Skip to content
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.

Sign up for GitHub

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

Jump to bottom

[msan] Use horizontal add to compute shadow for horizontal sub #124835

Merged
merged 2 commits into from
Jan 28, 2025
Merged

[msan] Use horizontal add to compute shadow for horizontal sub #124835

merged 2 commits into from
Jan 28, 2025

Conversation

thurstond
Copy link
Contributor

@thurstond thurstond commented Jan 28, 2025 •
edited
Loading

This improves the horizontal sub handling (from #124159), by always using horizontal add for the shadow, as recommended by Vitaly.

Fixes #124662

@thurstond
[msan] Use horizontal sub to compute shadow for horizontal add
5dcb768 
This improves the horizontal sub handling (from llvm#124159) by using horizontal add for the shadow, as recommended by Vitaly.

Fixes llvm#124662
@llvmbot
Copy link
Member

llvmbot commented Jan 28, 2025 •
edited
Loading

@llvm/pr-subscribers-llvm-transforms

@llvm/pr-subscribers-compiler-rt-sanitizer

Author: Thurston Dang (thurstond)

Changes

This improves the horizontal sub handling (from #124159) by using horizontal add for the shadow, as recommended by Vitaly.

Fixes #124662

Full diff: https://github.com/llvm/llvm-project/pull/124835.diff

7 Files Affected:

  • (modified) llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp (+72-9)
  • (modified) llvm/test/Instrumentation/MemorySanitizer/X86/avx-intrinsics-x86.ll (+2-2)
  • (modified) llvm/test/Instrumentation/MemorySanitizer/X86/avx2-intrinsics-x86.ll (+3-3)
  • (modified) llvm/test/Instrumentation/MemorySanitizer/X86/mmx-intrinsics.ll (+3-3)
  • (modified) llvm/test/Instrumentation/MemorySanitizer/i386/avx-intrinsics-i386.ll (+2-2)
  • (modified) llvm/test/Instrumentation/MemorySanitizer/i386/avx2-intrinsics-i386.ll (+3-3)
  • (modified) llvm/test/Instrumentation/MemorySanitizer/i386/mmx-intrinsics.ll (+3-3)
diff --git a/llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp b/llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp
index fadebc3ae42660..8f014ce195adbd 100644
--- a/llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp
+++ b/llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp
@@ -4037,20 +4037,83 @@ struct MemorySanitizerVisitor : public InstVisitor<MemorySanitizerVisitor> {
 
