[Clang] VectorExprEvaluator::VisitCallExpr / InterpretBuiltin - allow AVX/AVX512 subvector extraction intrinsics to be used in constexpr #157712 #158853
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
|
Thank you for submitting a Pull Request (PR) to the LLVM Project! This PR will be automatically labeled and the relevant teams will be notified. If you wish to, you can add reviewers by using the "Reviewers" section on this page. If this is not working for you, it is probably because you do not have write permissions for the repository. In which case you can instead tag reviewers by name in a comment by using If you have received no comments on your PR for a week, you can request a review by "ping"ing the PR by adding a comment “Ping”. The common courtesy "ping" rate is once a week. Please remember that you are asking for valuable time from other developers. If you have further questions, they may be answered by the LLVM GitHub User Guide. You can also ask questions in a comment on this PR, on the LLVM Discord or on the forums. |
SeongjaeP
force-pushed
the
issue157712-avx-extract
branch
from
e9b71e5 to
3028c5f
Compare
SeongjaeP
changed the title
SeongjaeP
force-pushed
the
issue157712-avx-extract
branch
from
9412543 to
1292451
Compare
|
Hi @RKSimon, could you please take a look at this PR when you have time? |
|
@SeongjaeP Sorry but I missed this in my notifications - please can you resolves the merge conflicts? |
RKSimon
requested changes
Oct 6, 2025
tbaederr
reviewed
Oct 7, 2025
tbaederr
reviewed
Oct 9, 2025
| pushInteger(S, N == Value.getBitWidth() ? 0 : N + 1, Call->getType()); | ||
| return true; | ||
| } | ||
|
|
There was a problem hiding this comment.
Those two functions were removed upstream.
There was a problem hiding this comment.
Looks like a merge went wrong somewhere, causing these issues.
I’ll resolve the conflicts and fix it soon — thanks for the review!
tbaederr
added
clang:frontend
You can test this locally with the following command:git-clang-format --diff origin/main HEAD --extensions h,cpp,c -- clang/lib/AST/ByteCode/InterpBuiltin.cpp clang/lib/AST/ExprConstant.cpp clang/lib/Headers/avx512dqintrin.h clang/lib/Headers/avx512fintrin.h clang/lib/Headers/avx512vldqintrin.h clang/lib/Headers/avx512vlintrin.h clang/test/CodeGen/X86/avx-builtins.c clang/test/CodeGen/X86/avx2-builtins.c clang/test/CodeGen/X86/avx512dq-builtins.c clang/test/CodeGen/X86/avx512f-builtins.c clang/test/CodeGen/X86/avx512vl-builtins.c clang/test/CodeGen/X86/avx512vldq-builtins.c
View the diff from clang-format here.diff --git a/clang/lib/AST/ByteCode/InterpBuiltin.cpp b/clang/lib/AST/ByteCode/InterpBuiltin.cpp
index c8479b9b0..da5cd25ec 100644
--- a/clang/lib/AST/ByteCode/InterpBuiltin.cpp
+++ b/clang/lib/AST/ByteCode/InterpBuiltin.cpp
@@ -736,7 +736,6 @@ static bool interp__builtin_expect(InterpState &S, CodePtr OpPC,
return true;
}
-
/// rotateleft(value, amount)
static bool interp__builtin_rotate(InterpState &S, CodePtr OpPC,
