release/19.x: FastISel: Fix incorrectly using getPointerTy (#110465) #110490
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
|
@nikic What do you think about merging this PR to the release branch? |
|
@llvm/pr-subscribers-backend-x86 Author: None (llvmbot) ChangesBackport 81ba95c Requested by: @arsenm Full diff: https://github.com/llvm/llvm-project/pull/110490.diff 4 Files Affected:
diff --git a/llvm/include/llvm/CodeGen/FastISel.h b/llvm/include/llvm/CodeGen/FastISel.h
index 3cbc35400181dd..f3c4cc8d0511d4 100644
--- a/llvm/include/llvm/CodeGen/FastISel.h
+++ b/llvm/include/llvm/CodeGen/FastISel.h
@@ -275,7 +275,7 @@ class FastISel {
/// This is a wrapper around getRegForValue that also takes care of
/// truncating or sign-extending the given getelementptr index value.
- Register getRegForGEPIndex(const Value *Idx);
+ Register getRegForGEPIndex(MVT PtrVT, const Value *Idx);
/// We're checking to see if we can fold \p LI into \p FoldInst. Note
/// that we could have a sequence where multiple LLVM IR instructions are
diff --git a/llvm/lib/CodeGen/SelectionDAG/FastISel.cpp b/llvm/lib/CodeGen/SelectionDAG/FastISel.cpp
index ef9f7833551905..246acc7f405837 100644
--- a/llvm/lib/CodeGen/SelectionDAG/FastISel.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/FastISel.cpp
@@ -380,14 +380,13 @@ void FastISel::updateValueMap(const Value *I, Register Reg, unsigned NumRegs) {
}
}
-Register FastISel::getRegForGEPIndex(const Value *Idx) {
+Register FastISel::getRegForGEPIndex(MVT PtrVT, const Value *Idx) {
Register IdxN = getRegForValue(Idx);
if (!IdxN)
// Unhandled operand. Halt "fast" selection and bail.
return Register();
// If the index is smaller or larger than intptr_t, truncate or extend it.
- MVT PtrVT = TLI.getPointerTy(DL);
EVT IdxVT = EVT::getEVT(Idx->getType(), /*HandleUnknown=*/false);
if (IdxVT.bitsLT(PtrVT)) {
IdxN = fastEmit_r(IdxVT.getSimpleVT(), PtrVT, ISD::SIGN_EXTEND, IdxN);
@@ -543,7 +542,8 @@ bool FastISel::selectGetElementPtr(const User *I) {
uint64_t TotalOffs = 0;
// FIXME: What's a good SWAG number for MaxOffs?
uint64_t MaxOffs = 2048;
- MVT VT = TLI.getPointerTy(DL);
+ MVT VT = TLI.getValueType(DL, I->getType()).getSimpleVT();
+
for (gep_type_iterator GTI = gep_type_begin(I), E = gep_type_end(I);
GTI != E; ++GTI) {
const Value *Idx = GTI.getOperand();
@@ -584,7 +584,7 @@ bool FastISel::selectGetElementPtr(const User *I) {
// N = N + Idx * ElementSize;
uint64_t ElementSize = GTI.getSequentialElementStride(DL);
- Register IdxN = getRegForGEPIndex(Idx);
+ Register IdxN = getRegForGEPIndex(VT, Idx);
if (!IdxN) // Unhandled operand. Halt "fast" selection and bail.
return false;
diff --git a/llvm/lib/Target/X86/X86FastISel.cpp b/llvm/lib/Target/X86/X86FastISel.cpp
index 2eae155956368f..5d594bd54fbfc4 100644
--- a/llvm/lib/Target/X86/X86FastISel.cpp
+++ b/llvm/lib/Target/X86/X86FastISel.cpp
@@ -902,6 +902,8 @@ bool X86FastISel::X86SelectAddress(const Value *V, X86AddressMode &AM) {
uint64_t Disp = (int32_t)AM.Disp;
unsigned IndexReg = AM.IndexReg;
unsigned Scale = AM.Scale;
+ MVT PtrVT = TLI.getValueType(DL, U->getType()).getSimpleVT();
+
gep_type_iterator GTI = gep_type_begin(U);
// Iterate through the indices, folding what we can. Constants can be
// folded, and one dynamic index can be handled, if the scale is supported.
