[WebAssembly] Add support for memcmp expansion#148298
Merged
badumbatish merged 8 commits intollvm:mainfrom Jul 20, 2025
Merged
Conversation
Member
|
@llvm/pr-subscribers-backend-webassembly Author: jjasmine (badumbatish) ChangesFixes #61400 Added test case in llvm/test/CodeGen/WebAssembly/memcmp-expand.ll Full diff: https://github.com/llvm/llvm-project/pull/148298.diff 4 Files Affected:
diff --git a/llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp b/llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp
index bf2e04caa0a61..a91a58db2a422 100644
--- a/llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp
+++ b/llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp
@@ -46,6 +46,10 @@ WebAssemblyTargetLowering::WebAssemblyTargetLowering(
: TargetLowering(TM), Subtarget(&STI) {
auto MVTPtr = Subtarget->hasAddr64() ? MVT::i64 : MVT::i32;
+ // Set the load count for memcmp expand optimization
+ MaxLoadsPerMemcmp = 3;
+ MaxLoadsPerMemcmpOptSize = 2;
+
// Booleans always contain 0 or 1.
setBooleanContents(ZeroOrOneBooleanContent);
// Except in SIMD vectors
diff --git a/llvm/lib/Target/WebAssembly/WebAssemblyTargetTransformInfo.cpp b/llvm/lib/Target/WebAssembly/WebAssemblyTargetTransformInfo.cpp
index 4f159996e4c6c..3686fce33f3ca 100644
--- a/llvm/lib/Target/WebAssembly/WebAssemblyTargetTransformInfo.cpp
+++ b/llvm/lib/Target/WebAssembly/WebAssemblyTargetTransformInfo.cpp
@@ -141,6 +141,16 @@ InstructionCost WebAssemblyTTIImpl::getCastInstrCost(
return BaseT::getCastInstrCost(Opcode, Dst, Src, CCH, CostKind, I);
}
+WebAssemblyTTIImpl::TTI::MemCmpExpansionOptions
+WebAssemblyTTIImpl::enableMemCmpExpansion(bool OptSize, bool IsZeroCmp) const {
+ TTI::MemCmpExpansionOptions Options;
+ // INFO: I'm not sure what determines this, setting 2 conservatively
+ Options.NumLoadsPerBlock = 2;
+ Options.LoadSizes.append({8, 4, 2, 1});
+ Options.MaxNumLoads = TLI->getMaxExpandSizeMemcmp(OptSize);
+ return Options;
+}
+
InstructionCost WebAssemblyTTIImpl::getMemoryOpCost(
unsigned Opcode, Type *Ty, Align Alignment, unsigned AddressSpace,
TTI::TargetCostKind CostKind, TTI::OperandValueInfo OpInfo,
diff --git a/llvm/lib/Target/WebAssembly/WebAssemblyTargetTransformInfo.h b/llvm/lib/Target/WebAssembly/WebAssemblyTargetTransformInfo.h
index d83b8d1f45dbd..c915eeb07d4fd 100644
--- a/llvm/lib/Target/WebAssembly/WebAssemblyTargetTransformInfo.h
+++ b/llvm/lib/Target/WebAssembly/WebAssemblyTargetTransformInfo.h
@@ -73,6 +73,10 @@ class WebAssemblyTTIImpl final : public BasicTTIImplBase<WebAssemblyTTIImpl> {
getCastInstrCost(unsigned Opcode, Type *Dst, Type *Src,
TTI::CastContextHint CCH, TTI::TargetCostKind CostKind,
const Instruction *I = nullptr) const override;
+
+ TTI::MemCmpExpansionOptions
