diff --git a/README.md b/README.md
--- a/README.md
+++ b/README.md
@@ -23,18 +23,39 @@
Here are the throughput numbers with some different selections of data and parity shards. For reference each shard is 1MB random data, and 1 CPU core is used for encoding.
-| Data | Parity | Parity | SSSE3 MB/s | AVX2 MB/s |
-|------|--------|--------|------------|-----------|
-| 5 | 2 | 40% | 3641,66 | 3987,24 |
-| 10 | 2 | 20% | 3951,01 | 4444,44 |
-| 10 | 4 | 40% | 1821,16 | 1927,90 |
-| 50 | 20 | 40% | 398,09 | 431,78 |
+
+
+
+ | Data |
+ Parity |
+ Parity |
+ SSSE3 MB/s |
+ AVX2 MB/s |
+
+
+
+ | 5 | 2 | 40% | 3641,66 | 3987,24 |
+ | 10 | 2 | 20% | 3951,01 | 4444,44 |
+ | 10 | 4 | 40% | 1821,16 | 1927,90 |
+ | 50 | 20 | 40% | 398,09 | 431,78 |
+
+
Example of performance on Intel(R) Core(TM) i7-4600U CPU @ 3.30GHz - 2 physical cores, 4 logical cores (note: `/proc/cpuinfo` mentions 2.10GHz only). The example uses 10 blocks with 16MB data each and 4 parity blocks.
-| Threads | SSSE3 MB/s | AVX2 MB/s | Speed |
-|---------|------------|-----------|-------|
-| 1 | 1551,89 | 1588,88 | 100% |
+
+
+
+ | Threads |
+ SSSE3 MB/s |
+ AVX2 MB/s |
+ Speed |
+
+
+
+ | 1 | 1551,89 | 1588,88 | 100% |
+
+
# Links
* [Backblaze Open Sources Reed-Solomon Erasure Coding Source Code](https://www.backblaze.com/blog/reed-solomon/).
diff --git a/bench/Main.hs b/bench/Main.hs
--- a/bench/Main.hs
+++ b/bench/Main.hs
@@ -10,7 +10,6 @@
import Control.Monad.ST (ST)
#ifdef SIMD
-import Data.Bits ((.&.), shiftL)
import Data.Maybe (catMaybes)
#endif
import Data.Word (Word8)
@@ -19,13 +18,6 @@
import Foreign.C (CSize(..))
#endif
-#ifdef SIMD
-import System.Cpuid (cpuid)
-# if RS_HAVE_AVX2
-import System.Cpuid (cpuidWithIndex)
-# endif
-#endif
-
import Criterion.Main
import qualified Data.Vector.Generic as V
@@ -34,27 +26,23 @@
#ifdef SIMD
import qualified Data.Vector.Generic.Sized as S
+import Data.ReedSolomon (SIMDInstructions(..))
#endif
+import qualified Data.ReedSolomon as RS
import qualified Data.ReedSolomon.Galois.NoAsm as NoAsm
#ifdef SIMD
import qualified Data.ReedSolomon.Galois.Amd64 as Amd64
#endif
main :: IO ()
-#ifdef SIMD
main = do
- (_, _, ecx1, edx1) <- cpuid 1
- let avx = ecx1 .&. (1 `shiftL` 28) /= 0
- ssse3 = ecx1 .&. (1 `shiftL` 9) /= 0
- sse2 = edx1 .&. (1 `shiftL` 26) /= 0
- dependOn b c = if b then Just c else Nothing
-# if RS_HAVE_AVX2
- (_, ebx7, _, _) <- cpuidWithIndex 7 0
- let avx2 = ebx7 .&. (1 `shiftL` 5) /= 0
-# endif
-#else
-main =
+ level <- RS.simdInstructions
+ putStrLn $ "Native instructions: " ++ maybe "None" show level
+
+#ifdef SIMD
+ let dependOn l prop = maybe Nothing (\lvl -> if l <= lvl then Just prop else Nothing ) level
#endif
+
defaultMain [
bgroup "Galois/galMulSlice/1048576" [
bench "NoAsm" $ whnf (benchGalMulSlice NoAsm.galMulSlice 177) v1048576
@@ -66,11 +54,11 @@
, bgroup "reedsolomon_gal_mul" $ catMaybes [
Just $ bench "Native" $ whnf (benchRGM c_reedsolomon_gal_mul) v1048576
#if RS_HAVE_AVX2
- , dependOn avx2 $ bench "AVX2" $ whnf (benchRGM c_reedsolomon_gal_mul_avx2) v1048576
+ , dependOn AVX2 $ bench "AVX2" $ whnf (benchRGM c_reedsolomon_gal_mul_avx2) v1048576
#endif
- , dependOn avx $ bench "AVX" $ whnf (benchRGM c_reedsolomon_gal_mul_avx) v1048576
- , dependOn ssse3 $ bench "SSSE3" $ whnf (benchRGM c_reedsolomon_gal_mul_ssse3) v1048576
- , dependOn sse2 $ bench "SSE2" $ whnf (benchRGM c_reedsolomon_gal_mul_sse2) v1048576
+ , dependOn AVX $ bench "AVX" $ whnf (benchRGM c_reedsolomon_gal_mul_avx) v1048576
+ , dependOn SSSE3 $ bench "SSSE3" $ whnf (benchRGM c_reedsolomon_gal_mul_ssse3) v1048576
+ , dependOn SSE2 $ bench "SSE2" $ whnf (benchRGM c_reedsolomon_gal_mul_sse2) v1048576
, Just $ bench "Generic" $ whnf (benchRGM c_reedsolomon_gal_mul_generic) v1048576
]