@@ -937,7 +939,7 @@ bool X86FastISel::X86SelectAddress(const Value *V, X86AddressMode &AM) {
(S == 1 || S == 2 || S == 4 || S == 8)) {
// Scaled-index addressing.
Scale = S;
- IndexReg = getRegForGEPIndex(Op);
+ IndexReg = getRegForGEPIndex(PtrVT, Op);
if (IndexReg == 0)
return false;
break;
diff --git a/llvm/test/CodeGen/X86/issue56055.ll b/llvm/test/CodeGen/X86/issue56055.ll
new file mode 100644
index 00000000000000..27eaf13e3b00be
--- /dev/null
+++ b/llvm/test/CodeGen/X86/issue56055.ll
@@ -0,0 +1,81 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -fast-isel < %s | FileCheck -check-prefixes=CHECK,FASTISEL %s
+; RUN: llc < %s | FileCheck -check-prefixes=CHECK,SDAG %s
+
+target datalayout = "e-m:w-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-windows-msvc"
+
+define void @issue56055(ptr addrspace(270) %ptr, ptr %out) {
+; CHECK-LABEL: issue56055:
+; CHECK: # %bb.0:
+; CHECK-NEXT: addl $2, %ecx
+; CHECK-NEXT: movl %ecx, (%rdx)
+; CHECK-NEXT: retq
+ %add.ptr = getelementptr inbounds i8, ptr addrspace(270) %ptr, i32 2
+ store ptr addrspace(270) %add.ptr, ptr %out
+ ret void
+}
+
+define void @issue56055_vector(<2 x ptr addrspace(270)> %ptr, ptr %out) {
+; CHECK-LABEL: issue56055_vector:
+; CHECK: # %bb.0:
+; CHECK-NEXT: movdqa (%rcx), %xmm0
+; CHECK-NEXT: paddd __xmm@00000000000000000000000200000002(%rip), %xmm0
+; CHECK-NEXT: movq %xmm0, (%rdx)
+; CHECK-NEXT: retq
+ %add.ptr = getelementptr inbounds i8, <2 x ptr addrspace(270)> %ptr, <2 x i32> <i32 2, i32 2>
+ store <2 x ptr addrspace(270)> %add.ptr, ptr %out
+ ret void
+}
+
+define void @issue56055_small_idx(ptr addrspace(270) %ptr, ptr %out, i16 %idx) {
+; CHECK-LABEL: issue56055_small_idx:
+; CHECK: # %bb.0:
+; CHECK-NEXT: movswl %r8w, %eax
+; CHECK-NEXT: addl %ecx, %eax
+; CHECK-NEXT: movl %eax, (%rdx)
+; CHECK-NEXT: retq
+ %add.ptr = getelementptr inbounds i8, ptr addrspace(270) %ptr, i16 %idx
+ store ptr addrspace(270) %add.ptr, ptr %out
+ ret void
+}
+
+define void @issue56055_small_idx_vector(<2 x ptr addrspace(270)> %ptr, ptr %out, <2 x i16> %idx) {
+; CHECK-LABEL: issue56055_small_idx_vector:
+; CHECK: # %bb.0:
+; CHECK-NEXT: pshuflw {{.*#+}} xmm0 = mem[0,0,2,1,4,5,6,7]
+; CHECK-NEXT: psrad $16, %xmm0
+; CHECK-NEXT: paddd (%rcx), %xmm0
+; CHECK-NEXT: movq %xmm0, (%rdx)
+; CHECK-NEXT: retq
+ %add.ptr = getelementptr inbounds i8, <2 x ptr addrspace(270)> %ptr, <2 x i16> %idx
+ store <2 x ptr addrspace(270)> %add.ptr, ptr %out
+ ret void
+}
+
+define void @issue56055_large_idx(ptr addrspace(270) %ptr, ptr %out, i64 %idx) {
+; CHECK-LABEL: issue56055_large_idx:
+; CHECK: # %bb.0:
+; CHECK-NEXT: addl %ecx, %r8d
