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/LLVMBuild.txt2
-rw-r--r--lib/Target/X86/Utils/X86ShuffleDecode.cpp177
-rw-r--r--lib/Target/X86/Utils/X86ShuffleDecode.h31
3 files changed, 204 insertions, 6 deletions
diff --git a/lib/Target/X86/Utils/LLVMBuild.txt b/lib/Target/X86/Utils/LLVMBuild.txt
index fdb886f..de0a30f 100644
--- a/lib/Target/X86/Utils/LLVMBuild.txt
+++ b/lib/Target/X86/Utils/LLVMBuild.txt
@@ -19,5 +19,5 @@
type = Library
name = X86Utils
parent = X86
-required_libraries = Support
+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 5f2441c..ba6cbc8 100644
--- a/lib/Target/X86/Utils/X86ShuffleDecode.cpp
+++ b/lib/Target/X86/Utils/X86ShuffleDecode.cpp
@@ -13,6 +13,7 @@
//===----------------------------------------------------------------------===//
#include "X86ShuffleDecode.h"
+#include "llvm/IR/Constants.h"
#include "llvm/CodeGen/MachineValueType.h"
//===----------------------------------------------------------------------===//
@@ -62,6 +63,51 @@ void DecodeMOVLHPSMask(unsigned NElts, SmallVectorImpl<int> &ShuffleMask) {
ShuffleMask.push_back(NElts+i);
}
+void DecodeMOVSLDUPMask(MVT VT, SmallVectorImpl<int> &ShuffleMask) {
+ unsigned NumElts = VT.getVectorNumElements();
+ for (int i = 0, e = NumElts / 2; i < e; ++i) {
+ ShuffleMask.push_back(2 * i);
+ ShuffleMask.push_back(2 * i);
+ }
+}
+
+void DecodeMOVSHDUPMask(MVT VT, SmallVectorImpl<int> &ShuffleMask) {
+ unsigned NumElts = VT.getVectorNumElements();
+ for (int i = 0, e = NumElts / 2; i < e; ++i) {
+ ShuffleMask.push_back(2 * i + 1);
+ ShuffleMask.push_back(2 * i + 1);
+ }
+}
+
+void DecodePSLLDQMask(MVT VT, unsigned Imm, SmallVectorImpl<int> &ShuffleMask) {
+ unsigned VectorSizeInBits = VT.getSizeInBits();
+ unsigned NumElts = VectorSizeInBits / 8;
+ unsigned NumLanes = VectorSizeInBits / 128;
+ unsigned NumLaneElts = NumElts / NumLanes;
+
+ for (unsigned l = 0; l < NumElts; l += NumLaneElts)
+ for (unsigned i = 0; i < NumLaneElts; ++i) {
+ int M = SM_SentinelZero;
+ if (i >= Imm) M = i - Imm + l;
+ ShuffleMask.push_back(M);
+ }
+}
+
+void DecodePSRLDQMask(MVT VT, unsigned Imm, SmallVectorImpl<int> &ShuffleMask) {
+ unsigned VectorSizeInBits = VT.getSizeInBits();
+ unsigned NumElts = VectorSizeInBits / 8;
+ unsigned NumLanes = VectorSizeInBits / 128;
+ unsigned NumLaneElts = NumElts / NumLanes;
+
+ for (unsigned l = 0; l < NumElts; l += NumLaneElts)
+ for (unsigned i = 0; i < NumLaneElts; ++i) {
+ unsigned Base = i + Imm;
+ int M = Base + l;
+ if (Base >= NumLaneElts) M = SM_SentinelZero;
+ ShuffleMask.push_back(M);
+ }
+}
+
void DecodePALIGNRMask(MVT VT, unsigned Imm,
SmallVectorImpl<int> &ShuffleMask) {
unsigned NumElts = VT.getVectorNumElements();
@@ -207,6 +253,97 @@ void DecodeVPERM2X128Mask(MVT VT, unsigned Imm,
}
}
+void DecodePSHUFBMask(const Constant *C, SmallVectorImpl<int> &ShuffleMask) {
+ Type *MaskTy = C->getType();
+ assert(MaskTy->isVectorTy() && "Expected a vector constant mask!");
+ assert(MaskTy->getVectorElementType()->isIntegerTy(8) &&
+ "Expected i8 constant mask elements!");
+ int NumElements = MaskTy->getVectorNumElements();
+ // FIXME: Add support for AVX-512.
+ assert((NumElements == 16 || NumElements == 32) &&
+ "Only 128-bit and 256-bit vectors supported!");
+ ShuffleMask.reserve(NumElements);
+
+ if (auto *CDS = dyn_cast<ConstantDataSequential>(C)) {
+ assert((unsigned)NumElements == CDS->getNumElements() &&
+ "Constant mask has a different number of elements!");
+
+ for (int i = 0; i < NumElements; ++i) {
+ // For AVX vectors with 32 bytes the base of the shuffle is the 16-byte
+ // lane of the vector we're inside.