#endif
diff --git a/cbits/configure b/cbits/configure
--- a/cbits/configure
+++ b/cbits/configure
@@ -1,6 +1,6 @@
#! /bin/sh
# Guess values for system-dependent variables and create Makefiles.
-# Generated by GNU Autoconf 2.69 for reedsolomon 0.0.1.2.
+# Generated by GNU Autoconf 2.69 for reedsolomon 0.0.2.0.
#
# Report bugs to .
#
@@ -590,8 +590,8 @@
# Identity of this package.
PACKAGE_NAME='reedsolomon'
PACKAGE_TARNAME='reedsolomon'
-PACKAGE_VERSION='0.0.1.2'
-PACKAGE_STRING='reedsolomon 0.0.1.2'
+PACKAGE_VERSION='0.0.2.0'
+PACKAGE_STRING='reedsolomon 0.0.2.0'
PACKAGE_BUGREPORT='https://github.com/NicolasT/reedsolomon/issues'
PACKAGE_URL='https://github.com/NicolasT/reedsolomon'
@@ -1316,7 +1316,7 @@
# Omit some internal or obsolete options to make the list less imposing.
# This message is too long to be a string in the A/UX 3.1 sh.
cat <<_ACEOF
-\`configure' configures reedsolomon 0.0.1.2 to adapt to many kinds of systems.
+\`configure' configures reedsolomon 0.0.2.0 to adapt to many kinds of systems.
Usage: $0 [OPTION]... [VAR=VALUE]...
@@ -1386,7 +1386,7 @@
if test -n "$ac_init_help"; then
case $ac_init_help in
- short | recursive ) echo "Configuration of reedsolomon 0.0.1.2:";;
+ short | recursive ) echo "Configuration of reedsolomon 0.0.2.0:";;
esac
cat <<\_ACEOF
@@ -1495,7 +1495,7 @@
test -n "$ac_init_help" && exit $ac_status
if $ac_init_version; then
cat <<\_ACEOF
-reedsolomon configure 0.0.1.2
+reedsolomon configure 0.0.2.0
generated by GNU Autoconf 2.69
Copyright (C) 2012 Free Software Foundation, Inc.
@@ -1972,7 +1972,7 @@
This file contains any messages produced by compilers while
running configure, to aid debugging if configure makes a mistake.
-It was created by reedsolomon $as_me 0.0.1.2, which was
+It was created by reedsolomon $as_me 0.0.2.0, which was
generated by GNU Autoconf 2.69. Invocation command line was
$ $0 $@
@@ -2840,7 +2840,7 @@
# Define the identity of the package.
PACKAGE='reedsolomon'
- VERSION='0.0.1.2'
+ VERSION='0.0.2.0'