+; CHECK-NEXT: movl %r8d, (%rdx)
+; CHECK-NEXT: retq
+ %add.ptr = getelementptr inbounds i8, ptr addrspace(270) %ptr, i64 %idx
+ store ptr addrspace(270) %add.ptr, ptr %out
+ ret void
+}
+
+define void @issue56055_large_idx_vector(<2 x ptr addrspace(270)> %ptr, ptr %out, <2 x i64> %idx) {
+; CHECK-LABEL: issue56055_large_idx_vector:
+; CHECK: # %bb.0:
+; CHECK-NEXT: pshufd {{.*#+}} xmm0 = mem[0,2,2,3]
+; CHECK-NEXT: paddd (%rcx), %xmm0
+; CHECK-NEXT: movq %xmm0, (%rdx)
+; CHECK-NEXT: retq
+ %add.ptr = getelementptr inbounds i8, <2 x ptr addrspace(270)> %ptr, <2 x i64> %idx
+ store <2 x ptr addrspace(270)> %add.ptr, ptr %out
+ ret void
+}
+
+;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+; FASTISEL: {{.*}}
+; SDAG: {{.*}}
|
|
@llvm/pr-subscribers-llvm-selectiondag Author: None (llvmbot) ChangesBackport 81ba95c Requested by: @arsenm Full diff: https://github.com/llvm/llvm-project/pull/110490.diff 4 Files Affected:
diff --git a/llvm/include/llvm/CodeGen/FastISel.h b/llvm/include/llvm/CodeGen/FastISel.h
index 3cbc35400181dd..f3c4cc8d0511d4 100644
--- a/llvm/include/llvm/CodeGen/FastISel.h
+++ b/llvm/include/llvm/CodeGen/FastISel.h
@@ -275,7 +275,7 @@ class FastISel {
/// This is a wrapper around getRegForValue that also takes care of
/// truncating or sign-extending the given getelementptr index value.
- Register getRegForGEPIndex(const Value *Idx);
+ Register getRegForGEPIndex(MVT PtrVT, const Value *Idx);
/// We're checking to see if we can fold \p LI into \p FoldInst. Note
/// that we could have a sequence where multiple LLVM IR instructions are
diff --git a/llvm/lib/CodeGen/SelectionDAG/FastISel.cpp b/llvm/lib/CodeGen/SelectionDAG/FastISel.cpp
index ef9f7833551905..246acc7f405837 100644
--- a/llvm/lib/CodeGen/SelectionDAG/FastISel.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/FastISel.cpp
@@ -380,14 +380,13 @@ void FastISel::updateValueMap(const Value *I, Register Reg, unsigned NumRegs) {
}
}
-Register FastISel::getRegForGEPIndex(const Value *Idx) {
+Register FastISel::getRegForGEPIndex(MVT PtrVT, const Value *Idx) {
Register IdxN = getRegForValue(Idx);
if (!IdxN)
// Unhandled operand. Halt "fast" selection and bail.
return Register();
// If the index is smaller or larger than intptr_t, truncate or extend it.
- MVT PtrVT = TLI.getPointerTy(DL);
EVT IdxVT = EVT::getEVT(Idx->getType(), /*HandleUnknown=*/false);
if (IdxVT.bitsLT(PtrVT)) {
IdxN = fastEmit_r(IdxVT.getSimpleVT(), PtrVT, ISD::SIGN_EXTEND, IdxN);
@@ -543,7 +542,8 @@ bool FastISel::selectGetElementPtr(const User *I) {
uint64_t TotalOffs = 0;
// FIXME: What's a good SWAG number for MaxOffs?