const InterpFrame *Frame,
@@ -2839,8 +2838,8 @@ static bool interp__builtin_elementwise_triop(
//_builtin_extract
static bool interp__builtin_x86_extract_vector(InterpState &S, CodePtr OpPC,
- const CallExpr *Call,
- unsigned ID) {
+ const CallExpr *Call,
+ unsigned ID) {
assert(Call->getNumArgs() == 2);
APSInt ImmAPS = popToAPSInt(S, Call->getArg(1));
@@ -2878,7 +2877,8 @@ static bool interp__builtin_x86_extract_vector(InterpState &S, CodePtr OpPC,
return true;
}
-static bool interp__builtin_x86_extract_vector_masked(InterpState &S, CodePtr OpPC,
+static bool interp__builtin_x86_extract_vector_masked(InterpState &S,
+ CodePtr OpPC,
const CallExpr *Call,
unsigned ID) {
assert(Call->getNumArgs() == 4);
@@ -2888,7 +2888,8 @@ static bool interp__builtin_x86_extract_vector_masked(InterpState &S, CodePtr Op
APSInt ImmAPS = popToAPSInt(S, Call->getArg(1));
const Pointer &Src = S.Stk.pop<Pointer>();
- if (!Src.getFieldDesc()->isPrimitiveArray() || !Merge.getFieldDesc()->isPrimitiveArray())
+ if (!Src.getFieldDesc()->isPrimitiveArray() ||
+ !Merge.getFieldDesc()->isPrimitiveArray())
return false;
const Pointer &Dst = S.Stk.peek<Pointer>();
@@ -2916,7 +2917,7 @@ static bool interp__builtin_x86_extract_vector_masked(InterpState &S, CodePtr Op
if ((Mask >> I) & 1)
Dst.elem<T>(I) = Src.elem<T>(Base + I);
else
- Dst.elem<T>(I) = Merge.elem<T>(I);
+ Dst.elem<T>(I) = Merge.elem<T>(I);
}
});
@@ -2924,7 +2925,6 @@ static bool interp__builtin_x86_extract_vector_masked(InterpState &S, CodePtr Op
return true;
}
-
static bool interp__builtin_x86_insert_subvector(InterpState &S, CodePtr OpPC,
const CallExpr *Call,
unsigned ID) {
@@ -3444,13 +3444,13 @@ bool InterpretBuiltin(InterpState &S, CodePtr OpPC, const CallExpr *Call,
return LHS.isSigned() ? LHS.ssub_sat(RHS) : LHS.usub_sat(RHS);
});
- case X86::BI__builtin_ia32_extract128i256: // _mm256_extracti128
- case X86::BI__builtin_ia32_vextractf128_pd256: // _mm256_extractf128_ps
- case X86::BI__builtin_ia32_vextractf128_ps256: // _mm256_extractf128_pd
- case X86::BI__builtin_ia32_vextractf128_si256: // _mm256_extracti128_si256
+ case X86::BI__builtin_ia32_extract128i256: // _mm256_extracti128
+ case X86::BI__builtin_ia32_vextractf128_pd256: // _mm256_extractf128_ps
+ case X86::BI__builtin_ia32_vextractf128_ps256: // _mm256_extractf128_pd
+ case X86::BI__builtin_ia32_vextractf128_si256: // _mm256_extracti128_si256
return interp__builtin_x86_extract_vector(S, OpPC, Call, BuiltinID);
- // AVX-512 / AVX-512VL / AVX-512DQ
+ // AVX-512 / AVX-512VL / AVX-512DQ
case X86::BI__builtin_ia32_extractf32x4_256_mask:
case X86::BI__builtin_ia32_extractf32x4_mask:
case X86::BI__builtin_ia32_extractf32x8_mask:
@@ -3465,7 +3465,6 @@ bool InterpretBuiltin(InterpState &S, CodePtr OpPC, const CallExpr *Call,
case X86::BI__builtin_ia32_extracti64x4_mask:
return interp__builtin_x86_extract_vector_masked(S, OpPC, Call, BuiltinID);
-
case clang::X86::BI__builtin_ia32_pavgb128:
case clang::X86::BI__builtin_ia32_pavgw128:
