diff options
author | Craig Topper <craig.topper@gmail.com> | 2011-11-22 01:57:35 +0000 |
---|---|---|
committer | Craig Topper <craig.topper@gmail.com> | 2011-11-22 01:57:35 +0000 |
commit | f7de577a08a705970f0fd8f3c1bb40f7040e4476 (patch) | |
tree | 07c02500aa979a61dc173746394c1db53cabe477 /lib/Target/X86/Utils | |
parent | c0d82857e095de15b31b0d654856aebea60dd339 (diff) | |
download | external_llvm-f7de577a08a705970f0fd8f3c1bb40f7040e4476.zip external_llvm-f7de577a08a705970f0fd8f3c1bb40f7040e4476.tar.gz external_llvm-f7de577a08a705970f0fd8f3c1bb40f7040e4476.tar.bz2 |
Fix shuffle decoding logic to handle UNPCKLPS/UNPCKLPD on 256-bit vectors correctly. Add support for decoding UNPCKHPS/UNPCKHPD for AVX 128-bit and 256-bit forms.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145055 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'lib/Target/X86/Utils')
-rw-r--r-- | lib/Target/X86/Utils/X86ShuffleDecode.cpp | 45 | ||||
-rw-r--r-- | lib/Target/X86/Utils/X86ShuffleDecode.h | 15 |
2 files changed, 42 insertions, 18 deletions
diff --git a/lib/Target/X86/Utils/X86ShuffleDecode.cpp b/lib/Target/X86/Utils/X86ShuffleDecode.cpp index aeb3309..8acd3c3 100644 --- a/lib/Target/X86/Utils/X86ShuffleDecode.cpp +++ b/lib/Target/X86/Utils/X86ShuffleDecode.cpp @@ -142,11 +142,32 @@ void DecodeSHUFPSMask(unsigned NElts, unsigned Imm, } } -void DecodeUNPCKHPMask(unsigned NElts, - SmallVectorImpl<unsigned> &ShuffleMask) { - for (unsigned i = 0; i != NElts/2; ++i) { - ShuffleMask.push_back(i+NElts/2); // Reads from dest - ShuffleMask.push_back(i+NElts+NElts/2); // Reads from src +void DecodeUNPCKHPSMask(unsigned NElts, + SmallVectorImpl<unsigned> &ShuffleMask) { + DecodeUNPCKHPMask(MVT::getVectorVT(MVT::i32, NElts), ShuffleMask); +} + +void DecodeUNPCKHPDMask(unsigned NElts, + SmallVectorImpl<unsigned> &ShuffleMask) { + DecodeUNPCKHPMask(MVT::getVectorVT(MVT::i64, NElts), ShuffleMask); +} + +void DecodeUNPCKHPMask(EVT VT, SmallVectorImpl<unsigned> &ShuffleMask) { + unsigned NumElts = VT.getVectorNumElements(); + + // Handle 128 and 256-bit vector lengths. AVX defines UNPCK* to operate + // independently on 128-bit lanes. + unsigned NumLanes = VT.getSizeInBits() / 128; + if (NumLanes == 0 ) NumLanes = 1; // Handle MMX + unsigned NumLaneElts = NumElts / NumLanes; + + for (unsigned s = 0; s < NumLanes; ++s) { + unsigned Start = s * NumLaneElts + NumLaneElts/2; + unsigned End = s * NumLaneElts + NumLaneElts; + for (unsigned i = Start; i != End; ++i) { + ShuffleMask.push_back(i); // Reads from dest/src1 + ShuffleMask.push_back(i+NumElts); // Reads from src/src2 + } } } @@ -163,8 +184,7 @@ void DecodeUNPCKLPDMask(unsigned NElts, /// DecodeUNPCKLPMask - This decodes the shuffle masks for unpcklps/unpcklpd /// etc. VT indicates the type of the vector allowing it to handle different /// datatypes and vector widths. -void DecodeUNPCKLPMask(EVT VT, - SmallVectorImpl<unsigned> &ShuffleMask) { +void DecodeUNPCKLPMask(EVT VT, SmallVectorImpl<unsigned> &ShuffleMask) { unsigned NumElts = VT.getVectorNumElements(); // Handle 128 and 256-bit vector lengths. AVX defines UNPCK* to operate @@ -173,16 +193,13 @@ void DecodeUNPCKLPMask(EVT VT, if (NumLanes == 0 ) NumLanes = 1; // Handle MMX unsigned NumLaneElts = NumElts / NumLanes; - unsigned Start = 0; - unsigned End = NumLaneElts / 2; for (unsigned s = 0; s < NumLanes; ++s) { + unsigned Start = s * NumLaneElts; + unsigned End = s * NumLaneElts + NumLaneElts/2; for (unsigned i = Start; i != End; ++i) { - ShuffleMask.push_back(i); // Reads from dest/src1 - ShuffleMask.push_back(i+NumLaneElts); // Reads from src/src2 + ShuffleMask.push_back(i); // Reads from dest/src1 + ShuffleMask.push_back(i+NumElts); // Reads from src/src2 } - // Process the next 128 bits. - Start += NumLaneElts; - End += NumLaneElts; } } diff --git a/lib/Target/X86/Utils/X86ShuffleDecode.h b/lib/Target/X86/Utils/X86ShuffleDecode.h index 58193e6..d715016 100644 --- a/lib/Target/X86/Utils/X86ShuffleDecode.h +++ b/lib/Target/X86/Utils/X86ShuffleDecode.h @@ -67,8 +67,16 @@ void DecodePUNPCKHMask(unsigned NElts, void DecodeSHUFPSMask(unsigned NElts, unsigned Imm, SmallVectorImpl<unsigned> &ShuffleMask); -void DecodeUNPCKHPMask(unsigned NElts, - SmallVectorImpl<unsigned> &ShuffleMask); +void DecodeUNPCKHPSMask(unsigned NElts, + SmallVectorImpl<unsigned> &ShuffleMask); + +void DecodeUNPCKHPDMask(unsigned NElts, + SmallVectorImpl<unsigned> &ShuffleMask); + +/// DecodeUNPCKHPMask - This decodes the shuffle masks for unpckhps/unpckhpd +/// etc. VT indicates the type of the vector allowing it to handle different +/// datatypes and vector widths. +void DecodeUNPCKHPMask(EVT VT, SmallVectorImpl<unsigned> &ShuffleMask); void DecodeUNPCKLPSMask(unsigned NElts, SmallVectorImpl<unsigned> &ShuffleMask); @@ -79,8 +87,7 @@ void DecodeUNPCKLPDMask(unsigned NElts, /// DecodeUNPCKLPMask - This decodes the shuffle masks for unpcklps/unpcklpd /// etc. VT indicates the type of the vector allowing it to handle different /// datatypes and vector widths. -void DecodeUNPCKLPMask(EVT VT, - SmallVectorImpl<unsigned> &ShuffleMask); +void DecodeUNPCKLPMask(EVT VT, SmallVectorImpl<unsigned> &ShuffleMask); // DecodeVPERMILPSMask - Decodes VPERMILPS permutes for any 128-bit 32-bit |