+ enableMemCmpExpansion(bool OptSize, bool IsZeroCmp) const override;
+
InstructionCost getMemoryOpCost(
unsigned Opcode, Type *Src, Align Alignment, unsigned AddressSpace,
TTI::TargetCostKind CostKind,
diff --git a/llvm/test/CodeGen/WebAssembly/memcmp-expand.ll b/llvm/test/CodeGen/WebAssembly/memcmp-expand.ll
new file mode 100644
index 0000000000000..64476e3b0b844
--- /dev/null
+++ b/llvm/test/CodeGen/WebAssembly/memcmp-expand.ll
@@ -0,0 +1,146 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc < %s -O3 -verify-machineinstrs -disable-wasm-fallthrough-return-opt -wasm-disable-explicit-locals -wasm-keep-registers | FileCheck %s
+
+target triple = "wasm32-unknown-unknown"
+
+declare i32 @memcmp(ptr, ptr, i32)
+
+define i1 @memcmp_expand_3(ptr %a, ptr %b) {
+; CHECK-LABEL: memcmp_expand_3:
+; CHECK: .functype memcmp_expand_3 (i32, i32) -> (i32)
+; CHECK-NEXT: # %bb.0:
+; CHECK-NEXT: i32.load16_u $push7=, 0($0):p2align=0
+; CHECK-NEXT: i32.load16_u $push6=, 0($1):p2align=0
+; CHECK-NEXT: i32.xor $push8=, $pop7, $pop6
+; CHECK-NEXT: i32.const $push0=, 2
+; CHECK-NEXT: i32.add $push3=, $0, $pop0
+; CHECK-NEXT: i32.load8_u $push4=, 0($pop3)
+; CHECK-NEXT: i32.const $push13=, 2
+; CHECK-NEXT: i32.add $push1=, $1, $pop13
+; CHECK-NEXT: i32.load8_u $push2=, 0($pop1)
+; CHECK-NEXT: i32.xor $push5=, $pop4, $pop2
+; CHECK-NEXT: i32.or $push9=, $pop8, $pop5
+; CHECK-NEXT: i32.const $push10=, 65535
+; CHECK-NEXT: i32.and $push11=, $pop9, $pop10
+; CHECK-NEXT: i32.eqz $push12=, $pop11
+; CHECK-NEXT: return $pop12
+ %cmp_3 = call i32 @memcmp(ptr %a, ptr %b, i32 3)
+ %res = icmp eq i32 %cmp_3, 0
+ ret i1 %res
+}
+
+define i1 @memcmp_expand_5(ptr %a, ptr %b) {
+; CHECK-LABEL: memcmp_expand_5:
+; CHECK: .functype memcmp_expand_5 (i32, i32) -> (i32)
+; CHECK-NEXT: # %bb.0:
+; CHECK-NEXT: i32.load $push7=, 0($0):p2align=0
+; CHECK-NEXT: i32.load $push6=, 0($1):p2align=0
+; CHECK-NEXT: i32.xor $push8=, $pop7, $pop6
+; CHECK-NEXT: i32.const $push0=, 4
+; CHECK-NEXT: i32.add $push3=, $0, $pop0
+; CHECK-NEXT: i32.load8_u $push4=, 0($pop3)
+; CHECK-NEXT: i32.const $push11=, 4
+; CHECK-NEXT: i32.add $push1=, $1, $pop11
+; CHECK-NEXT: i32.load8_u $push2=, 0($pop1)
+; CHECK-NEXT: i32.xor $push5=, $pop4, $pop2
+; CHECK-NEXT: i32.or $push9=, $pop8, $pop5
+; CHECK-NEXT: i32.eqz $push10=, $pop9
+; CHECK-NEXT: return $pop10
+ %cmp_5 = call i32 @memcmp(ptr %a, ptr %b, i32 5)
+ %res = icmp eq i32 %cmp_5, 0
+ ret i1 %res
+}
+
+define i1 @memcmp_expand_7(ptr %a, ptr %b) {
+; CHECK-LABEL: memcmp_expand_7:
+; CHECK: .functype memcmp_expand_7 (i32, i32) -> (i32)
+; CHECK-NEXT: # %bb.0:
+; CHECK-NEXT: block
+; CHECK-NEXT: block
+; CHECK-NEXT: i32.load $push7=, 0($0):p2align=0
+; CHECK-NEXT: i32.load $push6=, 0($1):p2align=0