+ int Base = i < 16 ? 0 : 16;
+ uint64_t Element = CDS->getElementAsInteger(i);
+ // If the high bit (7) of the byte is set, the element is zeroed.
+ if (Element & (1 << 7))
+ ShuffleMask.push_back(SM_SentinelZero);
+ else {
+ // Only the least significant 4 bits of the byte are used.
+ int Index = Base + (Element & 0xf);
+ ShuffleMask.push_back(Index);
+ }
+ }
+ } else if (auto *CV = dyn_cast<ConstantVector>(C)) {
+ assert((unsigned)NumElements == CV->getNumOperands() &&
+ "Constant mask has a different number of elements!");
+
+ for (int i = 0; i < NumElements; ++i) {
+ // For AVX vectors with 32 bytes the base of the shuffle is the 16-byte
+ // lane of the vector we're inside.
+ int Base = i < 16 ? 0 : 16;
+ Constant *COp = CV->getOperand(i);
+ if (isa<UndefValue>(COp)) {
+ ShuffleMask.push_back(SM_SentinelUndef);
+ continue;
+ }
+ uint64_t Element = cast<ConstantInt>(COp)->getZExtValue();
+ // If the high bit (7) of the byte is set, the element is zeroed.
+ if (Element & (1 << 7))
+ ShuffleMask.push_back(SM_SentinelZero);
+ else {
+ // Only the least significant 4 bits of the byte are used.
+ int Index = Base + (Element & 0xf);
+ ShuffleMask.push_back(Index);
+ }
+ }
+ }
+}
+
+void DecodePSHUFBMask(ArrayRef<uint64_t> RawMask,
+ SmallVectorImpl<int> &ShuffleMask) {
+ for (int i = 0, e = RawMask.size(); i < e; ++i) {
+ uint64_t M = RawMask[i];
+ if (M == (uint64_t)SM_SentinelUndef) {
+ ShuffleMask.push_back(M);
+ continue;
+ }
+ // For AVX vectors with 32 bytes the base of the shuffle is the half of
+ // the vector we're inside.
+ int Base = i < 16 ? 0 : 16;
+ // If the high bit (7) of the byte is set, the element is zeroed.
+ if (M & (1 << 7))
+ ShuffleMask.push_back(SM_SentinelZero);
+ else {
+ // Only the least significant 4 bits of the byte are used.
+ int Index = Base + (M & 0xf);
+ ShuffleMask.push_back(Index);
+ }
+ }
+}
+
+void DecodeBLENDMask(MVT VT, unsigned Imm, SmallVectorImpl<int> &ShuffleMask) {
+ int ElementBits = VT.getScalarSizeInBits();
+ int NumElements = VT.getVectorNumElements();
+ for (int i = 0; i < NumElements; ++i) {
+ // If there are more than 8 elements in the vector, then any immediate blend
+ // mask applies to each 128-bit lane. There can never be more than
+ // 8 elements in a 128-bit lane with an immediate blend.
+ int Bit = NumElements > 8 ? i % (128 / ElementBits) : i;
+ assert(Bit < 8 &&
+ "Immediate blends only operate over 8 elements at a time!");
+ ShuffleMask.push_back(((Imm >> Bit) & 1) ? NumElements + i : i);
+ }
+}
+
/// DecodeVPERMMask - this decodes the shuffle masks for VPERMQ/VPERMPD.
/// No VT provided since it only works on 256-bit, 4 element vectors.
void DecodeVPERMMask(unsigned Imm, SmallVectorImpl<int> &ShuffleMask) {
@@ -215,4 +352,44 @@ void DecodeVPERMMask(unsigned Imm, SmallVectorImpl<int> &ShuffleMask) {
}
}
+void DecodeVPERMILPMask(const Constant *C, SmallVectorImpl<int> &ShuffleMask) {
+ Type *MaskTy = C->getType();
+ assert(MaskTy->isVectorTy() && "Expected a vector constant mask!");
+ assert(MaskTy->getVectorElementType()->isIntegerTy() &&
+ "Expected integer constant mask elements!");
+ int ElementBits = MaskTy->getScalarSizeInBits();
+ int NumElements = MaskTy->getVectorNumElements();
+ assert((NumElements == 2 || NumElements == 4 || NumElements == 8) &&
+ "Unexpected number of vector elements.");
+ ShuffleMask.reserve(NumElements);
+ if (auto *CDS = dyn_cast<ConstantDataSequential>(C)) {
+ assert((unsigned)NumElements == CDS->getNumElements() &&
+ "Constant mask has a different number of elements!");
+
+ for (int i = 0; i < NumElements; ++i) {
+ int Base = (i * ElementBits / 128) * (128 / ElementBits);
+ uint64_t Element = CDS->getElementAsInteger(i);
+ // Only the least significant 2 bits of the integer are used.