case clang::X86::BI__builtin_ia32_pavgb256:
diff --git a/clang/lib/AST/ExprConstant.cpp b/clang/lib/AST/ExprConstant.cpp
index 281f9a360..510011d0f 100644
--- a/clang/lib/AST/ExprConstant.cpp
+++ b/clang/lib/AST/ExprConstant.cpp
@@ -12037,56 +12037,59 @@ bool VectorExprEvaluator::VisitCallExpr(const CallExpr *E) {
if (!EvaluateAsRValue(Info, E->getArg(0), SourceVec) ||
!EvaluateAsRValue(Info, E->getArg(1), SourceImm))
return false;
-
+
if (!SourceVec.isVector())
return false;
const auto *RetVT = E->getType()->castAs<VectorType>();
- if (!RetVT) return false;
+ if (!RetVT)
+ return false;
unsigned RetLen = RetVT->getNumElements();
unsigned SrcLen = SourceVec.getVectorLength();
- if (SrcLen != RetLen * 2)
+ if (SrcLen != RetLen * 2)
return false;
unsigned Idx = SourceImm.getInt().getZExtValue() & 1;
-
+
SmallVector<APValue, 32> ResultElements;
ResultElements.reserve(RetLen);
for (unsigned I = 0; I < RetLen; I++)
ResultElements.push_back(SourceVec.getVectorElt(Idx * RetLen + I));
-
+
return Success(APValue(ResultElements.data(), RetLen), E);
}
- case X86::BI__builtin_ia32_extracti32x4_256_mask:
+ case X86::BI__builtin_ia32_extracti32x4_256_mask:
case X86::BI__builtin_ia32_extractf32x4_256_mask:
- case X86::BI__builtin_ia32_extracti32x4_mask:
- case X86::BI__builtin_ia32_extractf32x4_mask:
- case X86::BI__builtin_ia32_extracti32x8_mask:
- case X86::BI__builtin_ia32_extractf32x8_mask:
- case X86::BI__builtin_ia32_extracti64x2_256_mask:
- case X86::BI__builtin_ia32_extractf64x2_256_mask:
- case X86::BI__builtin_ia32_extracti64x2_512_mask:
+ case X86::BI__builtin_ia32_extracti32x4_mask:
+ case X86::BI__builtin_ia32_extractf32x4_mask:
+ case X86::BI__builtin_ia32_extracti32x8_mask:
+ case X86::BI__builtin_ia32_extractf32x8_mask:
+ case X86::BI__builtin_ia32_extracti64x2_256_mask:
+ case X86::BI__builtin_ia32_extractf64x2_256_mask:
+ case X86::BI__builtin_ia32_extracti64x2_512_mask:
case X86::BI__builtin_ia32_extractf64x2_512_mask:
case X86::BI__builtin_ia32_extracti64x4_mask:
- case X86::BI__builtin_ia32_extractf64x4_mask:{
+ case X86::BI__builtin_ia32_extractf64x4_mask: {
APValue SourceVec, MergeVec;
APSInt Imm, MaskImm;
- if (!EvaluateAsRValue(Info, E->getArg(0), SourceVec) ||
- !EvaluateInteger(E->getArg(1), Imm, Info) ||
- !EvaluateAsRValue(Info, E->getArg(2), MergeVec) ||
- !EvaluateInteger(E->getArg(3), MaskImm, Info))
- return false;
+ if (!EvaluateAsRValue(Info, E->getArg(0), SourceVec) ||
+ !EvaluateInteger(E->getArg(1), Imm, Info) ||
+ !EvaluateAsRValue(Info, E->getArg(2), MergeVec) ||
+ !EvaluateInteger(E->getArg(3), MaskImm, Info))
+ return false;
const auto *RetVT = E->getType()->castAs<VectorType>();
unsigned RetLen = RetVT->getNumElements();
- if (!SourceVec.isVector() || !MergeVec.isVector()) return false;
+ if (!SourceVec.isVector() || !MergeVec.isVector())
+ return false;
unsigned SrcLen = SourceVec.getVectorLength();
- if (!SrcLen || !RetLen || (SrcLen % RetLen) != 0) return false;
+ if (!SrcLen || !RetLen || (SrcLen % RetLen) != 0)
