aboutsummaryrefslogtreecommitdiffstats
path: root/lib/Target/X86/Utils
diff options
context:
space:
mode:
Diffstat (limited to 'lib/Target/X86/Utils')
-rw-r--r--lib/Target/X86/Utils/CMakeLists.txt5
-rw-r--r--lib/Target/X86/Utils/LLVMBuild.txt23
-rw-r--r--lib/Target/X86/Utils/X86ShuffleDecode.cpp149
-rw-r--r--lib/Target/X86/Utils/X86ShuffleDecode.h56
4 files changed, 91 insertions, 142 deletions
diff --git a/lib/Target/X86/Utils/CMakeLists.txt b/lib/Target/X86/Utils/CMakeLists.txt
index caffd8b..2e72c34 100644
--- a/lib/Target/X86/Utils/CMakeLists.txt
+++ b/lib/Target/X86/Utils/CMakeLists.txt
@@ -4,9 +4,4 @@ add_llvm_library(LLVMX86Utils
X86ShuffleDecode.cpp
)
-add_llvm_library_dependencies(LLVMX86Utils
- LLVMCore
- LLVMSupport
- )
-
add_dependencies(LLVMX86Utils X86CommonTableGen)
diff --git a/lib/Target/X86/Utils/LLVMBuild.txt b/lib/Target/X86/Utils/LLVMBuild.txt
new file mode 100644
index 0000000..de0a30f
--- /dev/null
+++ b/lib/Target/X86/Utils/LLVMBuild.txt
@@ -0,0 +1,23 @@
+;===- ./lib/Target/X86/Utils/LLVMBuild.txt ---------------------*- Conf -*--===;
+;
+; The LLVM Compiler Infrastructure
+;
+; This file is distributed under the University of Illinois Open Source
+; License. See LICENSE.TXT for details.
+;
+;===------------------------------------------------------------------------===;
+;
+; This is an LLVMBuild description file for the components in this subdirectory.
+;
+; For more information on the LLVMBuild system, please see:
+;
+; http://llvm.org/docs/LLVMBuild.html
+;
+;===------------------------------------------------------------------------===;
+
+[component_0]
+type = Library
+name = X86Utils
+parent = X86
+required_libraries = Core Support
+add_to_library_groups = X86
diff --git a/lib/Target/X86/Utils/X86ShuffleDecode.cpp b/lib/Target/X86/Utils/X86ShuffleDecode.cpp
index aeb3309..e7631b6 100644
--- a/lib/Target/X86/Utils/X86ShuffleDecode.cpp
+++ b/lib/Target/X86/Utils/X86ShuffleDecode.cpp
@@ -95,76 +95,53 @@ void DecodePSHUFLWMask(unsigned Imm,
ShuffleMask.push_back(7);
}
-void DecodePUNPCKLBWMask(unsigned NElts,
- SmallVectorImpl<unsigned> &ShuffleMask) {
- DecodeUNPCKLPMask(MVT::getVectorVT(MVT::i8, NElts), ShuffleMask);
-}
-
-void DecodePUNPCKLWDMask(unsigned NElts,
- SmallVectorImpl<unsigned> &ShuffleMask) {
- DecodeUNPCKLPMask(MVT::getVectorVT(MVT::i16, NElts), ShuffleMask);
-}
-
-void DecodePUNPCKLDQMask(unsigned NElts,
- SmallVectorImpl<unsigned> &ShuffleMask) {
- DecodeUNPCKLPMask(MVT::getVectorVT(MVT::i32, NElts), ShuffleMask);
-}
-
-void DecodePUNPCKLQDQMask(unsigned NElts,
- SmallVectorImpl<unsigned> &ShuffleMask) {
- DecodeUNPCKLPMask(MVT::getVectorVT(MVT::i64, NElts), ShuffleMask);
-}
-
-void DecodePUNPCKLMask(EVT VT,
- SmallVectorImpl<unsigned> &ShuffleMask) {
- DecodeUNPCKLPMask(VT, ShuffleMask);
-}
+void DecodeSHUFPMask(EVT VT, unsigned Imm,
+ SmallVectorImpl<unsigned> &ShuffleMask) {
+ unsigned NumElts = VT.getVectorNumElements();
-void DecodePUNPCKHMask(unsigned NElts,
- SmallVectorImpl<unsigned> &ShuffleMask) {
- for (unsigned i = 0; i != NElts/2; ++i) {
- ShuffleMask.push_back(i+NElts/2);
- ShuffleMask.push_back(i+NElts+NElts/2);
- }
-}
+ unsigned NumLanes = VT.getSizeInBits() / 128;
+ unsigned NumLaneElts = NumElts / NumLanes;
-void DecodeSHUFPSMask(unsigned NElts, unsigned Imm,
- SmallVectorImpl<unsigned> &ShuffleMask) {
- // Part that reads from dest.