cat >>confdefs.h <<_ACEOF
@@ -13290,7 +13290,7 @@
# report actual input values of CONFIG_FILES etc. instead of their
# values after options handling.
ac_log="
-This file was extended by reedsolomon $as_me 0.0.1.2, which was
+This file was extended by reedsolomon $as_me 0.0.2.0, which was
generated by GNU Autoconf 2.69. Invocation command line was
CONFIG_FILES = $CONFIG_FILES
@@ -13357,7 +13357,7 @@
cat >>$CONFIG_STATUS <<_ACEOF || ac_write_fail=1
ac_cs_config="`$as_echo "$ac_configure_args" | sed 's/^ //; s/[\\""\`\$]/\\\\&/g'`"
ac_cs_version="\\
-reedsolomon config.status 0.0.1.2
+reedsolomon config.status 0.0.2.0
configured by $0, generated by GNU Autoconf 2.69,
with options \\"\$ac_cs_config\\"
diff --git a/cbits/reedsolomon.h b/cbits/reedsolomon.h
--- a/cbits/reedsolomon.h
+++ b/cbits/reedsolomon.h
@@ -15,6 +15,8 @@
PROTO(reedsolomon_gal_mul_avx2);
PROTO(reedsolomon_gal_mul_xor_avx2);
+PROTO(reedsolomon_gal_mul_avx);
+PROTO(reedsolomon_gal_mul_xor_avx);
PROTO(reedsolomon_gal_mul_ssse3);
PROTO(reedsolomon_gal_mul_xor_ssse3);
PROTO(reedsolomon_gal_mul_sse2);
@@ -24,3 +26,13 @@
PROTO(reedsolomon_gal_mul);
PROTO(reedsolomon_gal_mul_xor);
+
+typedef enum {
+ REEDSOLOMON_CPU_GENERIC = 0,
+ REEDSOLOMON_CPU_SSE2 = 1,
+ REEDSOLOMON_CPU_SSSE3 = 2,
+ REEDSOLOMON_CPU_AVX = 3,
+ REEDSOLOMON_CPU_AVX2 = 4,
+} reedsolomon_cpu_support;
+
+reedsolomon_cpu_support reedsolomon_determine_cpu_support(void);
diff --git a/cbits/reedsolomon_dispatch.c b/cbits/reedsolomon_dispatch.c
--- a/cbits/reedsolomon_dispatch.c
+++ b/cbits/reedsolomon_dispatch.c
@@ -38,57 +38,82 @@
return 1;
}
+struct cpuid_registers {
+ unsigned int eax, ebx, ecx, edx;
+};
+
+reedsolomon_cpu_support reedsolomon_determine_cpu_support(void) {
+ struct cpuid_registers cpuid1 = { 0, 0, 0, 0 };
+ unsigned int rc1 = 0;
+ reedsolomon_cpu_support result = REEDSOLOMON_CPU_GENERIC;
+
+ rc1 = __get_cpuid(1, &cpuid1.eax, &cpuid1.ebx, &cpuid1.ecx, &cpuid1.edx);
+
+ if(rc1 == 0) {
+ result = REEDSOLOMON_CPU_GENERIC;
+ }
+ else {
#if RS_HAVE_AVX2
-# define IFUNC_AVX2(n, result) \
- do { \
- int avx2_rc = 0; \
- struct { \
- unsigned int eax, ebx, ecx, edx; \
- } cpuid7 = { 0, 0, 0, 0 }; \
- \
- avx2_rc = __get_cpuid_count(7, 0, &cpuid7.eax, &cpuid7.ebx, &cpuid7.ecx, &cpuid7.edx); \
- \
- if(avx2_rc != 0 && (cpuid7.ebx & bit_AVX2) != 0) { \
- LOG("reedsolomon: using " #n "_avx2"); \
- result = n ## _avx2; \
- } \
- } while(0)
+ struct cpuid_registers cpuid7 = { 0, 0, 0, 0 };
+ unsigned int rc7 = 0;
+
+ rc7 = __get_cpuid_count(7, 0, &cpuid7.eax, &cpuid7.ebx, &cpuid7.ecx, &cpuid7.edx);
+ if(rc7 != 0 && (cpuid7.ebx & bit_AVX2) != 0) {
+ result = REEDSOLOMON_CPU_AVX2;
+ }
+ else {
#else
-# define IFUNC_AVX2(n, result)
+ if(1) {
#endif
+ if((cpuid1.ecx & bit_AVX) != 0) {
+ result = REEDSOLOMON_CPU_AVX;
+ }
+ else if((cpuid1.ecx & bit_SSSE3) != 0) {
+ result = REEDSOLOMON_CPU_SSSE3;
+ }
+ else if((cpuid1.edx & bit_SSE2) != 0) {
+ result = REEDSOLOMON_CPU_SSE2;
+ }
+ else {
+ result = REEDSOLOMON_CPU_GENERIC;
+ }
+ }
+ }
-#define IFUNC(n, proto) \
- static proto n ## _ifunc(void) { \
- struct { \
- unsigned int eax, ebx, ecx, edx; \
- } cpuid1 = { 0, 0, 0, 0 }; \
- int rc = 0; \
- proto func = NULL; \
- \
- rc = __get_cpuid(1, &cpuid1.eax, &cpuid1.ebx, &cpuid1.ecx, &cpuid1.edx); \