+ return false;
unsigned Lanes = SrcLen / RetLen;
unsigned Lane = static_cast<unsigned>(Imm.getZExtValue() % Lanes);
@@ -12099,11 +12102,10 @@ bool VectorExprEvaluator::VisitCallExpr(const CallExpr *E) {
if ((Mask >> I) & 1)
ResultElements.push_back(SourceVec.getVectorElt(Base + I));
else
- ResultElements.push_back(MergeVec.getVectorElt(I));
+ ResultElements.push_back(MergeVec.getVectorElt(I));
}
return Success(APValue(ResultElements.data(), ResultElements.size()), E);
}
-
case X86::BI__builtin_ia32_vpshldd128:
case X86::BI__builtin_ia32_vpshldd256:
diff --git a/clang/lib/Headers/avx512dqintrin.h b/clang/lib/Headers/avx512dqintrin.h
index 0ff776b36..b2fb02ab1 100644
--- a/clang/lib/Headers/avx512dqintrin.h
+++ b/clang/lib/Headers/avx512dqintrin.h
@@ -1212,10 +1212,10 @@ _mm512_maskz_broadcast_i64x2(__mmask8 __M, __m128i __A)
(__v8di)_mm512_setzero_si512());
}
-#define _mm512_extractf32x8_ps(A, imm) \
- ((__m256)__builtin_ia32_extractf32x8_mask((__v16sf)(__m512)(A), (int)(imm), \
- (__v8sf)_mm256_setzero_ps(), \
- (__mmask8)-1))
+#define _mm512_extractf32x8_ps(A, imm) \
+ ((__m256)__builtin_ia32_extractf32x8_mask((__v16sf)(__m512)(A), (int)(imm), \
+ (__v8sf)_mm256_setzero_ps(), \
+ (__mmask8) - 1))
#define _mm512_mask_extractf32x8_ps(W, U, A, imm) \
((__m256)__builtin_ia32_extractf32x8_mask((__v16sf)(__m512)(A), (int)(imm), \
@@ -1227,11 +1227,10 @@ _mm512_maskz_broadcast_i64x2(__mmask8 __M, __m128i __A)
(__v8sf)_mm256_setzero_ps(), \
(__mmask8)(U)))
-#define _mm512_extractf64x2_pd(A, imm) \
- ((__m128d)__builtin_ia32_extractf64x2_512_mask((__v8df)(__m512d)(A), \
- (int)(imm), \
- (__v2df)_mm_setzero_pd(), \
- (__mmask8)-1))
+#define _mm512_extractf64x2_pd(A, imm) \
+ ((__m128d)__builtin_ia32_extractf64x2_512_mask( \
+ (__v8df)(__m512d)(A), (int)(imm), (__v2df)_mm_setzero_pd(), \
+ (__mmask8) - 1))
#define _mm512_mask_extractf64x2_pd(W, U, A, imm) \
((__m128d)__builtin_ia32_extractf64x2_512_mask((__v8df)(__m512d)(A), \
@@ -1245,10 +1244,10 @@ _mm512_maskz_broadcast_i64x2(__mmask8 __M, __m128i __A)
(__v2df)_mm_setzero_pd(), \
(__mmask8)(U)))
-#define _mm512_extracti32x8_epi32(A, imm) \
- ((__m256i)__builtin_ia32_extracti32x8_mask((__v16si)(__m512i)(A), (int)(imm), \
- (__v8si)_mm256_setzero_si256(), \
- (__mmask8)-1))
+#define _mm512_extracti32x8_epi32(A, imm) \
+ ((__m256i)__builtin_ia32_extracti32x8_mask( \
+ (__v16si)(__m512i)(A), (int)(imm), (__v8si)_mm256_setzero_si256(), \
+ (__mmask8) - 1))
#define _mm512_mask_extracti32x8_epi32(W, U, A, imm) \
((__m256i)__builtin_ia32_extracti32x8_mask((__v16si)(__m512i)(A), (int)(imm), \
@@ -1260,11 +1259,10 @@ _mm512_maskz_broadcast_i64x2(__mmask8 __M, __m128i __A)
(__v8si)_mm256_setzero_si256(), \
(__mmask8)(U)))
-#define _mm512_extracti64x2_epi64(A, imm) \
- ((__m128i)__builtin_ia32_extracti64x2_512_mask((__v8di)(__m512i)(A), \
- (int)(imm), \
- (__v2di)_mm_setzero_si128(), \
- (__mmask8)-1))
+#define _mm512_extracti64x2_epi64(A, imm) \
+ ((__m128i)__builtin_ia32_extracti64x2_512_mask( \