+; CHECK-NEXT: i32.xor $push8=, $pop7, $pop6
+; CHECK-NEXT: i32.const $push0=, 4
+; CHECK-NEXT: i32.add $push3=, $0, $pop0
+; CHECK-NEXT: i32.load16_u $push4=, 0($pop3):p2align=0
+; CHECK-NEXT: i32.const $push17=, 4
+; CHECK-NEXT: i32.add $push1=, $1, $pop17
+; CHECK-NEXT: i32.load16_u $push2=, 0($pop1):p2align=0
+; CHECK-NEXT: i32.xor $push5=, $pop4, $pop2
+; CHECK-NEXT: i32.or $push9=, $pop8, $pop5
+; CHECK-NEXT: br_if 0, $pop9 # 0: down to label1
+; CHECK-NEXT: # %bb.1: # %loadbb1
+; CHECK-NEXT: i32.const $2=, 0
+; CHECK-NEXT: i32.const $push10=, 6
+; CHECK-NEXT: i32.add $push13=, $0, $pop10
+; CHECK-NEXT: i32.load8_u $push14=, 0($pop13)
+; CHECK-NEXT: i32.const $push18=, 6
+; CHECK-NEXT: i32.add $push11=, $1, $pop18
+; CHECK-NEXT: i32.load8_u $push12=, 0($pop11)
+; CHECK-NEXT: i32.eq $push15=, $pop14, $pop12
+; CHECK-NEXT: br_if 1, $pop15 # 1: down to label0
+; CHECK-NEXT: .LBB2_2: # %res_block
+; CHECK-NEXT: end_block # label1:
+; CHECK-NEXT: i32.const $2=, 1
+; CHECK-NEXT: .LBB2_3: # %endblock
+; CHECK-NEXT: end_block # label0:
+; CHECK-NEXT: i32.eqz $push16=, $2
+; CHECK-NEXT: return $pop16
+ %cmp_7 = call i32 @memcmp(ptr %a, ptr %b, i32 7)
+ %res = icmp eq i32 %cmp_7, 0
+ ret i1 %res
+}
+
+define i1 @memcmp_expand_2(ptr %a, ptr %b) {
+; CHECK-LABEL: memcmp_expand_2:
+; CHECK: .functype memcmp_expand_2 (i32, i32) -> (i32)
+; CHECK-NEXT: # %bb.0:
+; CHECK-NEXT: i32.load16_u $push1=, 0($0):p2align=0
+; CHECK-NEXT: i32.load16_u $push0=, 0($1):p2align=0
+; CHECK-NEXT: i32.eq $push2=, $pop1, $pop0
+; CHECK-NEXT: return $pop2
+ %cmp_2 = call i32 @memcmp(ptr %a, ptr %b, i32 2)
+ %res = icmp eq i32 %cmp_2, 0
+ ret i1 %res
+}
+
+
+define i1 @memcmp_expand_8(ptr %a, ptr %b) {
+; CHECK-LABEL: memcmp_expand_8:
+; CHECK: .functype memcmp_expand_8 (i32, i32) -> (i32)
+; CHECK-NEXT: # %bb.0:
+; CHECK-NEXT: i64.load $push1=, 0($0):p2align=0
+; CHECK-NEXT: i64.load $push0=, 0($1):p2align=0
+; CHECK-NEXT: i64.eq $push2=, $pop1, $pop0
+; CHECK-NEXT: return $pop2
+ %cmp_8 = call i32 @memcmp(ptr %a, ptr %b, i32 8)
+ %res = icmp eq i32 %cmp_8, 0
+ ret i1 %res
+}
+
+
+define i1 @memcmp_expand_16(ptr %a, ptr %b) {
+; CHECK-LABEL: memcmp_expand_16:
+; CHECK: .functype memcmp_expand_16 (i32, i32) -> (i32)
+; CHECK-NEXT: # %bb.0:
+; CHECK-NEXT: i64.load $push7=, 0($0):p2align=0
+; CHECK-NEXT: i64.load $push6=, 0($1):p2align=0
+; CHECK-NEXT: i64.xor $push8=, $pop7, $pop6
+; CHECK-NEXT: i32.const $push0=, 8
+; CHECK-NEXT: i32.add $push3=, $0, $pop0
+; CHECK-NEXT: i64.load $push4=, 0($pop3):p2align=0
+; CHECK-NEXT: i32.const $push11=, 8
+; CHECK-NEXT: i32.add $push1=, $1, $pop11
+; CHECK-NEXT: i64.load $push2=, 0($pop1):p2align=0
+; CHECK-NEXT: i64.xor $push5=, $pop4, $pop2
+; CHECK-NEXT: i64.or $push9=, $pop8, $pop5