   void handleAVXHorizontalAddSubIntrinsic(IntrinsicInst &I) {
     // Approximation only:
-    //    output         = horizontal_add(A, B)
+    //    output         = horizontal_add/sub(A, B)
     // => shadow[output] = horizontal_add(shadow[A], shadow[B])
     //
-    // - If we add/subtract two adjacent zero (initialized) shadow values, the
+    // We always use horizontal add instead of subtract, because subtracting
+    // a fully uninitialized shadow would result in a fully initialized shadow.
+    //
+    // - If we add two adjacent zero (initialized) shadow values, the
     //   result always be zero i.e., no false positives.
-    // - If we add/subtract two shadows, one of which is uninitialized, the
-    //   result will always be non-zero i.e., no false negative.
-    // - However, we can have false negatives if we subtract two non-zero
-    //   shadows of the same value (or do an addition that wraps to zero); we
-    //   consider this an acceptable tradeoff for performance.
+    // - If we add two shadows, one of which is uninitialized, the
+    //   result will always be non-zero i.e., no false negatives.
+    // - However, we can have false negatives if we do an addition that wraps
+    //   to zero; we consider this an acceptable tradeoff for performance.
+    //
     // To make shadow propagation precise, we want the equivalent of
-    // "horizontal OR", but this is not available.
+    // "horizontal OR", but this is not available for SSE3/SSSE3/AVX/AVX2.
+
+    Intrinsic::ID shadowIntrinsicID = I.getIntrinsicID();
+
+    switch(I.getIntrinsicID()) {
+    case Intrinsic::x86_sse3_hsub_ps:
+      shadowIntrinsicID = Intrinsic::x86_sse3_hadd_ps;
+      break;
+
+    case Intrinsic::x86_sse3_hsub_pd:
+      shadowIntrinsicID = Intrinsic::x86_sse3_hadd_pd;
+      break;
+
+    case Intrinsic::x86_ssse3_phsub_d:
+      shadowIntrinsicID = Intrinsic::x86_ssse3_phadd_d;
+      break;
+
+    case Intrinsic::x86_ssse3_phsub_d_128:
+      shadowIntrinsicID = Intrinsic::x86_ssse3_phadd_d_128;
+      break;
+
+    case Intrinsic::x86_ssse3_phsub_w:
+      shadowIntrinsicID = Intrinsic::x86_ssse3_phadd_w;
+      break;
+
+    case Intrinsic::x86_ssse3_phsub_w_128:
+      shadowIntrinsicID = Intrinsic::x86_ssse3_phadd_w_128;
+      break;
+
+    case Intrinsic::x86_ssse3_phsub_sw:
+      shadowIntrinsicID = Intrinsic::x86_ssse3_phadd_sw;
+      break;
+
+    case Intrinsic::x86_ssse3_phsub_sw_128:
+      shadowIntrinsicID = Intrinsic::x86_ssse3_phadd_sw_128;
+      break;
+
+    case Intrinsic::x86_avx_hsub_pd_256:
+      shadowIntrinsicID = Intrinsic::x86_avx_hadd_pd_256;
+      break;
+
+    case Intrinsic::x86_avx_hsub_ps_256:
+      shadowIntrinsicID = Intrinsic::x86_avx_hadd_ps_256;
+      break;
+
+    case Intrinsic::x86_avx2_phsub_d:
+      shadowIntrinsicID = Intrinsic::x86_avx2_phadd_d;
+      break;
+
+    case Intrinsic::x86_avx2_phsub_w:
+      shadowIntrinsicID = Intrinsic::x86_avx2_phadd_w;
+      break;
+
+    case Intrinsic::x86_avx2_phsub_sw:
+      shadowIntrinsicID = Intrinsic::x86_avx2_phadd_sw;
+      break;
+
+    default:
+      break;
+   }
+
     return handleIntrinsicByApplyingToShadow(
-        I, /*shadowIntrinsicID=*/I.getIntrinsicID(),
+        I, shadowIntrinsicID,
         /*trailingVerbatimArgs*/ 0);
   }
 
