[llvm] [msan] Handle Arm NEON floating-point min/max (vector) (PR #125778)
via llvm-commits
llvm-commits at lists.llvm.org
Tue Feb 4 15:06:35 PST 2025
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-compiler-rt-sanitizer
Author: Thurston Dang (thurstond)
<details>
<summary>Changes</summary>
Apply handleVectorReduceIntrinsic() to Intrinsic::aarch64_neon_f{min,max}(mn)?v. Previously, these were unknown intrinsics handled suboptimally by visitInstruction.
Updates the tests from https://github.com/llvm/llvm-project/pull/125729.
---
Full diff: https://github.com/llvm/llvm-project/pull/125778.diff
2 Files Affected:
- (modified) llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp (+7)
- (modified) llvm/test/Instrumentation/MemorySanitizer/AArch64/arm64-fminv.ll (+24-105)
``````````diff
diff --git a/llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp b/llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp
index f3f2e5041fb1d3..2771f806fa227e 100644
--- a/llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp
+++ b/llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp
@@ -4367,6 +4367,13 @@ struct MemorySanitizerVisitor : public InstVisitor<MemorySanitizerVisitor> {
case Intrinsic::aarch64_neon_faddv:
case Intrinsic::aarch64_neon_saddv:
case Intrinsic::aarch64_neon_uaddv:
+ // Floating-point min/max (vector)
+ // The f{min,max}"nm"v variants handle NaN differently than f{min,max}v,
+ // but our shadow propagation is the same.
+ case Intrinsic::aarch64_neon_fmaxv:
+ case Intrinsic::aarch64_neon_fminv:
+ case Intrinsic::aarch64_neon_fmaxnmv:
+ case Intrinsic::aarch64_neon_fminnmv:
handleVectorReduceIntrinsic(I);
break;
case Intrinsic::vector_reduce_fadd:
diff --git a/llvm/test/Instrumentation/MemorySanitizer/AArch64/arm64-fminv.ll b/llvm/test/Instrumentation/MemorySanitizer/AArch64/arm64-fminv.ll
index 508decfc3c6d2f..ff5784ada8c482 100644
--- a/llvm/test/Instrumentation/MemorySanitizer/AArch64/arm64-fminv.ll
+++ b/llvm/test/Instrumentation/MemorySanitizer/AArch64/arm64-fminv.ll
@@ -2,12 +2,6 @@
; RUN: opt < %s -passes=msan -S | FileCheck %s
;
; Forked from llvm/test/CodeGen/AArch64/arm64-fminv.ll
-;
-; Currently handled (suboptimally) by handleUnknownInstruction:
-; - llvm.aarch64.neon.fmaxv
-; - llvm.aarch64.neon.fminv
-; - llvm.aarch64.neon.fmaxnmv
-; - llvm.aarch64.neon.fminnmv
target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
target triple = "aarch64--linux-android9001"
@@ -17,15 +11,9 @@ define float @test_fminv_v2f32(<2 x float> %in) #0 {
; CHECK-SAME: <2 x float> [[IN:%.*]]) #[[ATTR0:[0-9]+]] {
; CHECK-NEXT: [[TMP1:%.*]] = load <2 x i32>, ptr @__msan_param_tls, align 8
; CHECK-NEXT: call void @llvm.donothing()
-; CHECK-NEXT: [[TMP2:%.*]] = bitcast <2 x i32> [[TMP1]] to i64
-; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i64 [[TMP2]], 0
-; CHECK-NEXT: br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1:![0-9]+]]
-; CHECK: 3:
-; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR3:[0-9]+]]
-; CHECK-NEXT: unreachable
-; CHECK: 4:
+; CHECK-NEXT: [[TMP2:%.*]] = call i32 @llvm.vector.reduce.or.v2i32(<2 x i32> [[TMP1]])
; CHECK-NEXT: [[MIN:%.*]] = call float @llvm.aarch64.neon.fminv.f32.v2f32(<2 x float> [[IN]])
-; CHECK-NEXT: store i32 0, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT: store i32 [[TMP2]], ptr @__msan_retval_tls, align 8
; CHECK-NEXT: ret float [[MIN]]
;
%min = call float @llvm.aarch64.neon.fminv.f32.v2f32(<2 x float> %in)
@@ -37,15 +25,9 @@ define float @test_fminv_v4f32(<4 x float> %in) #0 {
; CHECK-SAME: <4 x float> [[IN:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8
; CHECK-NEXT: call void @llvm.donothing()
-; CHECK-NEXT: [[TMP2:%.*]] = bitcast <4 x i32> [[TMP1]] to i128
-; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i128 [[TMP2]], 0
-; CHECK-NEXT: br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
-; CHECK: 3:
-; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR3]]
-; CHECK-NEXT: unreachable
-; CHECK: 4:
+; CHECK-NEXT: [[TMP2:%.*]] = call i32 @llvm.vector.reduce.or.v4i32(<4 x i32> [[TMP1]])
; CHECK-NEXT: [[MIN:%.*]] = call float @llvm.aarch64.neon.fminv.f32.v4f32(<4 x float> [[IN]])
-; CHECK-NEXT: store i32 0, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT: store i32 [[TMP2]], ptr @__msan_retval_tls, align 8
; CHECK-NEXT: ret float [[MIN]]
;
%min = call float @llvm.aarch64.neon.fminv.f32.v4f32(<4 x float> %in)
@@ -57,15 +39,9 @@ define double @test_fminv_v2f64(<2 x double> %in) #0 {
; CHECK-SAME: <2 x double> [[IN:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[TMP1:%.*]] = load <2 x i64>, ptr @__msan_param_tls, align 8
; CHECK-NEXT: call void @llvm.donothing()
-; CHECK-NEXT: [[TMP2:%.*]] = bitcast <2 x i64> [[TMP1]] to i128
-; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i128 [[TMP2]], 0
-; CHECK-NEXT: br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
-; CHECK: 3:
-; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR3]]
-; CHECK-NEXT: unreachable
-; CHECK: 4:
+; CHECK-NEXT: [[TMP2:%.*]] = call i64 @llvm.vector.reduce.or.v2i64(<2 x i64> [[TMP1]])
; CHECK-NEXT: [[MIN:%.*]] = call double @llvm.aarch64.neon.fminv.f64.v2f64(<2 x double> [[IN]])
-; CHECK-NEXT: store i64 0, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT: store i64 [[TMP2]], ptr @__msan_retval_tls, align 8
; CHECK-NEXT: ret double [[MIN]]
;
%min = call double @llvm.aarch64.neon.fminv.f64.v2f64(<2 x double> %in)
@@ -81,15 +57,9 @@ define float @test_fmaxv_v2f32(<2 x float> %in) #0 {
; CHECK-SAME: <2 x float> [[IN:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[TMP1:%.*]] = load <2 x i32>, ptr @__msan_param_tls, align 8
; CHECK-NEXT: call void @llvm.donothing()
-; CHECK-NEXT: [[TMP2:%.*]] = bitcast <2 x i32> [[TMP1]] to i64
-; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i64 [[TMP2]], 0
-; CHECK-NEXT: br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
-; CHECK: 3:
-; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR3]]
-; CHECK-NEXT: unreachable
-; CHECK: 4:
+; CHECK-NEXT: [[TMP2:%.*]] = call i32 @llvm.vector.reduce.or.v2i32(<2 x i32> [[TMP1]])
; CHECK-NEXT: [[MAX:%.*]] = call float @llvm.aarch64.neon.fmaxv.f32.v2f32(<2 x float> [[IN]])
-; CHECK-NEXT: store i32 0, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT: store i32 [[TMP2]], ptr @__msan_retval_tls, align 8
; CHECK-NEXT: ret float [[MAX]]
;
%max = call float @llvm.aarch64.neon.fmaxv.f32.v2f32(<2 x float> %in)
@@ -101,15 +71,9 @@ define float @test_fmaxv_v4f32(<4 x float> %in) #0 {
; CHECK-SAME: <4 x float> [[IN:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8
; CHECK-NEXT: call void @llvm.donothing()
-; CHECK-NEXT: [[TMP2:%.*]] = bitcast <4 x i32> [[TMP1]] to i128
-; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i128 [[TMP2]], 0
-; CHECK-NEXT: br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
-; CHECK: 3:
-; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR3]]
-; CHECK-NEXT: unreachable
-; CHECK: 4:
+; CHECK-NEXT: [[TMP2:%.*]] = call i32 @llvm.vector.reduce.or.v4i32(<4 x i32> [[TMP1]])
; CHECK-NEXT: [[MAX:%.*]] = call float @llvm.aarch64.neon.fmaxv.f32.v4f32(<4 x float> [[IN]])
-; CHECK-NEXT: store i32 0, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT: store i32 [[TMP2]], ptr @__msan_retval_tls, align 8
; CHECK-NEXT: ret float [[MAX]]
;
%max = call float @llvm.aarch64.neon.fmaxv.f32.v4f32(<4 x float> %in)
@@ -121,15 +85,9 @@ define double @test_fmaxv_v2f64(<2 x double> %shareholder_value) #0 {
