Overflow and saturating intrinsics

[eugenis]

No description provided.

[llvmbot]

@llvm/pr-subscribers-llvm-transforms

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

Author: Evgenii Stepanov (eugenis)

Changes

---

Full diff: https://github.com/llvm/llvm-project/pull/88068.diff

3 Files Affected:

• (modified) llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp (+29)
• (added) llvm/test/Instrumentation/MemorySanitizer/overflow.ll (+130)
• (added) llvm/test/Instrumentation/MemorySanitizer/saturating.ll (+120)

diff --git a/llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp b/llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp
index 46b9181c8922e0..b92d358b4aa16f 100644
--- a/llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp
+++ b/llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp
@@ -3715,8 +3715,37 @@ struct MemorySanitizerVisitor : public InstVisitor<MemorySanitizerVisitor> {
     setOrigin(&I, getOrigin(&I, 0));
   }
 
+  void handleArithmeticWithOverflow(IntrinsicInst &I) {
+    IRBuilder<> IRB(&I);
+    Value *Shadow0 = IRB.CreateOr(getShadow(&I, 0), getShadow(&I, 1));
+    Value *Shadow1 = IRB.CreateICmpNE(Shadow0, getCleanShadow(Shadow0));
+
+    Value *Shadow = PoisonValue::get(getShadowTy(&I));
+    Shadow = IRB.CreateInsertValue(Shadow, Shadow0, 0);
+    Shadow = IRB.CreateInsertValue(Shadow, Shadow1, 1);
+
+    setShadow(&I, Shadow);
+    setOriginForNaryOp(I);
+  }
+
   void visitIntrinsicInst(IntrinsicInst &I) {
     switch (I.getIntrinsicID()) {
+    case Intrinsic::uadd_with_overflow:
+    case Intrinsic::sadd_with_overflow:
+    case Intrinsic::usub_with_overflow:
+    case Intrinsic::ssub_with_overflow:
+    case Intrinsic::umul_with_overflow:
+    case Intrinsic::smul_with_overflow:
+      handleArithmeticWithOverflow(I);
+      break;
+    case Intrinsic::sadd_sat:
+    case Intrinsic::uadd_sat:
+    case Intrinsic::ssub_sat:
+    case Intrinsic::usub_sat:
+    case Intrinsic::sshl_sat:
+    case Intrinsic::ushl_sat:
+      handleShadowOr(I);
+      break;
     case Intrinsic::abs:
       handleAbsIntrinsic(I);
       break;
diff --git a/llvm/test/Instrumentation/MemorySanitizer/overflow.ll b/llvm/test/Instrumentation/MemorySanitizer/overflow.ll
new file mode 100644
index 00000000000000..0cfae0008263f7
--- /dev/null
+++ b/llvm/test/Instrumentation/MemorySanitizer/overflow.ll
@@ -0,0 +1,130 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4
+; RUN: opt %s -S -passes=msan 2>&1 | FileCheck %s
+
+target datalayout = "e-m:o-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+define {i64, i1} @test_sadd_with_overflow(i64 %a, i64 %b) #0 {
+; CHECK-LABEL: define { i64, i1 } @test_sadd_with_overflow(
+; CHECK-SAME: i64 [[A:%.*]], i64 [[B:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = or i64 [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ne i64 [[TMP3]], 0
+; CHECK-NEXT:    [[TMP5:%.*]] = insertvalue { i64, i1 } poison, i64 [[TMP3]], 0
+; CHECK-NEXT:    [[TMP6:%.*]] = insertvalue { i64, i1 } [[TMP5]], i1 [[TMP4]], 1
+; CHECK-NEXT:    [[RES:%.*]] = call { i64, i1 } @llvm.sadd.with.overflow.i64(i64 [[A]], i64 [[B]])
+; CHECK-NEXT:    store { i64, i1 } [[TMP6]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret { i64, i1 } [[RES]]
