "x86" name - (return $ Unsafe.performIO $ check q) - -check :: - (CPUID.FlagSet CPUID.Feature1C -> CPUID.FlagSet CPUID.Feature1D -> Bool) -> - IO Bool -check q = - fmap (uncurry q) $ CPUID.features - - --- * target specific extensions - -sse1 :: Ext.Subtarget -sse1 = subtarget "sse" (\_ecx edx -> CPUID.testFlag CPUID.sse edx) - -sse2 :: Ext.Subtarget -sse2 = subtarget "sse2" (\_ecx edx -> CPUID.testFlag CPUID.sse2 edx) - -sse3 :: Ext.Subtarget -sse3 = subtarget "sse3" (\ecx _edx -> CPUID.testFlag CPUID.sse3 ecx) - -ssse3 :: Ext.Subtarget -ssse3 = subtarget "ssse3" (\ecx _edx -> CPUID.testFlag CPUID.ssse3 ecx) - -sse41 :: Ext.Subtarget -sse41 = subtarget "sse41" (\ecx _edx -> CPUID.testFlag CPUID.sse4_1 ecx) - -sse42 :: Ext.Subtarget -sse42 = subtarget "sse42" (\ecx _edx -> CPUID.testFlag CPUID.sse4_2 ecx) - -avx :: Ext.Subtarget -avx = subtarget "avx" (\ecx _edx -> CPUID.testFlag CPUID.avx ecx) - -avx2 :: Ext.Subtarget -avx2 = subtarget "avx2" (\ _ecx _edx -> False) - -avx512 :: Ext.Subtarget -avx512 = subtarget "avx512" (\ _ecx _edx -> False) - -fma :: Ext.Subtarget -fma = subtarget "fma" (\ ecx _edx -> CPUID.testFlag CPUID.fma ecx) - -amd3dnow :: Ext.Subtarget -amd3dnow = subtarget "3dnow" (\ _ecx _edx -> False) - -amd3dnowa :: Ext.Subtarget -amd3dnowa = subtarget "3dnowa" (\ _ecx _edx -> False) - -aes :: Ext.Subtarget -aes = subtarget "aesni" (\ _ecx _edx -> False) - -sse4a :: Ext.Subtarget -sse4a = subtarget "sse4a" (\ _ecx _edx -> False) diff --git a/x86/none/LLVM/Extra/ExtensionCheck/X86.hs b/x86/none/LLVM/Extra/ExtensionCheck/X86.hs deleted file mode 100644 --- a/x86/none/LLVM/Extra/ExtensionCheck/X86.hs +++ /dev/null @@ -1,55 +0,0 @@ -module LLVM.Extra.ExtensionCheck.X86 ( - sse1, sse2, sse3, ssse3, sse41, sse42, avx, avx2, avx512, - fma, amd3dnow, amd3dnowa, aes, sse4a, - ) where - -import qualified LLVM.Extra.Extension as Ext - -subtarget :: String -> Bool -> Ext.Subtarget -subtarget name q = - Ext.Subtarget "x86" name (return q) - - --- * target specific extensions - -sse1 :: Ext.Subtarget -sse1 = subtarget "sse" False - -sse2 :: Ext.Subtarget -sse2 = subtarget "sse2" False - -sse3 :: Ext.Subtarget -sse3 = subtarget "sse3" False - -ssse3 :: Ext.Subtarget -ssse3 = subtarget "ssse3" False - -sse41 :: Ext.Subtarget -sse41 = subtarget "sse41" False - -sse42 :: Ext.Subtarget -sse42 = subtarget "sse42" False - -avx :: Ext.Subtarget -avx = subtarget "avx" False - -avx2 :: Ext.Subtarget -avx2 = subtarget "avx2" False - -avx512 :: Ext.Subtarget -avx512 = subtarget "avx512" False - -fma :: Ext.Subtarget -fma = subtarget "fma" False - -amd3dnow :: Ext.Subtarget -amd3dnow = subtarget "3dnow" False - -amd3dnowa :: Ext.Subtarget -amd3dnowa = subtarget "3dnowa" False - -aes :: Ext.Subtarget -aes = subtarget "aesni" False - -sse4a :: Ext.Subtarget -sse4a = subtarget "sse4a" False