; CHECK-SAME: <2 x double> [[SHAREHOLDER_VALUE:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[TMP1:%.*]] = load <2 x i64>, ptr @__msan_param_tls, align 8
; CHECK-NEXT: call void @llvm.donothing()
-; CHECK-NEXT: [[TMP2:%.*]] = bitcast <2 x i64> [[TMP1]] to i128
-; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i128 [[TMP2]], 0
-; CHECK-NEXT: br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
-; CHECK: 3:
-; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR3]]
-; CHECK-NEXT: unreachable
-; CHECK: 4:
+; CHECK-NEXT: [[TMP2:%.*]] = call i64 @llvm.vector.reduce.or.v2i64(<2 x i64> [[TMP1]])
; CHECK-NEXT: [[MAX:%.*]] = call double @llvm.aarch64.neon.fmaxv.f64.v2f64(<2 x double> [[SHAREHOLDER_VALUE]])
-; CHECK-NEXT: store i64 0, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT: store i64 [[TMP2]], ptr @__msan_retval_tls, align 8
; CHECK-NEXT: ret double [[MAX]]
;
%max_sv = call double @llvm.aarch64.neon.fmaxv.f64.v2f64(<2 x double> %shareholder_value)
@@ -145,15 +103,9 @@ define float @test_fminnmv_v2f32(<2 x float> %in) #0 {
; CHECK-SAME: <2 x float> [[IN:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[TMP1:%.*]] = load <2 x i32>, ptr @__msan_param_tls, align 8
; CHECK-NEXT: call void @llvm.donothing()
-; CHECK-NEXT: [[TMP2:%.*]] = bitcast <2 x i32> [[TMP1]] to i64
-; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i64 [[TMP2]], 0
-; CHECK-NEXT: br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
-; CHECK: 3:
-; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR3]]
-; CHECK-NEXT: unreachable
-; CHECK: 4:
+; CHECK-NEXT: [[TMP2:%.*]] = call i32 @llvm.vector.reduce.or.v2i32(<2 x i32> [[TMP1]])
; CHECK-NEXT: [[MINNM:%.*]] = call float @llvm.aarch64.neon.fminnmv.f32.v2f32(<2 x float> [[IN]])
-; CHECK-NEXT: store i32 0, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT: store i32 [[TMP2]], ptr @__msan_retval_tls, align 8
; CHECK-NEXT: ret float [[MINNM]]
;
%minnm = call float @llvm.aarch64.neon.fminnmv.f32.v2f32(<2 x float> %in)
@@ -165,15 +117,9 @@ define float @test_fminnmv_v4f32(<4 x float> %in) #0 {
; CHECK-SAME: <4 x float> [[IN:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8
; CHECK-NEXT: call void @llvm.donothing()
-; CHECK-NEXT: [[TMP2:%.*]] = bitcast <4 x i32> [[TMP1]] to i128
-; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i128 [[TMP2]], 0
-; CHECK-NEXT: br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
-; CHECK: 3:
-; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR3]]
-; CHECK-NEXT: unreachable
-; CHECK: 4:
+; CHECK-NEXT: [[TMP2:%.*]] = call i32 @llvm.vector.reduce.or.v4i32(<4 x i32> [[TMP1]])
; CHECK-NEXT: [[MINNM:%.*]] = call float @llvm.aarch64.neon.fminnmv.f32.v4f32(<4 x float> [[IN]])
-; CHECK-NEXT: store i32 0, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT: store i32 [[TMP2]], ptr @__msan_retval_tls, align 8
; CHECK-NEXT: ret float [[MINNM]]
;
%minnm = call float @llvm.aarch64.neon.fminnmv.f32.v4f32(<4 x float> %in)
@@ -185,15 +131,9 @@ define double @test_fminnmv_v2f64(<2 x double> %in) #0 {
; CHECK-SAME: <2 x double> [[IN:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[TMP1:%.*]] = load <2 x i64>, ptr @__msan_param_tls, align 8
; CHECK-NEXT: call void @llvm.donothing()
-; CHECK-NEXT: [[TMP2:%.*]] = bitcast <2 x i64> [[TMP1]] to i128
-; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i128 [[TMP2]], 0
-; CHECK-NEXT: br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
-; CHECK: 3:
-; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR3]]
-; CHECK-NEXT: unreachable
-; CHECK: 4:
+; CHECK-NEXT: [[TMP2:%.*]] = call i64 @llvm.vector.reduce.or.v2i64(<2 x i64> [[TMP1]])