diff --git a/llvm/test/Instrumentation/MemorySanitizer/X86/avx-intrinsics-x86.ll b/llvm/test/Instrumentation/MemorySanitizer/X86/avx-intrinsics-x86.ll
index 43f51a810d0d2b..26e9c39696f707 100644
--- a/llvm/test/Instrumentation/MemorySanitizer/X86/avx-intrinsics-x86.ll
+++ b/llvm/test/Instrumentation/MemorySanitizer/X86/avx-intrinsics-x86.ll
@@ -475,7 +475,7 @@ define <4 x double> @test_x86_avx_hsub_pd_256(<4 x double> %a0, <4 x double> %a1
 ; CHECK-NEXT:    call void @llvm.donothing()
 ; CHECK-NEXT:    [[A0:%.*]] = bitcast <4 x i64> [[TMP1]] to <4 x double>
 ; CHECK-NEXT:    [[A1:%.*]] = bitcast <4 x i64> [[TMP2]] to <4 x double>
-; CHECK-NEXT:    [[RES:%.*]] = call <4 x double> @llvm.x86.avx.hsub.pd.256(<4 x double> [[A0]], <4 x double> [[A1]])
+; CHECK-NEXT:    [[RES:%.*]] = call <4 x double> @llvm.x86.avx.hadd.pd.256(<4 x double> [[A0]], <4 x double> [[A1]])
 ; CHECK-NEXT:    [[_MSPROP:%.*]] = bitcast <4 x double> [[RES]] to <4 x i64>
 ; CHECK-NEXT:    [[RES1:%.*]] = call <4 x double> @llvm.x86.avx.hsub.pd.256(<4 x double> [[A2:%.*]], <4 x double> [[A3:%.*]])
 ; CHECK-NEXT:    store <4 x i64> [[_MSPROP]], ptr @__msan_retval_tls, align 8
@@ -494,7 +494,7 @@ define <8 x float> @test_x86_avx_hsub_ps_256(<8 x float> %a0, <8 x float> %a1) #
 ; CHECK-NEXT:    call void @llvm.donothing()
 ; CHECK-NEXT:    [[A0:%.*]] = bitcast <8 x i32> [[TMP1]] to <8 x float>
 ; CHECK-NEXT:    [[A1:%.*]] = bitcast <8 x i32> [[TMP2]] to <8 x float>
-; CHECK-NEXT:    [[RES:%.*]] = call <8 x float> @llvm.x86.avx.hsub.ps.256(<8 x float> [[A0]], <8 x float> [[A1]])
+; CHECK-NEXT:    [[RES:%.*]] = call <8 x float> @llvm.x86.avx.hadd.ps.256(<8 x float> [[A0]], <8 x float> [[A1]])
 ; CHECK-NEXT:    [[_MSPROP:%.*]] = bitcast <8 x float> [[RES]] to <8 x i32>
 ; CHECK-NEXT:    [[RES1:%.*]] = call <8 x float> @llvm.x86.avx.hsub.ps.256(<8 x float> [[A2:%.*]], <8 x float> [[A3:%.*]])
 ; CHECK-NEXT:    store <8 x i32> [[_MSPROP]], ptr @__msan_retval_tls, align 8
diff --git a/llvm/test/Instrumentation/MemorySanitizer/X86/avx2-intrinsics-x86.ll b/llvm/test/Instrumentation/MemorySanitizer/X86/avx2-intrinsics-x86.ll
index c68461dd367ee9..5597a9c96611f0 100644
--- a/llvm/test/Instrumentation/MemorySanitizer/X86/avx2-intrinsics-x86.ll
+++ b/llvm/test/Instrumentation/MemorySanitizer/X86/avx2-intrinsics-x86.ll
@@ -617,7 +617,7 @@ define <8 x i32> @test_x86_avx2_phsub_d(<8 x i32> %a0, <8 x i32> %a1) #0 {
 ; CHECK-NEXT:    [[TMP1:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8
 ; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 32) to ptr), align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = call <8 x i32> @llvm.x86.avx2.phsub.d(<8 x i32> [[TMP1]], <8 x i32> [[TMP2]])
+; CHECK-NEXT:    [[_MSPROP:%.*]] = call <8 x i32> @llvm.x86.avx2.phadd.d(<8 x i32> [[TMP1]], <8 x i32> [[TMP2]])
 ; CHECK-NEXT:    [[RES:%.*]] = call <8 x i32> @llvm.x86.avx2.phsub.d(<8 x i32> [[A0:%.*]], <8 x i32> [[A1:%.*]])
 ; CHECK-NEXT:    store <8 x i32> [[_MSPROP]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <8 x i32> [[RES]]
@@ -633,7 +633,7 @@ define <16 x i16> @test_x86_avx2_phsub_sw(<16 x i16> %a0, <16 x i16> %a1) #0 {
 ; CHECK-NEXT:    [[TMP1:%.*]] = load <16 x i16>, ptr @__msan_param_tls, align 8
 ; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 32) to ptr), align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = call <16 x i16> @llvm.x86.avx2.phsub.sw(<16 x i16> [[TMP1]], <16 x i16> [[TMP2]])
+; CHECK-NEXT:    [[_MSPROP:%.*]] = call <16 x i16> @llvm.x86.avx2.phadd.sw(<16 x i16> [[TMP1]], <16 x i16> [[TMP2]])
 ; CHECK-NEXT:    [[RES:%.*]] = call <16 x i16> @llvm.x86.avx2.phsub.sw(<16 x i16> [[A0:%.*]], <16 x i16> [[A1:%.*]])