- for (unsigned i = 0; i != NElts/2; ++i) {
- ShuffleMask.push_back(Imm % NElts);
- Imm /= NElts;
- }
- // Part that reads from src.
- for (unsigned i = 0; i != NElts/2; ++i) {
- ShuffleMask.push_back(Imm % NElts + NElts);
- Imm /= NElts;
+ int NewImm = Imm;
+ for (unsigned l = 0; l < NumLanes; ++l) {
+ unsigned LaneStart = l * NumLaneElts;
+ // Part that reads from dest.
+ for (unsigned i = 0; i != NumLaneElts/2; ++i) {
+ ShuffleMask.push_back(NewImm % NumLaneElts + LaneStart);
+ NewImm /= NumLaneElts;
+ }
+ // Part that reads from src.
+ for (unsigned i = 0; i != NumLaneElts/2; ++i) {
+ ShuffleMask.push_back(NewImm % NumLaneElts + NumElts + LaneStart);
+ NewImm /= NumLaneElts;
+ }
+ if (NumLaneElts == 4) NewImm = Imm; // reload 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 DecodeUNPCKHMask(EVT VT, SmallVectorImpl<unsigned> &ShuffleMask) {
+ unsigned NumElts = VT.getVectorNumElements();
-void DecodeUNPCKLPSMask(unsigned NElts,
- SmallVectorImpl<unsigned> &ShuffleMask) {
- DecodeUNPCKLPMask(MVT::getVectorVT(MVT::i32, NElts), ShuffleMask);
-}
+ // 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;
-void DecodeUNPCKLPDMask(unsigned NElts,
- SmallVectorImpl<unsigned> &ShuffleMask) {
- DecodeUNPCKLPMask(MVT::getVectorVT(MVT::i64, NElts), ShuffleMask);
+ 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
+ }
+ }
}
-/// DecodeUNPCKLPMask - This decodes the shuffle masks for unpcklps/unpcklpd
+/// DecodeUNPCKLMask - 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 DecodeUNPCKLMask(EVT VT, SmallVectorImpl<unsigned> &ShuffleMask) {
unsigned NumElts = VT.getVectorNumElements();
// Handle 128 and 256-bit vector lengths. AVX defines UNPCK* to operate
@@ -173,49 +150,33 @@ 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;
}
}
-// DecodeVPERMILPSMask - Decodes VPERMILPS permutes for any 128-bit 32-bit
-// elements. For 256-bit vectors, it's considered as two 128 lanes, the
-// referenced elements can't cross lanes and the mask of the first lane must
-// be the same of the second.
-void DecodeVPERMILPSMask(unsigned NumElts, unsigned Imm,
- SmallVectorImpl<unsigned> &ShuffleMask) {
- unsigned NumLanes = (NumElts*32)/128;
- unsigned LaneSize = NumElts/NumLanes;
-
- for (unsigned l = 0; l != NumLanes; ++l) {
- for (unsigned i = 0; i != LaneSize; ++i) {
- unsigned Idx = (Imm >> (i*2)) & 0x3 ;
- ShuffleMask.push_back(Idx+(l*LaneSize));
- }
- }
-}
+// DecodeVPERMILPMask - Decodes VPERMILPS/ VPERMILPD permutes for any 128-bit
+// 32-bit or 64-bit elements. For 256-bit vectors, it's considered as two 128
+// lanes. For VPERMILPS, referenced elements can't cross lanes and the mask of
+// the first lane must be the same of the second.
+void DecodeVPERMILPMask(EVT VT, unsigned Imm,
+ SmallVectorImpl<unsigned> &ShuffleMask) {
+ unsigned NumElts = VT.getVectorNumElements();
-// DecodeVPERMILPDMask - Decodes VPERMILPD permutes for any 128-bit 64-bit
-// elements. For 256-bit vectors, it's considered as two 128 lanes, the
-// referenced elements can't cross lanes but the mask of the first lane can
-// be the different of the second (not like VPERMILPS).