; CHECK-NEXT: [[MINNM:%.*]] = call double @llvm.aarch64.neon.fminnmv.f64.v2f64(<2 x double> [[IN]])
-; CHECK-NEXT: store i64 0, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT: store i64 [[TMP2]], ptr @__msan_retval_tls, align 8
; CHECK-NEXT: ret double [[MINNM]]
;
%minnm = call double @llvm.aarch64.neon.fminnmv.f64.v2f64(<2 x double> %in)
@@ -209,15 +149,9 @@ define float @test_fmaxnmv_v2f32(<2 x float> %in) #0 {
; CHECK-SAME: <2 x float> [[IN:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[TMP1:%.*]] = load <2 x i32>, ptr @__msan_param_tls, align 8
; CHECK-NEXT: call void @llvm.donothing()
-; CHECK-NEXT: [[TMP2:%.*]] = bitcast <2 x i32> [[TMP1]] to i64
-; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i64 [[TMP2]], 0
-; CHECK-NEXT: br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
-; CHECK: 3:
-; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR3]]
-; CHECK-NEXT: unreachable
-; CHECK: 4:
+; CHECK-NEXT: [[TMP2:%.*]] = call i32 @llvm.vector.reduce.or.v2i32(<2 x i32> [[TMP1]])
; CHECK-NEXT: [[MAXNM:%.*]] = call float @llvm.aarch64.neon.fmaxnmv.f32.v2f32(<2 x float> [[IN]])
-; CHECK-NEXT: store i32 0, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT: store i32 [[TMP2]], ptr @__msan_retval_tls, align 8
; CHECK-NEXT: ret float [[MAXNM]]
;
%maxnm = call float @llvm.aarch64.neon.fmaxnmv.f32.v2f32(<2 x float> %in)
@@ -229,15 +163,9 @@ define float @test_fmaxnmv_v4f32(<4 x float> %in) #0 {
; CHECK-SAME: <4 x float> [[IN:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8
; CHECK-NEXT: call void @llvm.donothing()
-; CHECK-NEXT: [[TMP2:%.*]] = bitcast <4 x i32> [[TMP1]] to i128
-; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i128 [[TMP2]], 0
-; CHECK-NEXT: br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
-; CHECK: 3:
-; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR3]]
-; CHECK-NEXT: unreachable
-; CHECK: 4:
+; CHECK-NEXT: [[TMP2:%.*]] = call i32 @llvm.vector.reduce.or.v4i32(<4 x i32> [[TMP1]])
; CHECK-NEXT: [[MAXNM:%.*]] = call float @llvm.aarch64.neon.fmaxnmv.f32.v4f32(<4 x float> [[IN]])
-; CHECK-NEXT: store i32 0, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT: store i32 [[TMP2]], ptr @__msan_retval_tls, align 8
; CHECK-NEXT: ret float [[MAXNM]]
;
%maxnm = call float @llvm.aarch64.neon.fmaxnmv.f32.v4f32(<4 x float> %in)
@@ -249,15 +177,9 @@ define double @test_fmaxnmv_v2f64(<2 x double> %in) #0 {
; CHECK-SAME: <2 x double> [[IN:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[TMP1:%.*]] = load <2 x i64>, ptr @__msan_param_tls, align 8
; CHECK-NEXT: call void @llvm.donothing()
-; CHECK-NEXT: [[TMP2:%.*]] = bitcast <2 x i64> [[TMP1]] to i128
-; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i128 [[TMP2]], 0
-; CHECK-NEXT: br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
-; CHECK: 3:
-; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR3]]
-; CHECK-NEXT: unreachable
-; CHECK: 4:
+; CHECK-NEXT: [[TMP2:%.*]] = call i64 @llvm.vector.reduce.or.v2i64(<2 x i64> [[TMP1]])
; CHECK-NEXT: [[MAXNM:%.*]] = call double @llvm.aarch64.neon.fmaxnmv.f64.v2f64(<2 x double> [[IN]])
-; CHECK-NEXT: store i64 0, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT: store i64 [[TMP2]], ptr @__msan_retval_tls, align 8
; CHECK-NEXT: ret double [[MAXNM]]
;
%maxnm = call double @llvm.aarch64.neon.fmaxnmv.f64.v2f64(<2 x double> %in)
@@ -269,6 +191,3 @@ declare float @llvm.aarch64.neon.fmaxnmv.f32.v4f32(<4 x float>)
declare double @llvm.aarch64.neon.fmaxnmv.f64.v2f64(<2 x double>)
attributes #0 = { sanitize_memory }
-;.
-; CHECK: [[PROF1]] = !{!"branch_weights", i32 1, i32 1048575}
-;.
``````````
</details>
https://github.com/llvm/llvm-project/pull/125778
More information about the llvm-commits
mailing list