+ (__v8di)(__m512i)(A), (int)(imm), (__v2di)_mm_setzero_si128(), \
+ (__mmask8) - 1))
#define _mm512_mask_extracti64x2_epi64(W, U, A, imm) \
((__m128i)__builtin_ia32_extracti64x2_512_mask((__v8di)(__m512i)(A), \
diff --git a/clang/lib/Headers/avx512fintrin.h b/clang/lib/Headers/avx512fintrin.h
index 2768a5bae..13370aaef 100644
--- a/clang/lib/Headers/avx512fintrin.h
+++ b/clang/lib/Headers/avx512fintrin.h
@@ -3164,10 +3164,10 @@ _mm512_maskz_permutex2var_epi64(__mmask8 __U, __m512i __A, __m512i __I,
(__v16si)_mm512_setzero_si512()))
/* Vector Extract */
-#define _mm512_extractf64x4_pd(A, I) \
- ((__m256d)__builtin_ia32_extractf64x4_mask((__v8df)(__m512d)(A), (int)(I), \
- (__v4df)_mm256_setzero_pd(), \
- (__mmask8)-1))
+#define _mm512_extractf64x4_pd(A, I) \
+ ((__m256d)__builtin_ia32_extractf64x4_mask((__v8df)(__m512d)(A), (int)(I), \
+ (__v4df)_mm256_setzero_pd(), \
+ (__mmask8) - 1))
#define _mm512_mask_extractf64x4_pd(W, U, A, imm) \
((__m256d)__builtin_ia32_extractf64x4_mask((__v8df)(__m512d)(A), (int)(imm), \
@@ -3179,10 +3179,10 @@ _mm512_maskz_permutex2var_epi64(__mmask8 __U, __m512i __A, __m512i __I,
(__v4df)_mm256_setzero_pd(), \
(__mmask8)(U)))
-#define _mm512_extractf32x4_ps(A, I) \
- ((__m128)__builtin_ia32_extractf32x4_mask((__v16sf)(__m512)(A), (int)(I), \
- (__v4sf)_mm_setzero_ps(), \
- (__mmask8)-1))
+#define _mm512_extractf32x4_ps(A, I) \
+ ((__m128)__builtin_ia32_extractf32x4_mask((__v16sf)(__m512)(A), (int)(I), \
+ (__v4sf)_mm_setzero_ps(), \
+ (__mmask8) - 1))
#define _mm512_mask_extractf32x4_ps(W, U, A, imm) \
((__m128)__builtin_ia32_extractf32x4_mask((__v16sf)(__m512)(A), (int)(imm), \
@@ -7105,10 +7105,10 @@ _mm512_mask_cvtepi64_storeu_epi16 (void *__P, __mmask8 __M, __m512i __A)
__builtin_ia32_pmovqw512mem_mask ((__v8hi *) __P, (__v8di) __A, __M);
}
-#define _mm512_extracti32x4_epi32(A, imm) \
- ((__m128i)__builtin_ia32_extracti32x4_mask((__v16si)(__m512i)(A), (int)(imm), \
- (__v4si)_mm_setzero_si128(), \
- (__mmask8)-1))
+#define _mm512_extracti32x4_epi32(A, imm) \
+ ((__m128i)__builtin_ia32_extracti32x4_mask( \
+ (__v16si)(__m512i)(A), (int)(imm), (__v4si)_mm_setzero_si128(), \
+ (__mmask8) - 1))
#define _mm512_mask_extracti32x4_epi32(W, U, A, imm) \
((__m128i)__builtin_ia32_extracti32x4_mask((__v16si)(__m512i)(A), (int)(imm), \
@@ -7120,10 +7120,10 @@ _mm512_mask_cvtepi64_storeu_epi16 (void *__P, __mmask8 __M, __m512i __A)
(__v4si)_mm_setzero_si128(), \
(__mmask8)(U)))
-#define _mm512_extracti64x4_epi64(A, imm) \
+#define _mm512_extracti64x4_epi64(A, imm) \
((__m256i)__builtin_ia32_extracti64x4_mask((__v8di)(__m512i)(A), (int)(imm), \
- (__v4di)_mm256_setzero_si256(), \
- (__mmask8)-1))
+ (__v4di)_mm256_setzero_si256(), \
+ (__mmask8) - 1))
#define _mm512_mask_extracti64x4_epi64(W, U, A, imm) \
((__m256i)__builtin_ia32_extracti64x4_mask((__v8di)(__m512i)(A), (int)(imm), \
diff --git a/clang/lib/Headers/avx512vldqintrin.h b/clang/lib/Headers/avx512vldqintrin.h
index 2d3c4b551..8aded1c47 100644