 ; CHECK-NEXT:    store <16 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <16 x i16> [[RES]]
@@ -649,7 +649,7 @@ define <16 x i16> @test_x86_avx2_phsub_w(<16 x i16> %a0, <16 x i16> %a1) #0 {
 ; CHECK-NEXT:    [[TMP1:%.*]] = load <16 x i16>, ptr @__msan_param_tls, align 8
 ; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 32) to ptr), align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = call <16 x i16> @llvm.x86.avx2.phsub.w(<16 x i16> [[TMP1]], <16 x i16> [[TMP2]])
+; CHECK-NEXT:    [[_MSPROP:%.*]] = call <16 x i16> @llvm.x86.avx2.phadd.w(<16 x i16> [[TMP1]], <16 x i16> [[TMP2]])
 ; CHECK-NEXT:    [[RES:%.*]] = call <16 x i16> @llvm.x86.avx2.phsub.w(<16 x i16> [[A0:%.*]], <16 x i16> [[A1:%.*]])
 ; CHECK-NEXT:    store <16 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <16 x i16> [[RES]]
diff --git a/llvm/test/Instrumentation/MemorySanitizer/X86/mmx-intrinsics.ll b/llvm/test/Instrumentation/MemorySanitizer/X86/mmx-intrinsics.ll
index 61c90d0fb80d43..2b6c6ff2e2b929 100644
--- a/llvm/test/Instrumentation/MemorySanitizer/X86/mmx-intrinsics.ll
+++ b/llvm/test/Instrumentation/MemorySanitizer/X86/mmx-intrinsics.ll
@@ -3339,7 +3339,7 @@ define i64 @test6(<1 x i64> %a, <1 x i64> %b) #0 {
 ; CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[TMP1]] to <1 x i64>
 ; CHECK-NEXT:    [[TMP8:%.*]] = bitcast <4 x i16> [[TMP12]] to <1 x i64>
 ; CHECK-NEXT:    [[TMP17:%.*]] = bitcast <4 x i16> [[TMP0]] to <1 x i64>
-; CHECK-NEXT:    [[_MSPROP:%.*]] = call <1 x i64> @llvm.x86.ssse3.phsub.sw(<1 x i64> [[TMP16]], <1 x i64> [[TMP8]])
+; CHECK-NEXT:    [[_MSPROP:%.*]] = call <1 x i64> @llvm.x86.ssse3.phadd.sw(<1 x i64> [[TMP16]], <1 x i64> [[TMP8]])
 ; CHECK-NEXT:    [[TMP18:%.*]] = tail call <1 x i64> @llvm.x86.ssse3.phsub.sw(<1 x i64> [[TMP2]], <1 x i64> [[TMP17]]) #[[ATTR5]]
 ; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <1 x i64> [[_MSPROP]] to <4 x i16>
 ; CHECK-NEXT:    [[TMP19:%.*]] = bitcast <1 x i64> [[TMP18]] to <4 x i16>
@@ -3379,7 +3379,7 @@ define i64 @test5(<1 x i64> %a, <1 x i64> %b) #0 {
 ; CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[TMP1]] to <1 x i64>
 ; CHECK-NEXT:    [[TMP8:%.*]] = bitcast <2 x i32> [[TMP12]] to <1 x i64>
 ; CHECK-NEXT:    [[TMP17:%.*]] = bitcast <2 x i32> [[TMP0]] to <1 x i64>
-; CHECK-NEXT:    [[_MSPROP:%.*]] = call <1 x i64> @llvm.x86.ssse3.phsub.d(<1 x i64> [[TMP16]], <1 x i64> [[TMP8]])
+; CHECK-NEXT:    [[_MSPROP:%.*]] = call <1 x i64> @llvm.x86.ssse3.phadd.d(<1 x i64> [[TMP16]], <1 x i64> [[TMP8]])
 ; CHECK-NEXT:    [[TMP18:%.*]] = tail call <1 x i64> @llvm.x86.ssse3.phsub.d(<1 x i64> [[TMP2]], <1 x i64> [[TMP17]]) #[[ATTR5]]
 ; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <1 x i64> [[_MSPROP]] to <2 x i32>
 ; CHECK-NEXT:    [[TMP19:%.*]] = bitcast <1 x i64> [[TMP18]] to <2 x i32>
@@ -3419,7 +3419,7 @@ define i64 @test4(<1 x i64> %a, <1 x i64> %b) #0 {
 ; CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[TMP1]] to <1 x i64>
 ; CHECK-NEXT:    [[TMP8:%.*]] = bitcast <4 x i16> [[TMP12]] to <1 x i64>
 ; CHECK-NEXT:    [[TMP17:%.*]] = bitcast <4 x i16> [[TMP0]] to <1 x i64>
-; CHECK-NEXT:    [[_MSPROP:%.*]] = call <1 x i64> @llvm.x86.ssse3.phsub.w(<1 x i64> [[TMP16]], <1 x i64> [[TMP8]])
+; CHECK-NEXT:    [[_MSPROP:%.*]] = call <1 x i64> @llvm.x86.ssse3.phadd.w(<1 x i64> [[TMP16]], <1 x i64> [[TMP8]])
 ; CHECK-NEXT:    [[TMP18:%.*]] = tail call <1 x i64> @llvm.x86.ssse3.phsub.w(<1 x i64> [[TMP2]], <1 x i64> [[TMP17]]) #[[ATTR5]]
 ; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <1 x i64> [[_MSPROP]] to <4 x i16>
 ; CHECK-NEXT:    [[TMP19:%.*]] = bitcast <1 x i64> [[TMP18]] to <4 x i16>