- if(rc == 0) { \
- LOG("reedsolomon: cpuid failed, using " #n "_generic"); \
- func = n ## _generic; \
- } \
- else { \
- IFUNC_AVX2(n, func); \
- if(func == NULL) { \
- if((cpuid1.ecx & bit_SSSE3) != 0) { \
- LOG("reedsolomon: using " #n "_ssse3"); \
- func = n ## _ssse3; \
- } \
- else if((cpuid1.edx & bit_SSE2) != 0) { \
- LOG("reedsolomon: using " #n "_sse2"); \
- func = n ## _sse2; \
- } \
- else { \
- LOG("reedsolomon: using " #n "_generic"); \
- func = n ## _generic; \
- } \
- } \
- } \
- \
- return func; \
+ return result;
+}
+
+#define CASE(n, lower, upper) \
+ case REEDSOLOMON_CPU_ ## upper: { \
+ LOG("reedsolomon: using " #n "_" #lower); \
+ result = n ## _ ## lower; \
+ } break
+
+#if RS_HAVE_AVX2
+# define MAYBE_AVX2(n) \
+ CASE(n, avx2, AVX2)
+#else
+# define MAYBE_AVX2(n)
+#endif
+
+#define IFUNC(n, proto) \
+ static proto n ## _ifunc(void) { \
+ reedsolomon_cpu_support level = reedsolomon_determine_cpu_support(); \
+ proto result = n ## _generic; \
+ \
+ switch(level) { \
+ MAYBE_AVX2(n); \
+ CASE(n, avx, AVX); \
+ CASE(n, ssse3, SSSE3); \
+ CASE(n, sse2, SSE2); \
+ CASE(n, generic, GENERIC); \
+ default: { \
+ LOG("reedsolomon: using " #n "_generic"); \
+ result = n ## _generic; \
+ } \
+ } \
+ \
+ return result; \
}
typedef PROTO_RETURN(*gal_mul_proto)(PROTO_ARGS);
diff --git a/examples/simple-bench.hs b/examples/simple-bench.hs
--- a/examples/simple-bench.hs
+++ b/examples/simple-bench.hs
@@ -45,6 +45,9 @@
let [dataShards, parityShards, dataSize] = fmap read args
+ instructionSet <- RS.simdInstructions
+ printf "Using CPU instructions: %s\n" (maybe "None" show instructionSet)
+
putStrLn "generating encoder..."
encoder <- RS.new dataShards parityShards
diff --git a/reedsolomon.cabal b/reedsolomon.cabal
--- a/reedsolomon.cabal
+++ b/reedsolomon.cabal
@@ -1,5 +1,5 @@
Name: reedsolomon
-Version: 0.0.1.2
+Version: 0.0.2.0
Synopsis: Reed-Solomon Erasure Coding in Haskell
Description: Please see README.md
Homepage: http://github.com/NicolasT/reedsolomon
@@ -190,7 +190,6 @@
, exceptions
, bytestring
, profunctors
- , cpu >= 0.1 && < 0.2
, tasty >= 0.10 && < 0.12
, tasty-hunit > 0.9 && < 0.10
, tasty-quickcheck >= 0.8 && < 0.9
@@ -215,14 +214,21 @@
Hs-Source-Dirs: bench
, src
Main-Is: Main.hs
- Other-Modules: Data.ReedSolomon.Galois.GenTables
+ Other-Modules: Data.ReedSolomon
+ , Data.ReedSolomon.Galois
+ , Data.ReedSolomon.Galois.GenTables
, Data.ReedSolomon.Galois.NoAsm
+ , Data.ReedSolomon.Matrix
+ , Data.Vector.Generic.Compat
+ , Data.Vector.Generic.Exceptions
+ , Data.Vector.Generic.Lifted
, Data.Vector.Generic.Sized
Build-Depends: base
+ , mtl
, vector
, loop
, primitive
- , cpu >= 0.1 && < 0.2
+ , exceptions
, criterion >= 1.1 && < 1.2
, reedsolomon
Default-Language: Haskell2010
diff --git a/src/Data/ReedSolomon.lhs b/src/Data/ReedSolomon.lhs
--- a/src/Data/ReedSolomon.lhs
+++ b/src/Data/ReedSolomon.lhs
@@ -33,6 +33,9 @@
> -- * Exceptions
> , ShardType(..)
> , ValueError(..)
+> -- * Determining runtime SIMD instruction support
+> , SIMDInstructions(..)
+> , simdInstructions
> ) where
>
> import Prelude hiding (any)
@@ -45,6 +48,10 @@
> import Data.Typeable (Typeable)
> import Data.Word
>
+#if SIMD
+> import Foreign.C (CInt(..))
