[Headers][X86] Allow SLLDQ/SRLDQ byte shift intrinsics to be used in constexpr #164166
+136
−4
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…srli_si256 intrinsics
…slli_si256 intrinsics
llvmbot
added
clang
|
@llvm/pr-subscribers-clang @llvm/pr-subscribers-backend-x86 Author: Ye Tian (TianYe717) ChangesSupport constexpr usage for SLLDQ/SRLDQ byte shift intrinsics This draft PR adds support for using the following SRLDQ intrinsics in constant expressions:
Relevant tests are included. Addresses #156494. Full diff: https://github.com/llvm/llvm-project/pull/164166.diff 5 Files Affected:
diff --git a/clang/include/clang/Basic/BuiltinsX86.td b/clang/include/clang/Basic/BuiltinsX86.td
index d03c778740ad3..7f19f025d6af3 100644
--- a/clang/include/clang/Basic/BuiltinsX86.td
+++ b/clang/include/clang/Basic/BuiltinsX86.td
@@ -282,8 +282,6 @@ let Features = "sse2", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] i
def psllw128 : X86Builtin<"_Vector<8, short>(_Vector<8, short>, _Vector<8, short>)">;
def pslld128 : X86Builtin<"_Vector<4, int>(_Vector<4, int>, _Vector<4, int>)">;
def psllq128 : X86Builtin<"_Vector<2, long long int>(_Vector<2, long long int>, _Vector<2, long long int>)">;
- def pslldqi128_byteshift : X86Builtin<"_Vector<16, char>(_Vector<16, char>, _Constant int)">;
- def psrldqi128_byteshift : X86Builtin<"_Vector<16, char>(_Vector<16, char>, _Constant int)">;
}
let Features = "sse2",
@@ -302,6 +300,9 @@ let Features = "sse2",
def psrawi128 : X86Builtin<"_Vector<8, short>(_Vector<8, short>, int)">;
def psradi128 : X86Builtin<"_Vector<4, int>(_Vector<4, int>, int)">;
+
+ def pslldqi128_byteshift : X86Builtin<"_Vector<16, char>(_Vector<16, char>, _Constant int)">;
+ def psrldqi128_byteshift : X86Builtin<"_Vector<16, char>(_Vector<16, char>, _Constant int)">;
}
let Features = "sse3", Attributes = [NoThrow] in {
@@ -613,12 +614,10 @@ let Features = "avx2", Attributes = [NoThrow, Const, RequiredVectorWidth<256>] i
def psignw256 : X86Builtin<"_Vector<16, short>(_Vector<16, short>, _Vector<16, short>)">;
def psignd256 : X86Builtin<"_Vector<8, int>(_Vector<8, int>, _Vector<8, int>)">;
def psllw256 : X86Builtin<"_Vector<16, short>(_Vector<16, short>, _Vector<8, short>)">;
- def pslldqi256_byteshift : X86Builtin<"_Vector<32, char>(_Vector<32, char>, _Constant int)">;
def pslld256 : X86Builtin<"_Vector<8, int>(_Vector<8, int>, _Vector<4, int>)">;
def psllq256 : X86Builtin<"_Vector<4, long long int>(_Vector<4, long long int>, _Vector<2, long long int>)">;
def psraw256 : X86Builtin<"_Vector<16, short>(_Vector<16, short>, _Vector<8, short>)">;
def psrad256 : X86Builtin<"_Vector<8, int>(_Vector<8, int>, _Vector<4, int>)">;
- def psrldqi256_byteshift : X86Builtin<"_Vector<32, char>(_Vector<32, char>, _Constant int)">;
def psrlw256 : X86Builtin<"_Vector<16, short>(_Vector<16, short>, _Vector<8, short>)">;
def psrld256 : X86Builtin<"_Vector<8, int>(_Vector<8, int>, _Vector<4, int>)">;
def psrlq256 : X86Builtin<"_Vector<4, long long int>(_Vector<4, long long int>, _Vector<2, long long int>)">;
@@ -652,10 +651,12 @@ let Features = "avx2", Attributes = [NoThrow, Const, Constexpr, RequiredVectorWi