diff --git a/llvm/test/Instrumentation/MemorySanitizer/i386/avx-intrinsics-i386.ll b/llvm/test/Instrumentation/MemorySanitizer/i386/avx-intrinsics-i386.ll
index a22ca6dd15da4d..08dd27ffcaaf12 100644
--- a/llvm/test/Instrumentation/MemorySanitizer/i386/avx-intrinsics-i386.ll
+++ b/llvm/test/Instrumentation/MemorySanitizer/i386/avx-intrinsics-i386.ll
@@ -491,7 +491,7 @@ define <4 x double> @test_x86_avx_hsub_pd_256(<4 x double> %a0, <4 x double> %a1
 ; CHECK-NEXT:    call void @llvm.donothing()
 ; CHECK-NEXT:    [[A0:%.*]] = bitcast <4 x i64> [[TMP1]] to <4 x double>
 ; CHECK-NEXT:    [[A1:%.*]] = bitcast <4 x i64> [[TMP2]] to <4 x double>
-; CHECK-NEXT:    [[RES:%.*]] = call <4 x double> @llvm.x86.avx.hsub.pd.256(<4 x double> [[A0]], <4 x double> [[A1]])
+; CHECK-NEXT:    [[RES:%.*]] = call <4 x double> @llvm.x86.avx.hadd.pd.256(<4 x double> [[A0]], <4 x double> [[A1]])
 ; CHECK-NEXT:    [[_MSPROP:%.*]] = bitcast <4 x double> [[RES]] to <4 x i64>
 ; CHECK-NEXT:    [[RES1:%.*]] = call <4 x double> @llvm.x86.avx.hsub.pd.256(<4 x double> [[A2:%.*]], <4 x double> [[A3:%.*]])
 ; CHECK-NEXT:    store <4 x i64> [[_MSPROP]], ptr @__msan_retval_tls, align 8
@@ -511,7 +511,7 @@ define <8 x float> @test_x86_avx_hsub_ps_256(<8 x float> %a0, <8 x float> %a1) #
 ; CHECK-NEXT:    call void @llvm.donothing()
 ; CHECK-NEXT:    [[A0:%.*]] = bitcast <8 x i32> [[TMP1]] to <8 x float>
 ; CHECK-NEXT:    [[A1:%.*]] = bitcast <8 x i32> [[TMP2]] to <8 x float>
-; CHECK-NEXT:    [[RES:%.*]] = call <8 x float> @llvm.x86.avx.hsub.ps.256(<8 x float> [[A0]], <8 x float> [[A1]])
+; CHECK-NEXT:    [[RES:%.*]] = call <8 x float> @llvm.x86.avx.hadd.ps.256(<8 x float> [[A0]], <8 x float> [[A1]])
 ; CHECK-NEXT:    [[_MSPROP:%.*]] = bitcast <8 x float> [[RES]] to <8 x i32>
 ; CHECK-NEXT:    [[RES1:%.*]] = call <8 x float> @llvm.x86.avx.hsub.ps.256(<8 x float> [[A2:%.*]], <8 x float> [[A3:%.*]])
 ; CHECK-NEXT:    store <8 x i32> [[_MSPROP]], ptr @__msan_retval_tls, align 8
diff --git a/llvm/test/Instrumentation/MemorySanitizer/i386/avx2-intrinsics-i386.ll b/llvm/test/Instrumentation/MemorySanitizer/i386/avx2-intrinsics-i386.ll
index 442f0c422645af..109166975eee13 100644
--- a/llvm/test/Instrumentation/MemorySanitizer/i386/avx2-intrinsics-i386.ll
+++ b/llvm/test/Instrumentation/MemorySanitizer/i386/avx2-intrinsics-i386.ll
@@ -651,7 +651,7 @@ define <8 x i32> @test_x86_avx2_phsub_d(<8 x i32> %a0, <8 x i32> %a1) #0 {
 ; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 32) to ptr), align 8
 ; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr @__msan_va_arg_overflow_size_tls, align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = call <8 x i32> @llvm.x86.avx2.phsub.d(<8 x i32> [[TMP1]], <8 x i32> [[TMP2]])
+; CHECK-NEXT:    [[_MSPROP:%.*]] = call <8 x i32> @llvm.x86.avx2.phadd.d(<8 x i32> [[TMP1]], <8 x i32> [[TMP2]])
 ; CHECK-NEXT:    [[RES:%.*]] = call <8 x i32> @llvm.x86.avx2.phsub.d(<8 x i32> [[A0:%.*]], <8 x i32> [[A1:%.*]])
 ; CHECK-NEXT:    store <8 x i32> [[_MSPROP]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <8 x i32> [[RES]]
@@ -668,7 +668,7 @@ define <16 x i16> @test_x86_avx2_phsub_sw(<16 x i16> %a0, <16 x i16> %a1) #0 {
 ; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 32) to ptr), align 8
 ; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr @__msan_va_arg_overflow_size_tls, align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = call <16 x i16> @llvm.x86.avx2.phsub.sw(<16 x i16> [[TMP1]], <16 x i16> [[TMP2]])
+; CHECK-NEXT:    [[_MSPROP:%.*]] = call <16 x i16> @llvm.x86.avx2.phadd.sw(<16 x i16> [[TMP1]], <16 x i16> [[TMP2]])
 ; CHECK-NEXT:    [[RES:%.*]] = call <16 x i16> @llvm.x86.avx2.phsub.sw(<16 x i16> [[A0:%.*]], <16 x i16> [[A1:%.*]])