+;
+  %res = call { i64, i1 } @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
+  ret { i64, i1 } %res
+}
+
+define {i64, i1} @test_uadd_with_overflow(i64 %a, i64 %b) #0 {
+; CHECK-LABEL: define { i64, i1 } @test_uadd_with_overflow(
+; CHECK-SAME: i64 [[A:%.*]], i64 [[B:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = or i64 [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ne i64 [[TMP3]], 0
+; CHECK-NEXT:    [[TMP5:%.*]] = insertvalue { i64, i1 } poison, i64 [[TMP3]], 0
+; CHECK-NEXT:    [[TMP6:%.*]] = insertvalue { i64, i1 } [[TMP5]], i1 [[TMP4]], 1
+; CHECK-NEXT:    [[RES:%.*]] = call { i64, i1 } @llvm.uadd.with.overflow.i64(i64 [[A]], i64 [[B]])
+; CHECK-NEXT:    store { i64, i1 } [[TMP6]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret { i64, i1 } [[RES]]
+;
+  %res = call { i64, i1 } @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
+  ret { i64, i1 } %res
+}
+
+define {i64, i1} @test_smul_with_overflow(i64 %a, i64 %b) #0 {
+; CHECK-LABEL: define { i64, i1 } @test_smul_with_overflow(
+; CHECK-SAME: i64 [[A:%.*]], i64 [[B:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = or i64 [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ne i64 [[TMP3]], 0
+; CHECK-NEXT:    [[TMP5:%.*]] = insertvalue { i64, i1 } poison, i64 [[TMP3]], 0
+; CHECK-NEXT:    [[TMP6:%.*]] = insertvalue { i64, i1 } [[TMP5]], i1 [[TMP4]], 1
+; CHECK-NEXT:    [[RES:%.*]] = call { i64, i1 } @llvm.smul.with.overflow.i64(i64 [[A]], i64 [[B]])
+; CHECK-NEXT:    store { i64, i1 } [[TMP6]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret { i64, i1 } [[RES]]
+;
+  %res = call { i64, i1 } @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
+  ret { i64, i1 } %res
+}
+define {i64, i1} @test_umul_with_overflow(i64 %a, i64 %b) #0 {
+; CHECK-LABEL: define { i64, i1 } @test_umul_with_overflow(
+; CHECK-SAME: i64 [[A:%.*]], i64 [[B:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = or i64 [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ne i64 [[TMP3]], 0
+; CHECK-NEXT:    [[TMP5:%.*]] = insertvalue { i64, i1 } poison, i64 [[TMP3]], 0
+; CHECK-NEXT:    [[TMP6:%.*]] = insertvalue { i64, i1 } [[TMP5]], i1 [[TMP4]], 1
+; CHECK-NEXT:    [[RES:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 [[A]], i64 [[B]])
+; CHECK-NEXT:    store { i64, i1 } [[TMP6]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret { i64, i1 } [[RES]]
+;
+  %res = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
+  ret { i64, i1 } %res
+}
+define {i64, i1} @test_ssub_with_overflow(i64 %a, i64 %b) #0 {
+; CHECK-LABEL: define { i64, i1 } @test_ssub_with_overflow(
+; CHECK-SAME: i64 [[A:%.*]], i64 [[B:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = or i64 [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ne i64 [[TMP3]], 0
+; CHECK-NEXT:    [[TMP5:%.*]] = insertvalue { i64, i1 } poison, i64 [[TMP3]], 0
+; CHECK-NEXT:    [[TMP6:%.*]] = insertvalue { i64, i1 } [[TMP5]], i1 [[TMP4]], 1
+; CHECK-NEXT:    [[RES:%.*]] = call { i64, i1 } @llvm.ssub.with.overflow.i64(i64 [[A]], i64 [[B]])
+; CHECK-NEXT:    store { i64, i1 } [[TMP6]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret { i64, i1 } [[RES]]
+;