-void DecodeVPERMILPDMask(unsigned NumElts, unsigned Imm,
- SmallVectorImpl<unsigned> &ShuffleMask) {
- unsigned NumLanes = (NumElts*64)/128;
- unsigned LaneSize = NumElts/NumLanes;
+ unsigned NumLanes = VT.getSizeInBits() / 128;
+ unsigned NumLaneElts = NumElts / NumLanes;
- for (unsigned l = 0; l < NumLanes; ++l) {
- for (unsigned i = l*LaneSize; i < LaneSize*(l+1); ++i) {
- unsigned Idx = (Imm >> i) & 0x1;
- ShuffleMask.push_back(Idx+(l*LaneSize));
+ for (unsigned l = 0; l != NumLanes; ++l) {
+ unsigned LaneStart = l*NumLaneElts;
+ for (unsigned i = 0; i != NumLaneElts; ++i) {
+ unsigned Idx = NumLaneElts == 4 ? (Imm >> (i*2)) & 0x3
+ : (Imm >> (i+LaneStart)) & 0x1;
+ ShuffleMask.push_back(Idx+LaneStart);
}
}
}
diff --git a/lib/Target/X86/Utils/X86ShuffleDecode.h b/lib/Target/X86/Utils/X86ShuffleDecode.h
index 58193e6..243728f 100644
--- a/lib/Target/X86/Utils/X86ShuffleDecode.h
+++ b/lib/Target/X86/Utils/X86ShuffleDecode.h
@@ -46,55 +46,25 @@ void DecodePSHUFHWMask(unsigned Imm,
void DecodePSHUFLWMask(unsigned Imm,
SmallVectorImpl<unsigned> &ShuffleMask);
-void DecodePUNPCKLBWMask(unsigned NElts,
- SmallVectorImpl<unsigned> &ShuffleMask);
-
-void DecodePUNPCKLWDMask(unsigned NElts,
- SmallVectorImpl<unsigned> &ShuffleMask);
-
-void DecodePUNPCKLDQMask(unsigned NElts,
- SmallVectorImpl<unsigned> &ShuffleMask);
-
-void DecodePUNPCKLQDQMask(unsigned NElts,
- SmallVectorImpl<unsigned> &ShuffleMask);
-
-void DecodePUNPCKLMask(EVT VT,
- SmallVectorImpl<unsigned> &ShuffleMask);
-
-void DecodePUNPCKHMask(unsigned NElts,
- SmallVectorImpl<unsigned> &ShuffleMask);
-
-void DecodeSHUFPSMask(unsigned NElts, unsigned Imm,
- SmallVectorImpl<unsigned> &ShuffleMask);
-
-void DecodeUNPCKHPMask(unsigned NElts,
- SmallVectorImpl<unsigned> &ShuffleMask);
-
-void DecodeUNPCKLPSMask(unsigned NElts,
- SmallVectorImpl<unsigned> &ShuffleMask);
-
-void DecodeUNPCKLPDMask(unsigned NElts,
- SmallVectorImpl<unsigned> &ShuffleMask);
+void DecodeSHUFPMask(EVT VT, unsigned Imm,
+ SmallVectorImpl<unsigned> &ShuffleMask);
-/// DecodeUNPCKLPMask - This decodes the shuffle masks for unpcklps/unpcklpd
+/// DecodeUNPCKHMask - 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 DecodeUNPCKLPMask(EVT VT,
- SmallVectorImpl<unsigned> &ShuffleMask);
+void DecodeUNPCKHMask(EVT VT, SmallVectorImpl<unsigned> &ShuffleMask);
+/// DecodeUNPCKLMask - 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 DecodeUNPCKLMask(EVT VT, SmallVectorImpl<unsigned> &ShuffleMask);
-// DecodeVPERMILPSMask - Decodes VPERMILPS permutes for any 128-bit 32-bit
-// elements. For 256-bit vectors, it's considered as two 128 lanes, the
-// referenced elements can't cross lanes and the mask of the first lane must
-// be the same of the second.
-void DecodeVPERMILPSMask(unsigned NElts, unsigned Imm,
- SmallVectorImpl<unsigned> &ShuffleMask);
-// DecodeVPERMILPDMask - Decodes VPERMILPD permutes for any 128-bit 64-bit
-// elements. For 256-bit vectors, it's considered as two 128 lanes, the
-// referenced elements can't cross lanes but the mask of the first lane can
-// be the different of the second (not like VPERMILPS).
-void DecodeVPERMILPDMask(unsigned NElts, unsigned Imm,
+// DecodeVPERMILPMask - Decodes VPERMILPS/ VPERMILPD permutes for any 128-bit
+// 32-bit or 64-bit elements. For 256-bit vectors, it's considered as two 128
+// lanes. For VPERMILPS, referenced elements can't cross lanes and the mask of
+// the first lane must be the same of the second.
+void DecodeVPERMILPMask(EVT VT, unsigned Imm,
SmallVectorImpl<unsigned> &ShuffleMask);
void DecodeVPERM2F128Mask(unsigned Imm,
generated by cgit v1.1 at 2025-11-11 01:08:29 +0000