 ; CHECK-NEXT:    store <16 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <16 x i16> [[RES]]
@@ -685,7 +685,7 @@ define <16 x i16> @test_x86_avx2_phsub_w(<16 x i16> %a0, <16 x i16> %a1) #0 {
 ; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 32) to ptr), align 8
 ; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr @__msan_va_arg_overflow_size_tls, align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = call <16 x i16> @llvm.x86.avx2.phsub.w(<16 x i16> [[TMP1]], <16 x i16> [[TMP2]])
+; CHECK-NEXT:    [[_MSPROP:%.*]] = call <16 x i16> @llvm.x86.avx2.phadd.w(<16 x i16> [[TMP1]], <16 x i16> [[TMP2]])
 ; CHECK-NEXT:    [[RES:%.*]] = call <16 x i16> @llvm.x86.avx2.phsub.w(<16 x i16> [[A0:%.*]], <16 x i16> [[A1:%.*]])
 ; CHECK-NEXT:    store <16 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <16 x i16> [[RES]]
diff --git a/llvm/test/Instrumentation/MemorySanitizer/i386/mmx-intrinsics.ll b/llvm/test/Instrumentation/MemorySanitizer/i386/mmx-intrinsics.ll
index f5ef8a9837b70b..15bd1755d479bf 100644
--- a/llvm/test/Instrumentation/MemorySanitizer/i386/mmx-intrinsics.ll
+++ b/llvm/test/Instrumentation/MemorySanitizer/i386/mmx-intrinsics.ll
@@ -3426,7 +3426,7 @@ define i64 @test6(<1 x i64> %a, <1 x i64> %b) #0 {
 ; CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[TMP1]] to <1 x i64>
 ; CHECK-NEXT:    [[TMP8:%.*]] = bitcast <4 x i16> [[TMP12]] to <1 x i64>
 ; CHECK-NEXT:    [[TMP17:%.*]] = bitcast <4 x i16> [[TMP0]] to <1 x i64>
-; CHECK-NEXT:    [[_MSPROP:%.*]] = call <1 x i64> @llvm.x86.ssse3.phsub.sw(<1 x i64> [[TMP16]], <1 x i64> [[TMP8]])
+; CHECK-NEXT:    [[_MSPROP:%.*]] = call <1 x i64> @llvm.x86.ssse3.phadd.sw(<1 x i64> [[TMP16]], <1 x i64> [[TMP8]])
 ; CHECK-NEXT:    [[TMP18:%.*]] = tail call <1 x i64> @llvm.x86.ssse3.phsub.sw(<1 x i64> [[TMP2]], <1 x i64> [[TMP17]]) #[[ATTR5]]
 ; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <1 x i64> [[_MSPROP]] to <4 x i16>
 ; CHECK-NEXT:    [[TMP19:%.*]] = bitcast <1 x i64> [[TMP18]] to <4 x i16>
@@ -3467,7 +3467,7 @@ define i64 @test5(<1 x i64> %a, <1 x i64> %b) #0 {
 ; CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[TMP1]] to <1 x i64>
 ; CHECK-NEXT:    [[TMP8:%.*]] = bitcast <2 x i32> [[TMP12]] to <1 x i64>
 ; CHECK-NEXT:    [[TMP17:%.*]] = bitcast <2 x i32> [[TMP0]] to <1 x i64>
-; CHECK-NEXT:    [[_MSPROP:%.*]] = call <1 x i64> @llvm.x86.ssse3.phsub.d(<1 x i64> [[TMP16]], <1 x i64> [[TMP8]])
+; CHECK-NEXT:    [[_MSPROP:%.*]] = call <1 x i64> @llvm.x86.ssse3.phadd.d(<1 x i64> [[TMP16]], <1 x i64> [[TMP8]])
 ; CHECK-NEXT:    [[TMP18:%.*]] = tail call <1 x i64> @llvm.x86.ssse3.phsub.d(<1 x i64> [[TMP2]], <1 x i64> [[TMP17]]) #[[ATTR5]]
 ; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <1 x i64> [[_MSPROP]] to <2 x i32>
 ; CHECK-NEXT:    [[TMP19:%.*]] = bitcast <1 x i64> [[TMP18]] to <2 x i32>
@@ -3508,7 +3508,7 @@ define i64 @test4(<1 x i64> %a, <1 x i64> %b) #0 {
 ; CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[TMP1]] to <1 x i64>
 ; CHECK-NEXT:    [[TMP8:%.*]] = bitcast <4 x i16> [[TMP12]] to <1 x i64>
 ; CHECK-NEXT:    [[TMP17:%.*]] = bitcast <4 x i16> [[TMP0]] to <1 x i64>
-; CHECK-NEXT:    [[_MSPROP:%.*]] = call <1 x i64> @llvm.x86.ssse3.phsub.w(<1 x i64> [[TMP16]], <1 x i64> [[TMP8]])
+; CHECK-NEXT:    [[_MSPROP:%.*]] = call <1 x i64> @llvm.x86.ssse3.phadd.w(<1 x i64> [[TMP16]], <1 x i64> [[TMP8]])
 ; CHECK-NEXT:    [[TMP18:%.*]] = tail call <1 x i64> @llvm.x86.ssse3.phsub.w(<1 x i64> [[TMP2]], <1 x i64> [[TMP17]]) #[[ATTR5]]
 ; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <1 x i64> [[_MSPROP]] to <4 x i16>
 ; CHECK-NEXT:    [[TMP19:%.*]] = bitcast <1 x i64> [[TMP18]] to <4 x i16>