--- a/clang/lib/Headers/avx512vldqintrin.h
+++ b/clang/lib/Headers/avx512vldqintrin.h
@@ -1072,11 +1072,10 @@ _mm256_maskz_broadcast_i64x2 (__mmask8 __M, __m128i __A)
(__v4di)_mm256_setzero_si256());
}
-#define _mm256_extractf64x2_pd(A, imm) \
- ((__m128d)__builtin_ia32_extractf64x2_256_mask((__v4df)(__m256d)(A), \
- (int)(imm), \
- (__v2df)_mm_setzero_pd(), \
- (__mmask8)-1))
+#define _mm256_extractf64x2_pd(A, imm) \
+ ((__m128d)__builtin_ia32_extractf64x2_256_mask( \
+ (__v4df)(__m256d)(A), (int)(imm), (__v2df)_mm_setzero_pd(), \
+ (__mmask8) - 1))
#define _mm256_mask_extractf64x2_pd(W, U, A, imm) \
((__m128d)__builtin_ia32_extractf64x2_256_mask((__v4df)(__m256d)(A), \
@@ -1090,11 +1089,10 @@ _mm256_maskz_broadcast_i64x2 (__mmask8 __M, __m128i __A)
(__v2df)_mm_setzero_pd(), \
(__mmask8)(U)))
-#define _mm256_extracti64x2_epi64(A, imm) \
- ((__m128i)__builtin_ia32_extracti64x2_256_mask((__v4di)(__m256i)(A), \
- (int)(imm), \
- (__v2di)_mm_setzero_si128(), \
- (__mmask8)-1))
+#define _mm256_extracti64x2_epi64(A, imm) \
+ ((__m128i)__builtin_ia32_extracti64x2_256_mask( \
+ (__v4di)(__m256i)(A), (int)(imm), (__v2di)_mm_setzero_si128(), \
+ (__mmask8) - 1))
#define _mm256_mask_extracti64x2_epi64(W, U, A, imm) \
((__m128i)__builtin_ia32_extracti64x2_256_mask((__v4di)(__m256i)(A), \
diff --git a/clang/lib/Headers/avx512vlintrin.h b/clang/lib/Headers/avx512vlintrin.h
index 252fb1119..eefeb1dad 100644
--- a/clang/lib/Headers/avx512vlintrin.h
+++ b/clang/lib/Headers/avx512vlintrin.h
@@ -7606,11 +7606,10 @@ _mm256_mask_cvtepi64_storeu_epi16 (void * __P, __mmask8 __M, __m256i __A)
__builtin_ia32_pmovqw256mem_mask ((__v8hi *) __P, (__v4di) __A, __M);
}
-#define _mm256_extractf32x4_ps(A, imm) \
- ((__m128)__builtin_ia32_extractf32x4_256_mask((__v8sf)(__m256)(A), \
- (int)(imm), \
- (__v4sf)_mm_setzero_ps(), \
- (__mmask8)-1))
+#define _mm256_extractf32x4_ps(A, imm) \
+ ((__m128)__builtin_ia32_extractf32x4_256_mask( \
+ (__v8sf)(__m256)(A), (int)(imm), (__v4sf)_mm_setzero_ps(), \
+ (__mmask8) - 1))
#define _mm256_mask_extractf32x4_ps(W, U, A, imm) \
((__m128)__builtin_ia32_extractf32x4_256_mask((__v8sf)(__m256)(A), \
@@ -7624,11 +7623,10 @@ _mm256_mask_cvtepi64_storeu_epi16 (void * __P, __mmask8 __M, __m256i __A)
(__v4sf)_mm_setzero_ps(), \
(__mmask8)(U)))
-#define _mm256_extracti32x4_epi32(A, imm) \
- ((__m128i)__builtin_ia32_extracti32x4_256_mask((__v8si)(__m256i)(A), \
- (int)(imm), \
- (__v4si)_mm_setzero_si128(), \
- (__mmask8)-1))
+#define _mm256_extracti32x4_epi32(A, imm) \
+ ((__m128i)__builtin_ia32_extracti32x4_256_mask( \
+ (__v8si)(__m256i)(A), (int)(imm), (__v4si)_mm_setzero_si128(), \
+ (__mmask8) - 1))
#define _mm256_mask_extracti32x4_epi32(W, U, A, imm) \
((__m128i)__builtin_ia32_extracti32x4_256_mask((__v8si)(__m256i)(A), \
|
RKSimon
requested changes
Oct 9, 2025
| APSInt Val = popToAPSInt(S, Call->getArg(0)); | ||
| pushInteger(S, Val.reverseBits(), Call->getType()); | ||
| return true; | ||
| } |
There was a problem hiding this comment.
