-
Notifications
You must be signed in to change notification settings - Fork 15.4k
[X86] getTargetConstantBitsFromNode - early-out if the element bitsize doesn't align with the source bitsize #150184
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
…e doesn't align with the source bitsize As we use getTargetConstantBitsFromNode in a wider variety of places we can't guarantee that all the sources match (or are legal) anymore - better to early out than assert. Fixes llvm#150117
|
@llvm/pr-subscribers-backend-x86 Author: Simon Pilgrim (RKSimon) ChangesAs we use getTargetConstantBitsFromNode in a wider variety of places we can't guarantee that all the sources match (or are legal) anymore - better to early out than assert. Fixes #150117 Full diff: https://github.com/llvm/llvm-project/pull/150184.diff 2 Files Affected:
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 568a8c4cfed79..11ab8dc685ea8 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -5001,9 +5001,12 @@ static bool getTargetConstantBitsFromNode(SDValue Op, unsigned EltSizeInBits,
EVT VT = Op.getValueType();
unsigned SizeInBits = VT.getSizeInBits();
- assert((SizeInBits % EltSizeInBits) == 0 && "Can't split constant!");
unsigned NumElts = SizeInBits / EltSizeInBits;
+ // Can't split constant.
+ if ((SizeInBits % EltSizeInBits) != 0)
+ return false;
+
// Bitcast a source array of element bits to the target size.
auto CastBitData = [&](APInt &UndefSrcElts, ArrayRef<APInt> SrcEltBits) {
unsigned NumSrcElts = UndefSrcElts.getBitWidth();
diff --git a/llvm/test/CodeGen/X86/vec_extract.ll b/llvm/test/CodeGen/X86/vec_extract.ll
index 087cd30abee9b..9bd38db3e9c4c 100644
--- a/llvm/test/CodeGen/X86/vec_extract.ll
+++ b/llvm/test/CodeGen/X86/vec_extract.ll
@@ -104,6 +104,72 @@ entry:
}
declare <2 x double> @foo()
+define i64 @pr150117(<31 x i8> %a0) nounwind {
+; X86-LABEL: pr150117:
+; X86: # %bb.0:
+; X86-NEXT: pushl %ebx
+; X86-NEXT: pushl %edi
+; X86-NEXT: pushl %esi
+; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT: movzbl {{[0-9]+}}(%esp), %esi
+; X86-NEXT: movzbl {{[0-9]+}}(%esp), %edi
+; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ebx
+; X86-NEXT: movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT: shll $8, %edx
+; X86-NEXT: orl %ebx, %edx
+; X86-NEXT: shll $8, %edi
+; X86-NEXT: orl %esi, %edi
+; X86-NEXT: shll $16, %ecx
+; X86-NEXT: orl %edi, %ecx
+; X86-NEXT: movl {{[0-9]+}}(%esp), %esi
+; X86-NEXT: shll $24, %esi
+; X86-NEXT: orl %ecx, %esi
+; X86-NEXT: movd %esi, %xmm0
+; X86-NEXT: pinsrw $2, %edx, %xmm0
+; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT: shll $8, %ecx
+; X86-NEXT: orl %eax, %ecx
+; X86-NEXT: pinsrw $3, %ecx, %xmm0
+; X86-NEXT: movd %xmm0, %eax
+; X86-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,1,1]
+; X86-NEXT: movd %xmm0, %edx
+; X86-NEXT: popl %esi
+; X86-NEXT: popl %edi
+; X86-NEXT: popl %ebx
+; X86-NEXT: retl
+;
+; X64-LABEL: pr150117:
+; X64: # %bb.0:
+; X64-NEXT: movzbl {{[0-9]+}}(%rsp), %eax
+; X64-NEXT: movzbl {{[0-9]+}}(%rsp), %ecx
+; X64-NEXT: movzbl {{[0-9]+}}(%rsp), %edx
+; X64-NEXT: movzbl {{[0-9]+}}(%rsp), %esi
+; X64-NEXT: movzbl {{[0-9]+}}(%rsp), %edi
+; X64-NEXT: movl {{[0-9]+}}(%rsp), %r8d
+; X64-NEXT: shll $8, %r8d
+; X64-NEXT: orl %edi, %r8d
+; X64-NEXT: shll $8, %esi
+; X64-NEXT: orl %edx, %esi
+; X64-NEXT: shll $16, %ecx
+; X64-NEXT: orl %esi, %ecx
+; X64-NEXT: movl {{[0-9]+}}(%rsp), %edx
+; X64-NEXT: shll $24, %edx
+; X64-NEXT: orl %ecx, %edx
+; X64-NEXT: movd %edx, %xmm0
+; X64-NEXT: pinsrw $2, %r8d, %xmm0
+; X64-NEXT: movl {{[0-9]+}}(%rsp), %ecx
+; X64-NEXT: shll $8, %ecx
+; X64-NEXT: orl %eax, %ecx
+; X64-NEXT: pinsrw $3, %ecx, %xmm0
+; X64-NEXT: movq %xmm0, %rax
+; X64-NEXT: retq
+ %shuffle = shufflevector <31 x i8> %a0, <31 x i8> zeroinitializer, <32 x i32> <i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+ %bitcast = bitcast <32 x i8> %shuffle to <4 x i64>
+ %elt = extractelement <4 x i64> %bitcast, i64 0
+ ret i64 %elt
+}
+
; OSS-Fuzz #15662
; https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=15662
define <4 x i32> @ossfuzz15662(ptr %in) {
|
phoebewang
left a comment
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM.
…e doesn't align with the source bitsize (llvm#150184) As we use getTargetConstantBitsFromNode in a wider variety of places we can't guarantee that all the sources match (or are legal) anymore - better to early out than assert. Fixes llvm#150117 (cherry picked from commit 3345582)
…e doesn't align with the source bitsize (llvm#150184) As we use getTargetConstantBitsFromNode in a wider variety of places we can't guarantee that all the sources match (or are legal) anymore - better to early out than assert. Fixes llvm#150117
As we use getTargetConstantBitsFromNode in a wider variety of places we can't guarantee that all the sources match (or are legal) anymore - better to early out than assert.
Fixes #150117