uint64_t MaxOffs = 2048;
- MVT VT = TLI.getPointerTy(DL);
+ MVT VT = TLI.getValueType(DL, I->getType()).getSimpleVT();
+
for (gep_type_iterator GTI = gep_type_begin(I), E = gep_type_end(I);
GTI != E; ++GTI) {
const Value *Idx = GTI.getOperand();
@@ -584,7 +584,7 @@ bool FastISel::selectGetElementPtr(const User *I) {
// N = N + Idx * ElementSize;
uint64_t ElementSize = GTI.getSequentialElementStride(DL);
- Register IdxN = getRegForGEPIndex(Idx);
+ Register IdxN = getRegForGEPIndex(VT, Idx);
if (!IdxN) // Unhandled operand. Halt "fast" selection and bail.
return false;
diff --git a/llvm/lib/Target/X86/X86FastISel.cpp b/llvm/lib/Target/X86/X86FastISel.cpp
index 2eae155956368f..5d594bd54fbfc4 100644
--- a/llvm/lib/Target/X86/X86FastISel.cpp
+++ b/llvm/lib/Target/X86/X86FastISel.cpp
@@ -902,6 +902,8 @@ bool X86FastISel::X86SelectAddress(const Value *V, X86AddressMode &AM) {
uint64_t Disp = (int32_t)AM.Disp;
unsigned IndexReg = AM.IndexReg;
unsigned Scale = AM.Scale;
+ MVT PtrVT = TLI.getValueType(DL, U->getType()).getSimpleVT();
+
gep_type_iterator GTI = gep_type_begin(U);
// Iterate through the indices, folding what we can. Constants can be
// folded, and one dynamic index can be handled, if the scale is supported.
@@ -937,7 +939,7 @@ bool X86FastISel::X86SelectAddress(const Value *V, X86AddressMode &AM) {
(S == 1 || S == 2 || S == 4 || S == 8)) {
// Scaled-index addressing.
Scale = S;
- IndexReg = getRegForGEPIndex(Op);
+ IndexReg = getRegForGEPIndex(PtrVT, Op);
if (IndexReg == 0)
return false;
break;
diff --git a/llvm/test/CodeGen/X86/issue56055.ll b/llvm/test/CodeGen/X86/issue56055.ll
new file mode 100644
index 00000000000000..27eaf13e3b00be
--- /dev/null
+++ b/llvm/test/CodeGen/X86/issue56055.ll
@@ -0,0 +1,81 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -fast-isel < %s | FileCheck -check-prefixes=CHECK,FASTISEL %s
+; RUN: llc < %s | FileCheck -check-prefixes=CHECK,SDAG %s
+
+target datalayout = "e-m:w-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-windows-msvc"
+
+define void @issue56055(ptr addrspace(270) %ptr, ptr %out) {
+; CHECK-LABEL: issue56055:
+; CHECK: # %bb.0:
+; CHECK-NEXT: addl $2, %ecx
+; CHECK-NEXT: movl %ecx, (%rdx)
+; CHECK-NEXT: retq
+ %add.ptr = getelementptr inbounds i8, ptr addrspace(270) %ptr, i32 2
+ store ptr addrspace(270) %add.ptr, ptr %out
+ ret void
+}
+
+define void @issue56055_vector(<2 x ptr addrspace(270)> %ptr, ptr %out) {
+; CHECK-LABEL: issue56055_vector:
+; CHECK: # %bb.0:
+; CHECK-NEXT: movdqa (%rcx), %xmm0
+; CHECK-NEXT: paddd __xmm@00000000000000000000000200000002(%rip), %xmm0
+; CHECK-NEXT: movq %xmm0, (%rdx)
+; CHECK-NEXT: retq
+ %add.ptr = getelementptr inbounds i8, <2 x ptr addrspace(270)> %ptr, <2 x i32> <i32 2, i32 2>
+ store <2 x ptr addrspace(270)> %add.ptr, ptr %out
+ ret void
+}
+
+define void @issue56055_small_idx(ptr addrspace(270) %ptr, ptr %out, i16 %idx) {
+; CHECK-LABEL: issue56055_small_idx:
+; CHECK: # %bb.0:
+; CHECK-NEXT: movswl %r8w, %eax
+; CHECK-NEXT: addl %ecx, %eax
+; CHECK-NEXT: movl %eax, (%rdx)
+; CHECK-NEXT: retq
+ %add.ptr = getelementptr inbounds i8, ptr addrspace(270) %ptr, i16 %idx
+ store ptr addrspace(270) %add.ptr, ptr %out