+  %res = call { i64, i1 } @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
+  ret { i64, i1 } %res
+}
+define {i64, i1} @test_usub_with_overflow(i64 %a, i64 %b) #0 {
+; CHECK-LABEL: define { i64, i1 } @test_usub_with_overflow(
+; CHECK-SAME: i64 [[A:%.*]], i64 [[B:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = or i64 [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ne i64 [[TMP3]], 0
+; CHECK-NEXT:    [[TMP5:%.*]] = insertvalue { i64, i1 } poison, i64 [[TMP3]], 0
+; CHECK-NEXT:    [[TMP6:%.*]] = insertvalue { i64, i1 } [[TMP5]], i1 [[TMP4]], 1
+; CHECK-NEXT:    [[RES:%.*]] = call { i64, i1 } @llvm.usub.with.overflow.i64(i64 [[A]], i64 [[B]])
+; CHECK-NEXT:    store { i64, i1 } [[TMP6]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret { i64, i1 } [[RES]]
+;
+  %res = call { i64, i1 } @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
+  ret { i64, i1 } %res
+}
+
+define {<4 x i32>, <4 x i1>} @test_sadd_with_overflow_vec(<4 x i32> %a, <4 x i32> %b) #0 {
+; CHECK-LABEL: define { <4 x i32>, <4 x i1> } @test_sadd_with_overflow_vec(
+; CHECK-SAME: <4 x i32> [[A:%.*]], <4 x i32> [[B:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <4 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 16) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = or <4 x i32> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ne <4 x i32> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP5:%.*]] = insertvalue { <4 x i32>, <4 x i1> } poison, <4 x i32> [[TMP3]], 0
+; CHECK-NEXT:    [[TMP6:%.*]] = insertvalue { <4 x i32>, <4 x i1> } [[TMP5]], <4 x i1> [[TMP4]], 1
+; CHECK-NEXT:    [[RES:%.*]] = call { <4 x i32>, <4 x i1> } @llvm.sadd.with.overflow.v4i32(<4 x i32> [[A]], <4 x i32> [[B]])
+; CHECK-NEXT:    store { <4 x i32>, <4 x i1> } [[TMP6]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret { <4 x i32>, <4 x i1> } [[RES]]
+;
+  %res = call { <4 x i32>, <4 x i1> } @llvm.sadd.with.overflow.v4i32(<4 x i32> %a, <4 x i32> %b)
+  ret { <4 x i32>, <4 x i1> } %res
+}
+
+attributes #0 = { sanitize_memory }
diff --git a/llvm/test/Instrumentation/MemorySanitizer/saturating.ll b/llvm/test/Instrumentation/MemorySanitizer/saturating.ll
new file mode 100644
index 00000000000000..b707b3160d9e56
--- /dev/null
+++ b/llvm/test/Instrumentation/MemorySanitizer/saturating.ll
@@ -0,0 +1,120 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4
+; RUN: opt %s -S -passes=msan 2>&1 | FileCheck %s
+
+target datalayout = "e-m:o-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+define i64 @test_sadd_sat(i64 %a, i64 %b) #0 {
+; CHECK-LABEL: define i64 @test_sadd_sat(
+; CHECK-SAME: i64 [[A:%.*]], i64 [[B:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or i64 [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or i64 [[_MSPROP]], 0
+; CHECK-NEXT:    [[RES:%.*]] = call i64 @llvm.sadd.sat.i64(i64 [[A]], i64 [[B]])
+; CHECK-NEXT:    store i64 [[_MSPROP1]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i64 [[RES]]
+;
+  %res = call i64 @llvm.sadd.sat(i64 %a, i64 %b)
+  ret i64 %res
+}
+
+define i64 @test_uadd_sat(i64 %a, i64 %b) #0 {
+; CHECK-LABEL: define i64 @test_uadd_sat(
+; CHECK-SAME: i64 [[A:%.*]], i64 [[B:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or i64 [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or i64 [[_MSPROP]], 0