+ int Index = Base + (Element & 0x3);
+ ShuffleMask.push_back(Index);
+ }
+ } else if (auto *CV = dyn_cast<ConstantVector>(C)) {
+ assert((unsigned)NumElements == C->getNumOperands() &&
+ "Constant mask has a different number of elements!");
+
+ for (int i = 0; i < NumElements; ++i) {
+ int Base = (i * ElementBits / 128) * (128 / ElementBits);
+ Constant *COp = CV->getOperand(i);
+ if (isa<UndefValue>(COp)) {
+ ShuffleMask.push_back(SM_SentinelUndef);
+ continue;
+ }
+ uint64_t Element = cast<ConstantInt>(COp)->getZExtValue();
+ // Only the least significant 2 bits of the integer are used.
+ int Index = Base + (Element & 0x3);
+ ShuffleMask.push_back(Index);
+ }
+ }
+}
+
} // llvm namespace
diff --git a/lib/Target/X86/Utils/X86ShuffleDecode.h b/lib/Target/X86/Utils/X86ShuffleDecode.h
index 9e75b6b..6ba3c64 100644
--- a/lib/Target/X86/Utils/X86ShuffleDecode.h
+++ b/lib/Target/X86/Utils/X86ShuffleDecode.h
@@ -12,21 +12,21 @@
//
//===----------------------------------------------------------------------===//
-#ifndef X86_SHUFFLE_DECODE_H
-#define X86_SHUFFLE_DECODE_H
+#ifndef LLVM_LIB_TARGET_X86_UTILS_X86SHUFFLEDECODE_H
+#define LLVM_LIB_TARGET_X86_UTILS_X86SHUFFLEDECODE_H
#include "llvm/ADT/SmallVector.h"
+#include "llvm/ADT/ArrayRef.h"
//===----------------------------------------------------------------------===//
// Vector Mask Decoding
//===----------------------------------------------------------------------===//
namespace llvm {
+class Constant;
class MVT;
-enum {
- SM_SentinelZero = -1
-};
+enum { SM_SentinelUndef = -1, SM_SentinelZero = -2 };
void DecodeINSERTPSMask(unsigned Imm, SmallVectorImpl<int> &ShuffleMask);
@@ -36,6 +36,14 @@ void DecodeMOVHLPSMask(unsigned NElts, SmallVectorImpl<int> &ShuffleMask);
// <0,2> or <0,1,4,5>
void DecodeMOVLHPSMask(unsigned NElts, SmallVectorImpl<int> &ShuffleMask);
+void DecodeMOVSLDUPMask(MVT VT, SmallVectorImpl<int> &ShuffleMask);
+
+void DecodeMOVSHDUPMask(MVT VT, SmallVectorImpl<int> &ShuffleMask);
+
+void DecodePSLLDQMask(MVT VT, unsigned Imm, SmallVectorImpl<int> &ShuffleMask);
+
+void DecodePSRLDQMask(MVT VT, unsigned Imm, SmallVectorImpl<int> &ShuffleMask);
+
void DecodePALIGNRMask(MVT VT, unsigned Imm, SmallVectorImpl<int> &ShuffleMask);
void DecodePSHUFMask(MVT VT, unsigned Imm, SmallVectorImpl<int> &ShuffleMask);
@@ -59,6 +67,16 @@ void DecodeUNPCKHMask(MVT VT, SmallVectorImpl<int> &ShuffleMask);
/// different datatypes and vector widths.
void DecodeUNPCKLMask(MVT VT, SmallVectorImpl<int> &ShuffleMask);
+/// \brief Decode a PSHUFB mask from an IR-level vector constant.
+void DecodePSHUFBMask(const Constant *C, SmallVectorImpl<int> &ShuffleMask);
+
+/// \brief Decode a PSHUFB mask from a raw array of constants such as from
+/// BUILD_VECTOR.
+void DecodePSHUFBMask(ArrayRef<uint64_t> RawMask,
+ SmallVectorImpl<int> &ShuffleMask);
+
+/// \brief Decode a BLEND immediate mask into a shuffle mask.
+void DecodeBLENDMask(MVT VT, unsigned Imm, SmallVectorImpl<int> &ShuffleMask);
void DecodeVPERM2X128Mask(MVT VT, unsigned Imm,
SmallVectorImpl<int> &ShuffleMask);
@@ -67,6 +85,9 @@ void DecodeVPERM2X128Mask(MVT VT, unsigned Imm,
/// No VT provided since it only works on 256-bit, 4 element vectors.
void DecodeVPERMMask(unsigned Imm, SmallVectorImpl<int> &ShuffleMask);
+/// \brief Decode a VPERMILP variable mask from an IR-level vector constant.
+void DecodeVPERMILPMask(const Constant *C, SmallVectorImpl<int> &ShuffleMask);
+
} // llvm namespace
#endif
generated by cgit v1.1 at 2024-06-09 10:37:06 +0000