remove these - they have been removed from a recent patch
There was a problem hiding this comment.
Ah, I see — that makes sense. I must have messed up the merge at some point.
I’m still getting used to the rebase workflow, so mistakes like this happen sometimes.
I’ll fix it and clean up the patch soon.
There was a problem hiding this comment.
Out of curiosity, when merge conflicts like this happen, what’s the usual approach that people take?
Do most contributors usually rebase onto the latest upstream, or do they start a new PR from upstream/main and cherry-pick their commits?
There was a problem hiding this comment.
fixing the merge is the best approach, rebasing usually means a lot of the review comments get lost which can be terrible for continuing the review for some patches - often at the point its better not to rebase, create a new PR and refer to the old one for reference.
There was a problem hiding this comment.
Thanks for the guidance and your patience. This branch has become difficult to manage, so I'll open a fresh PR and try again.
Implements constexpr evaluation for: - _mm256_extracti128_si256 (AVX2, VEXTRACTI128) - _mm256_extractf128_ps - _mm256_extractf128_pd - _mm256_extractf128_si256 These now work correctly in constant expressions by extracting the appropriate 128-bit lane from a 256-bit vector.
std::realloc is declared there
…uiltins in InterpBuiltin
- Route AVX/AVX2 vextractf128/ extract128i256 to 2-arg extract helper.
- Route all AVX-512(VL/DQ) extract builtins to unified 4-arg masked helper:
* extractf32x4_{256,_}
* extractf32x8_
* extractf64x2_{256,512}
* extractf64x4_
* extracti32x4_{256,_}
* extracti32x8_
* extracti64x2_{256,512}
* extracti64x4_
- Implement mask/merge/all-ones(mask=plain)/maskz semantics.
- Initialize all elements in the destination vector.
NOTE:
Tests are not included yet. This patch wires up InterpBuiltin support only.
A follow-up patch will add constexpr tests under clang/test/AST/Interp/.
SeongjaeP
force-pushed
the
issue157712-avx-extract
branch
from
21f366b to
24e06be
Compare
RKSimon
requested changes
Oct 10, 2025
RKSimon
pushed a commit
that referenced
this pull request
Oct 20, 2025
… AVX/AVX512 subvector extraction intrinsics to be used in constexpr #157712 (#162836) **This PR supersedes and replaces PR #158853** The original branch diverged too far from the main branch, resulting in significant merge conflicts that were difficult to resolve cleanly. To provide a clean and reviewable history, this new PR was created by cherry-picking the necessary commits onto a fresh branch based on the latest `main`. --- *(Original Description)* This patch enables the use of AVX/AVX512 subvector extraction intrinsics within `constexpr` functions. This is achieved by implementing the evaluation logic for these intrinsics in `VectorExprEvaluator::VisitCallExpr` and `InterpretBuiltin`. The original discussion and review comments can be found in the previous pull request for context: #158853 Fixes #157712
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.
AVX/AVX512 extract intrinsics now support constexpr evaluation in both the AST evaluator and bytecode interpreter paths.
Fixes #157712