@github-actions GitHub Actions
Copy link

github-actions bot commented Jan 28, 2025 •
edited
Loading

✅ With the latest revision this PR passed the C/C++ code formatter.

@fmayer
Copy link
Contributor

fmayer commented Jan 28, 2025

Isn't the commit message the wrong way round?

@thurstond
Copy link
Contributor Author

Isn't the commit message the wrong way round?

I added an (Oxford?) comma to clarify

@fmayer
Copy link
Contributor

fmayer commented Jan 28, 2025

I added an (Oxford?) comma to clarify

I, meant, the, first, line

@thurstond thurstond changed the title [msan] Use horizontal sub to compute shadow for horizontal add [msan] Use horizontal add to compute shadow for horizontal sub Jan 28, 2025
@thurstond
Copy link
Contributor Author

I added an (Oxford?) comma to clarify

I, meant, the, first, line

Thanks, fixed,

@thurstond thurstond mentioned this pull request Jan 28, 2025
Merged
@thurstond
Copy link
Contributor Author

The CI failure is unrelated: Clang.CodeGen/extend-variable-liveness-except.cpp

@thurstond thurstond merged commit 4a42607 into llvm:main Jan 28, 2025
5 of 7 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@fmayer fmayer fmayer approved these changes

@vitalybuka vitalybuka Awaiting requested review from vitalybuka

@kstoimenov kstoimenov Awaiting requested review from kstoimenov

Assignees
No one assigned
Labels
compiler-rt:sanitizer llvm:transforms
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.

msan: Consider handling add/sub as add
3 participants
@thurstond @llvmbot @fmayer