+; CHECK-NEXT:    [[RES:%.*]] = call i64 @llvm.uadd.sat.i64(i64 [[A]], i64 [[B]])
+; CHECK-NEXT:    store i64 [[_MSPROP1]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i64 [[RES]]
+;
+  %res = call i64 @llvm.uadd.sat(i64 %a, i64 %b)
+  ret i64 %res
+}
+
+define i64 @test_ssub_sat(i64 %a, i64 %b) #0 {
+; CHECK-LABEL: define i64 @test_ssub_sat(
+; CHECK-SAME: i64 [[A:%.*]], i64 [[B:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or i64 [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or i64 [[_MSPROP]], 0
+; CHECK-NEXT:    [[RES:%.*]] = call i64 @llvm.ssub.sat.i64(i64 [[A]], i64 [[B]])
+; CHECK-NEXT:    store i64 [[_MSPROP1]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i64 [[RES]]
+;
+  %res = call i64 @llvm.ssub.sat(i64 %a, i64 %b)
+  ret i64 %res
+}
+
+define i64 @test_usub_sat(i64 %a, i64 %b) #0 {
+; CHECK-LABEL: define i64 @test_usub_sat(
+; CHECK-SAME: i64 [[A:%.*]], i64 [[B:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or i64 [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or i64 [[_MSPROP]], 0
+; CHECK-NEXT:    [[RES:%.*]] = call i64 @llvm.usub.sat.i64(i64 [[A]], i64 [[B]])
+; CHECK-NEXT:    store i64 [[_MSPROP1]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i64 [[RES]]
+;
+  %res = call i64 @llvm.usub.sat(i64 %a, i64 %b)
+  ret i64 %res
+}
+
+define i64 @test_sshl_sat(i64 %a, i64 %b) #0 {
+; CHECK-LABEL: define i64 @test_sshl_sat(
+; CHECK-SAME: i64 [[A:%.*]], i64 [[B:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or i64 [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or i64 [[_MSPROP]], 0
+; CHECK-NEXT:    [[RES:%.*]] = call i64 @llvm.sshl.sat.i64(i64 [[A]], i64 [[B]])
+; CHECK-NEXT:    store i64 [[_MSPROP1]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i64 [[RES]]
+;
+  %res = call i64 @llvm.sshl.sat(i64 %a, i64 %b)
+  ret i64 %res
+}
+
+define i64 @test_ushl_sat(i64 %a, i64 %b) #0 {
+; CHECK-LABEL: define i64 @test_ushl_sat(
+; CHECK-SAME: i64 [[A:%.*]], i64 [[B:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or i64 [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or i64 [[_MSPROP]], 0
+; CHECK-NEXT:    [[RES:%.*]] = call i64 @llvm.ushl.sat.i64(i64 [[A]], i64 [[B]])
+; CHECK-NEXT:    store i64 [[_MSPROP1]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i64 [[RES]]
+;
+  %res = call i64 @llvm.ushl.sat(i64 %a, i64 %b)
+  ret i64 %res
+}
+
+define <4 x i32> @test_sadd_sat_vec(<4 x i32> %a, <4 x i32> %b) #0 {
+; CHECK-LABEL: define <4 x i32> @test_sadd_sat_vec(
+; CHECK-SAME: <4 x i32> [[A:%.*]], <4 x i32> [[B:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <4 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 16) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <4 x i32> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <4 x i32> [[_MSPROP]], zeroinitializer
+; CHECK-NEXT:    [[RES:%.*]] = call <4 x i32> @llvm.sadd.sat.v4i32(<4 x i32> [[A]], <4 x i32> [[B]])
+; CHECK-NEXT:    store <4 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <4 x i32> [[RES]]
+;
+  %res = call <4 x i32> @llvm.sadd.sat.v4i32(<4 x i32> %a, <4 x i32> %b)
+  ret <4 x i32> %res
+}
+
+
+attributes #0 = { sanitize_memory }