+#endif
+>
> import Control.Monad.Catch (MonadThrow, throwM)
>
> import qualified Data.Vector as V (Vector, MVector)
@@ -841,3 +848,26 @@
> | V.length shards < rsDataShards r = throwM (InvalidNumberOfShards DataShard $ V.length shards)
> | V.sum (V.map V.length shards) < outSize = throwM InvalidDataSize
> | otherwise = return $ V.take outSize $ V.concat $ V.toList shards
+
+> -- | Enumeration of SIMD instruction sets.
+> data SIMDInstructions = Generic
+> | SSE2
+> | SSSE3
+> | AVX
+> | AVX2
+> deriving (Show, Eq, Ord, Enum, Bounded)
+>
+> -- | Retrieve the SIMD instruction set used by 'native' Galois field operations.
+> --
+> -- Returns 'Nothing' when the library is compiled without SIMD support,
+> -- otherwise 'Just' the runtime-determined SIMD instruction set used for
+> -- optimized Galois field calculations.
+> simdInstructions :: IO (Maybe SIMDInstructions)
+> simdInstructions =
+#if !SIMD
+> return Nothing
+#else
+> (Just . toEnum . fromIntegral) `fmap` c_reedsolomon_determine_cpu_support
+>
+> foreign import ccall unsafe "reedsolomon_determine_cpu_support" c_reedsolomon_determine_cpu_support :: IO CInt
+#endif
diff --git a/test/Galois.lhs b/test/Galois.lhs
--- a/test/Galois.lhs
+++ b/test/Galois.lhs
@@ -12,7 +12,6 @@
>
> import Control.Monad (foldM, void, when)
#ifdef SIMD
-> import Data.Bits ((.&.), shiftL)
> import Data.Maybe (catMaybes)
#endif
> import Data.Word (Word8)
@@ -28,13 +27,6 @@
> import qualified Data.Vector.Generic.Mutable as MV
> import qualified Data.Vector.Storable as SV
>
-#ifdef SIMD
-> import System.Cpuid (cpuid)
-# if RS_HAVE_AVX2
-> import System.Cpuid (cpuidWithIndex)
-# endif
-#endif
->
> import Test.Tasty (TestTree, testGroup)
> import Test.Tasty.HUnit (Assertion, (@?=), testCase)
#ifdef SIMD
@@ -47,6 +39,8 @@
>
#ifdef SIMD
> import qualified Data.Vector.Generic.Sized as S
+> import qualified Data.ReedSolomon as RS
+> import Data.ReedSolomon (SIMDInstructions(..))
#endif
> import Data.ReedSolomon.Galois
> import qualified Data.ReedSolomon.Galois.NoAsm as NoAsm
@@ -412,14 +406,14 @@
> , testGroup "cbits" [
> testGroup "reedsolomon_gal_mul" $ catMaybes [
#if RS_HAVE_AVX2
-> dependOn avx2 $ testProperty "native/avx2" $
+> dependOn AVX2 $ testProperty "native/avx2" $
> reedsolomonGalMulEquivalence c_rgm c_rgm_avx2,
#endif
-> dependOn avx $ testProperty "native/avx" $
+> dependOn AVX $ testProperty "native/avx" $
> reedsolomonGalMulEquivalence c_rgm c_rgm_avx
-> , dependOn ssse3 $ testProperty "native/ssse3" $
+> , dependOn SSSE3 $ testProperty "native/ssse3" $
> reedsolomonGalMulEquivalence c_rgm c_rgm_ssse3
-> , dependOn sse2 $ testProperty "native/sse2" $
+> , dependOn SSE2 $ testProperty "native/sse2" $
> reedsolomonGalMulEquivalence c_rgm c_rgm_sse2
> , Just $ testProperty "native/generic" $
> reedsolomonGalMulEquivalence c_rgm c_rgm_generic
@@ -428,15 +422,8 @@
> ]
#endif
> ]
-#ifdef SIMD
+#if SIMD
> where
-> (_, _, ecx1, edx1) = unsafePerformIO $ cpuid 1
-#if RS_HAVE_AVX2
-> (_, ebx7, _, _) = unsafePerformIO $ cpuidWithIndex 7 0
-> avx2 = ebx7 .&. (1 `shiftL` 5) /= 0
-#endif
-> avx = ecx1 .&. (1 `shiftL` 28) /= 0
-> ssse3 = ecx1 .&. (1 `shiftL` 9) /= 0
-> sse2 = edx1 .&. (1 `shiftL` 26) /= 0
-> dependOn b c = if b then Just c else Nothing
+> level = unsafePerformIO RS.simdInstructions
+> dependOn l prop = maybe Nothing (\lvl -> if l <= lvl then Just prop else Nothing) level
#endif