def psllwi256 : X86Builtin<"_Vector<16, short>(_Vector<16, short>, int)">;
def pslldi256 : X86Builtin<"_Vector<8, int>(_Vector<8, int>, int)">;
def psllqi256 : X86Builtin<"_Vector<4, long long int>(_Vector<4, long long int>, int)">;
+ def pslldqi256_byteshift : X86Builtin<"_Vector<32, char>(_Vector<32, char>, _Constant int)">;
def psrlwi256 : X86Builtin<"_Vector<16, short>(_Vector<16, short>, int)">;
def psrldi256 : X86Builtin<"_Vector<8, int>(_Vector<8, int>, int)">;
def psrlqi256 : X86Builtin<"_Vector<4, long long int>(_Vector<4, long long int>, int)">;
+ def psrldqi256_byteshift : X86Builtin<"_Vector<32, char>(_Vector<32, char>, _Constant int)">;
def psrawi256 : X86Builtin<"_Vector<16, short>(_Vector<16, short>, int)">;
def psradi256 : X86Builtin<"_Vector<8, int>(_Vector<8, int>, int)">;
diff --git a/clang/lib/AST/ByteCode/InterpBuiltin.cpp b/clang/lib/AST/ByteCode/InterpBuiltin.cpp
index 0cb491063057c..85e18f2436885 100644
--- a/clang/lib/AST/ByteCode/InterpBuiltin.cpp
+++ b/clang/lib/AST/ByteCode/InterpBuiltin.cpp
@@ -3128,6 +3128,67 @@ static bool interp__builtin_ia32_vpconflict(InterpState &S, CodePtr OpPC,
return true;
}
+static bool interp__builtin_x86_pslldq_byteshift(InterpState &S, CodePtr OpPC,
+ const CallExpr *Call,
+ unsigned ID) {
+ assert(Call->getNumArgs() == 2);
+
+ APSInt ImmAPS = popToAPSInt(S, Call->getArg(1));
+ uint64_t Shift = ImmAPS.getZExtValue();
+
+ const Pointer &Src = S.Stk.pop<Pointer>();
+ if (!Src.getFieldDesc()->isPrimitiveArray())
+ return false;
+
+ unsigned NumElems = Src.getNumElems();
+ const Pointer &Dst = S.Stk.peek<Pointer>();
+ PrimType ElemT = Src.getFieldDesc()->getPrimType();
+
+ TYPE_SWITCH(ElemT, {
+ for (unsigned I = 0; I != NumElems; ++I) {
+ if (I < Shift) {
+ Dst.elem<T>(I) = T();
+ } else {
+ Dst.elem<T>(I) = Src.elem<T>(I - Shift);
+ }
+ }
+ });
+
+ Dst.initializeAllElements();
+
+ return true;
+}
+
+static bool interp__builtin_x86_psrldq_byteshift(InterpState &S, CodePtr OpPC,
+ const CallExpr *Call,
+ unsigned ID) {
+ assert(Call->getNumArgs() == 2);
+
+ APSInt ImmAPS = popToAPSInt(S, Call->getArg(1));
+ uint64_t Shift = ImmAPS.getZExtValue();
+
+ const Pointer &Src = S.Stk.pop<Pointer>();
+ if (!Src.getFieldDesc()->isPrimitiveArray())
+ return false;
+
+ unsigned NumElems = Src.getNumElems();
+ const Pointer &Dst = S.Stk.peek<Pointer>();
+ PrimType ElemT = Src.getFieldDesc()->getPrimType();
+
+ TYPE_SWITCH(ElemT, {
+ for (unsigned I = 0; I != NumElems; ++I) {
+ if (I + Shift < NumElems)
+ Dst.elem<T>(I) = Src.elem<T>(I + Shift);
+ else
+ Dst.elem<T>(I) = T();
+ }
+ });
+
+ Dst.initializeAllElements();
+
+ return true;
+}
+
bool InterpretBuiltin(InterpState &S, CodePtr OpPC, const CallExpr *Call,
uint32_t BuiltinID) {
if (!S.getASTContext().BuiltinInfo.isConstantEvaluated(BuiltinID))
@@ -4149,6 +4210,14 @@ bool InterpretBuiltin(InterpState &S, CodePtr OpPC, const CallExpr *Call,
case X86::BI__builtin_ia32_vec_set_v4di:
return interp__builtin_vec_set(S, OpPC, Call, BuiltinID);
+ case X86::BI__builtin_ia32_pslldqi128_byteshift:
+ case X86::BI__builtin_ia32_pslldqi256_byteshift:
+ return interp__builtin_x86_pslldq_byteshift(S, OpPC, Call, BuiltinID);
+
+ case X86::BI__builtin_ia32_psrldqi128_byteshift:
+ case X86::BI__builtin_ia32_psrldqi256_byteshift:
+ return interp__builtin_x86_psrldq_byteshift(S, OpPC, Call, BuiltinID);
+
default:
S.FFDiag(S.Current->getLocation(OpPC),
diag::note_invalid_subexpr_in_const_expr)
diff --git a/clang/lib/AST/ExprConstant.cpp b/clang/lib/AST/ExprConstant.cpp
index e308c171ed551..db9ab7f0b9e33 100644
--- a/clang/lib/AST/ExprConstant.cpp
+++ b/clang/lib/AST/ExprConstant.cpp
@@ -12790,6 +12790,60 @@ bool VectorExprEvaluator::VisitCallExpr(const CallExpr *E) {
return Success(APValue(Elems.data(), NumElems), E);
}
+
+ case X86::BI__builtin_ia32_pslldqi128_byteshift:
+ case X86::BI__builtin_ia32_pslldqi256_byteshift: {
+ assert(E->getNumArgs() == 2);
+
+ APValue Src;
+ APSInt Imm;
+ if (!EvaluateAsRValue(Info, E->getArg(0), Src) ||
+ !EvaluateInteger(E->getArg(1), Imm, Info))
+ return false;
+
+ unsigned VecLen = Src.getVectorLength();
+ unsigned Shift = Imm.getZExtValue();
+
+ SmallVector<APValue> ResultElements;
+ for (unsigned I = 0; I != VecLen; ++I) {
+ if (I < Shift) {
+ APSInt Zero(8, /*isUnsigned=*/true);
+ Zero = 0;
+ ResultElements.push_back(APValue(Zero));
+ } else {
+ ResultElements.push_back(Src.getVectorElt(I - Shift));
+ }
+ }
+
+ return Success(APValue(ResultElements.data(), ResultElements.size()), E);
+ }
+
+ case X86::BI__builtin_ia32_psrldqi128_byteshift:
+ case X86::BI__builtin_ia32_psrldqi256_byteshift: {
+ assert(E->getNumArgs() == 2);
+
+ APValue Src;
+ APSInt Imm;
+ if (!EvaluateAsRValue(Info, E->getArg(0), Src) ||
+ !EvaluateInteger(E->getArg(1), Imm, Info))
+ return false;
+
+ unsigned VecLen = Src.getVectorLength();
+ unsigned Shift = Imm.getZExtValue();
+
+ SmallVector<APValue> ResultElements;
+ for (unsigned I = 0; I < VecLen; ++I) {
+ if (I + Shift < VecLen) {
+ ResultElements.push_back(Src.getVectorElt(I + Shift));
+ } else {
+ APSInt Zero(8, /*isUnsigned=*/true);
+ Zero = 0;
+ ResultElements.push_back(APValue(Zero));
+ }
+ }
+
+ return Success(APValue(ResultElements.data(), ResultElements.size()), E);
+ }
}
}
diff --git a/clang/test/CodeGen/X86/avx2-builtins.c b/clang/test/CodeGen/X86/avx2-builtins.c
index a505d70a98203..79ba172ad0938 100644
--- a/clang/test/CodeGen/X86/avx2-builtins.c
+++ b/clang/test/CodeGen/X86/avx2-builtins.c
@@ -1225,6 +1225,8 @@ __m256i test_mm256_slli_si256(__m256i a) {
// CHECK: shufflevector <32 x i8> zeroinitializer, <32 x i8> %{{.*}}, <32 x i32> <i32 13, i32 14, i32 15, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 29, i32 30, i32 31, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60>
return _mm256_slli_si256(a, 3);
}
+TEST_CONSTEXPR(match_v32qi(_mm256_slli_si256(((__m256i)(__v32qi){1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32}), 3), 0, 0, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29));
+TEST_CONSTEXPR(match_v32qi(_mm256_slli_si256(((__m256i)(__v32qi){1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32}), 32), 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0));
__m128i test_mm_sllv_epi32(__m128i a, __m128i b) {
// CHECK-LABEL: test_mm_sllv_epi32