+ ret void
+}
+
+define void @issue56055_small_idx_vector(<2 x ptr addrspace(270)> %ptr, ptr %out, <2 x i16> %idx) {
+; CHECK-LABEL: issue56055_small_idx_vector:
+; CHECK: # %bb.0:
+; CHECK-NEXT: pshuflw {{.*#+}} xmm0 = mem[0,0,2,1,4,5,6,7]
+; CHECK-NEXT: psrad $16, %xmm0
+; CHECK-NEXT: paddd (%rcx), %xmm0
+; CHECK-NEXT: movq %xmm0, (%rdx)
+; CHECK-NEXT: retq
+ %add.ptr = getelementptr inbounds i8, <2 x ptr addrspace(270)> %ptr, <2 x i16> %idx
+ store <2 x ptr addrspace(270)> %add.ptr, ptr %out
+ ret void
+}
+
+define void @issue56055_large_idx(ptr addrspace(270) %ptr, ptr %out, i64 %idx) {
+; CHECK-LABEL: issue56055_large_idx:
+; CHECK: # %bb.0:
+; CHECK-NEXT: addl %ecx, %r8d
+; CHECK-NEXT: movl %r8d, (%rdx)
+; CHECK-NEXT: retq
+ %add.ptr = getelementptr inbounds i8, ptr addrspace(270) %ptr, i64 %idx
+ store ptr addrspace(270) %add.ptr, ptr %out
+ ret void
+}
+
+define void @issue56055_large_idx_vector(<2 x ptr addrspace(270)> %ptr, ptr %out, <2 x i64> %idx) {
+; CHECK-LABEL: issue56055_large_idx_vector:
+; CHECK: # %bb.0:
+; CHECK-NEXT: pshufd {{.*#+}} xmm0 = mem[0,2,2,3]
+; CHECK-NEXT: paddd (%rcx), %xmm0
+; CHECK-NEXT: movq %xmm0, (%rdx)
+; CHECK-NEXT: retq
+ %add.ptr = getelementptr inbounds i8, <2 x ptr addrspace(270)> %ptr, <2 x i64> %idx
+ store <2 x ptr addrspace(270)> %add.ptr, ptr %out
+ ret void
+}
+
+;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+; FASTISEL: {{.*}}
+; SDAG: {{.*}}
|
|
@nikic what do you think? release is today - should we get this in before that? |
This was using the default address space instead of the correct one. Fixes llvm#56055 (cherry picked from commit 81ba95c)
|
@tru This looks like an ABI break to me, but not entirely sure (I think CodeGen is part of the ABI, unlike Target). The abi-compare job needs the correct version to be set, which only happened in fae11d4, so the previous job runs would incorrectly pass (they would only check C ABI). I've rebased this now to see what the job says with the correct version... |
|
Seems like abi-compare agrees with you and it's a breakage. I couldn't figure out how to download the report (why is github hiding that?). |
For some reason, github only makes the artifacts available once all jobs have completed. |
nikic
reviewed
Oct 1, 2024
| @@ -275,7 +275,7 @@ class FastISel { | |||
|
|
|||
| /// This is a wrapper around getRegForValue that also takes care of | |||
| /// truncating or sign-extending the given getelementptr index value. | |||
| Register getRegForGEPIndex(const Value *Idx); | |||
| Register getRegForGEPIndex(MVT PtrVT, const Value *Idx); | |||
There was a problem hiding this comment.
The way to avoid the ABI break would be to keep the old Register getRegForGEPIndex(const Value *Idx) overload and make it pass TLI.getPointerTy(DL) for the first param to the other overload.
|
Would it be prudent to wait for a fix and in that case accept that into 19.1.2 instead? No need to rush this right? |
|
@tru Yes, this is a fix for a long-standing issue, no need for urgency here. |
|
New version in #110827 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Backport 81ba95c
Requested by: @arsenm