+; CHECK-NEXT: i64.eqz $push10=, $pop9
+; CHECK-NEXT: return $pop10
+ %cmp_16 = call i32 @memcmp(ptr %a, ptr %b, i32 16)
+ %res = icmp eq i32 %cmp_16, 0
+ ret i1 %res
+}
+
+
+
+
|
Contributor
Author
|
@lukel97 hi Luke! would love your review :) |
dschuff
reviewed
Jul 11, 2025
lukel97
reviewed
Jul 12, 2025
badumbatish
changed the title
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
dschuff
approved these changes
Jul 17, 2025
lukel97
approved these changes
Jul 19, 2025
Contributor
lukel97
left a comment
lukel97
left a comment
There was a problem hiding this comment.
LGTM, some last little nits
| ret i1 %res | ||
| } | ||
|
|
||
| define i1 @memcmp_expand_2(ptr %a, ptr %b) { |
Contributor
There was a problem hiding this comment.
Can you add a test where the pointers are aligned and see if we omit the p2aligns?
So basically the same test as this but just with
define i1 @memcmp_expand_2_align(ptr align(2) %a, ptr align(2) %b) {
Contributor
Author
There was a problem hiding this comment.
yep looks like it works
define i1 @memcmp_expand_2(ptr %a, ptr %b) {
; CHECK-LABEL: memcmp_expand_2:
; CHECK: .functype memcmp_expand_2 (i32, i32) -> (i32)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: i32.load16_u $push1=, 0($0):p2align=0
; CHECK-NEXT: i32.load16_u $push0=, 0($1):p2align=0
; CHECK-NEXT: i32.eq $push2=, $pop1, $pop0
; CHECK-NEXT: return $pop2
%cmp_2 = call i32 @memcmp(ptr %a, ptr %b, i32 2)
%res = icmp eq i32 %cmp_2, 0
ret i1 %res
}
define i1 @memcmp_expand_2_align(ptr align(2) %a, ptr align(2) %b) {
; CHECK-LABEL: memcmp_expand_2_align:
; CHECK: .functype memcmp_expand_2_align (i32, i32) -> (i32)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: i32.load16_u $push1=, 0($0)
; CHECK-NEXT: i32.load16_u $push0=, 0($1)
; CHECK-NEXT: i32.eq $push2=, $pop1, $pop0
; CHECK-NEXT: return $pop2
%cmp_2 = call i32 @memcmp(ptr %a, ptr %b, i32 2)
%res = icmp eq i32 %cmp_2, 0
ret i1 %res
}
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/162/builds/27209 Here is the relevant piece of the build log for the reference |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/195/builds/12089 Here is the relevant piece of the build log for the reference |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/52/builds/9823 Here is the relevant piece of the build log for the reference |
This was referenced Jul 23, 2025
mahesh-attarde
pushed a commit
to mahesh-attarde/llvm-project
that referenced
this pull request
Jul 28, 2025
Fixes llvm#61400 Added test case in llvm/test/CodeGen/WebAssembly/memcmp-expand.ll
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 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.
Fixes #61400
Added test case in llvm/test/CodeGen/WebAssembly/memcmp-expand.ll