@@ -1368,6 +1370,8 @@ __m256i test_mm256_srli_si256(__m256i a) {
// CHECK: shufflevector <32 x i8> %{{.*}}, <32 x i8> zeroinitializer, <32 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 32, i32 33, i32 34, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 48, i32 49, i32 50>
return _mm256_srli_si256(a, 3);
}
+TEST_CONSTEXPR(match_v32qi(_mm256_srli_si256(((__m256i)(__v32qi){0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31}), 3), 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 0, 0, 0));
+TEST_CONSTEXPR(match_v32qi(_mm256_srli_si256(((__m256i)(__v32qi){0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31}), 32), 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0));
__m128i test_mm_srlv_epi32(__m128i a, __m128i b) {
// CHECK-LABEL: test_mm_srlv_epi32
diff --git a/clang/test/CodeGen/X86/sse2-builtins.c b/clang/test/CodeGen/X86/sse2-builtins.c
index ade7ef39a008a..4faae3f1f35d1 100644
--- a/clang/test/CodeGen/X86/sse2-builtins.c
+++ b/clang/test/CodeGen/X86/sse2-builtins.c
@@ -1413,6 +1413,8 @@ __m128i test_mm_slli_si128(__m128i A) {
// CHECK: shufflevector <16 x i8> zeroinitializer, <16 x i8> %{{.*}}, <16 x i32> <i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26>
return _mm_slli_si128(A, 5);
}
+TEST_CONSTEXPR(match_v16qi(_mm_slli_si128(((__m128i)(__v16qi){1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16}), 5), 0, 0, 0, 0, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11));
+TEST_CONSTEXPR(match_v16qi(_mm_slli_si128(((__m128i)(__v16qi){1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16}), 16), 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0));
__m128i test_mm_slli_si128_2(__m128i A) {
// CHECK-LABEL: test_mm_slli_si128_2
@@ -1565,6 +1567,8 @@ __m128i test_mm_srli_si128(__m128i A) {
// CHECK: shufflevector <16 x i8> %{{.*}}, <16 x i8> zeroinitializer, <16 x i32> <i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20>
return _mm_srli_si128(A, 5);
}
+TEST_CONSTEXPR(match_v16qi(_mm_srli_si128(((__m128i)(__v16qi){0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15}), 5), 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 0, 0, 0, 0, 0));
+TEST_CONSTEXPR(match_v16qi(_mm_srli_si128(((__m128i)(__v16qi){0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15}), 16), 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0));
__m128i test_mm_srli_si128_2(__m128i A) {
// CHECK-LABEL: test_mm_srli_si128_2
|
|
I don't think I have an immediate problem with the implementation (at least not the parts in |
RKSimon
requested changes
Oct 20, 2025
| PrimType ElemT = Src.getFieldDesc()->getPrimType(); | ||
|
|
||
| TYPE_SWITCH(ElemT, { | ||
| for (unsigned I = 0; I != NumElems; ++I) { |
There was a problem hiding this comment.
The byte shifts are done PER 128-bit vector lane - see
llvm-project/clang/lib/CodeGen/TargetBuiltins/X86.cpp
Lines 1813 to 1838 in 737e116
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Support constexpr usage for SLLDQ/SRLDQ byte shift intrinsics
This draft PR adds support for using the following SRLDQ intrinsics in constant expressions:
Relevant tests are included.
Fixes #156494