[WebAssembly] Don't stackify multi-def instructions - #200429

alexcrichton · 2026-05-29T15:12:46Z

This commit updates the WebAssemblyRegStackify.cpp pass to specifically exclude attempting to stackify the first def of a multi-def instruction. As the previous comments indicate this is possible to do in some situations, but the current logic is incomplete and has led to miscompilations such as #98323 and #199910. One option would be to make the logic more robust, but in lieu of that in the meantime the change here is to completely disable stackification in these situations. This provides at least a "known working" base to build on later and fixes the known regressions around this.

Closes #98323
Closes #199910

This commit updates the `WebAssemblyRegStackify.cpp` pass to specifically exclude attempting to stackify the first def of a multi-def instruction. As the previous comments indicate this is possible to do in some situations, but the current logic is incomplete and has led to miscompilations such as llvm#98323 and llvm#199910. One option would be to make the logic more robust, but in lieu of that in the meantime the change here is to completely disable stackification in these situations. This provides at least a "known working" base to build on later and fixes the known regressions around this. Closes llvm#98323 Closes llvm#199910

llvmorg-github-actions · 2026-05-29T15:36:51Z

@llvm/pr-subscribers-backend-webassembly

Author: Alex Crichton (alexcrichton)

Changes

This commit updates the WebAssemblyRegStackify.cpp pass to specifically exclude attempting to stackify the first def of a multi-def instruction. As the previous comments indicate this is possible to do in some situations, but the current logic is incomplete and has led to miscompilations such as #98323 and #199910. One option would be to make the logic more robust, but in lieu of that in the meantime the change here is to completely disable stackification in these situations. This provides at least a "known working" base to build on later and fixes the known regressions around this.

Closes #98323
Closes #199910

Patch is 40.70 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/200429.diff

6 Files Affected:

(modified) llvm/lib/Target/WebAssembly/WebAssemblyRegStackify.cpp (+8-30)
(added) llvm/test/CodeGen/WebAssembly/multivalue-do-not-stackify.ll (+33)
(modified) llvm/test/CodeGen/WebAssembly/multivalue-stackify.ll (+204-118)
(modified) llvm/test/CodeGen/WebAssembly/multivalue.ll (+30-8)
(modified) llvm/test/CodeGen/WebAssembly/multivalue_libcall.ll (+12)
(modified) llvm/test/CodeGen/WebAssembly/wide-arithmetic.ll (+40-7)

diff --git a/llvm/lib/Target/WebAssembly/WebAssemblyRegStackify.cpp b/llvm/lib/Target/WebAssembly/WebAssemblyRegStackify.cpp
index 9015ceab87fb7..69f66858c20d1 100644
--- a/llvm/lib/Target/WebAssembly/WebAssemblyRegStackify.cpp
+++ b/llvm/lib/Target/WebAssembly/WebAssemblyRegStackify.cpp
@@ -355,38 +355,16 @@ static bool isSafeToMove(const MachineOperand *Def, const MachineOperand *Use,
   assert(DefI->getParent() == Insert->getParent());
   assert(UseI->getParent() == Insert->getParent());
 
-  // The first def of a multivalue instruction can be stackified by moving,
-  // since the later defs can always be placed into locals if necessary. Later
-  // defs can only be stackified if all previous defs are already stackified
-  // since ExplicitLocals will not know how to place a def in a local if a
-  // subsequent def is stackified. But only one def can be stackified by moving
-  // the instruction, so it must be the first one.
-  //
-  // TODO: This could be loosened to be the first *live* def, but care would
-  // have to be taken to ensure the drops of the initial dead defs can be
-  // placed. This would require checking that no previous defs are used in the
-  // same instruction as subsequent defs.
-  if (Def != DefI->defs().begin())
+  // For now avoid stackifying any multi-def instructions. While it's
+  // theoretically possible to do so for the first def in some cases this has
+  // historically led to bugs such as #199910 and #98323. For now this
+  // conservatively skips all multi-def instructions as a consequence. Note that
+  // multi-def instructions are expected to be not all that common so this in
+  // theory doesn't have a massive impact, but nevertheless this'd still be
+  // something to optimize better in the future.
+  if (DefI->getNumExplicitDefs() > 1)
     return false;
 
-  // If any subsequent def is used prior to the current value by the same
-  // instruction in which the current value is used, we cannot
-  // stackify. Stackifying in this case would require that def moving below the
-  // current def in the stack, which cannot be achieved, even with locals.
-  // Also ensure we don't sink the def past any other prior uses.
-  for (const auto &SubsequentDef : drop_begin(DefI->defs())) {
-    auto I = std::next(MachineBasicBlock::const_iterator(DefI));
-    auto E = std::next(MachineBasicBlock::const_iterator(UseI));
-    for (; I != E; ++I) {
-      for (const auto &PriorUse : I->uses()) {
-        if (&PriorUse == Use)
-          break;
-        if (PriorUse.isReg() && SubsequentDef.getReg() == PriorUse.getReg())
-          return false;
-      }
-    }
-  }
-
   // If moving is a semantic nop, it is always allowed
   const MachineBasicBlock *MBB = DefI->getParent();
   auto NextI = std::next(MachineBasicBlock::const_iterator(DefI));
diff --git a/llvm/test/CodeGen/WebAssembly/multivalue-do-not-stackify.ll b/llvm/test/CodeGen/WebAssembly/multivalue-do-not-stackify.ll
new file mode 100644
index 0000000000000..53b36fb3bde0c
--- /dev/null
+++ b/llvm/test/CodeGen/WebAssembly/multivalue-do-not-stackify.ll
@@ -0,0 +1,33 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+
+; RUN: llc < %s -verify-machineinstrs -mattr=+multivalue -target-abi=experimental-mv -O2 | FileCheck %s
+
+target triple = "wasm32-unknown-unknown"
+
+; Regression test for #98323 where attempting to stackify the call to `@foo`
+; historically led to a miscompile.
+
+define i64 @test() {
+; CHECK-LABEL: test:
+; CHECK:         .functype test () -> (i64)
+; CHECK-NEXT:    .local i64, i64
+; CHECK-NEXT:  # %bb.0: # %entry
+; CHECK-NEXT:    call foo
+; CHECK-NEXT:    local.set 1
+; CHECK-NEXT:    local.set 0
+; CHECK-NEXT:    i64.const 42
+; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 0
+; CHECK-NEXT:    i64.eqz
+; CHECK-NEXT:    i64.select
+; CHECK-NEXT:    # fallthrough-return
+entry:
+  %pair = call { i64, i64 } @foo()
+  %v0 = extractvalue { i64, i64 } %pair, 0
+  %1 = icmp eq i64 %v0, 0
+  %v1 = extractvalue { i64, i64 } %pair, 1
+  %_0.sroa.0.0 = select i1 %1, i64 42, i64 %v1
+  ret i64 %_0.sroa.0.0
+}
+
+declare { i64, i64 } @foo()
diff --git a/llvm/test/CodeGen/WebAssembly/multivalue-stackify.ll b/llvm/test/CodeGen/WebAssembly/multivalue-stackify.ll
index 0b5a304589aa6..82a8ea739493b 100644
--- a/llvm/test/CodeGen/WebAssembly/multivalue-stackify.ll
+++ b/llvm/test/CodeGen/WebAssembly/multivalue-stackify.ll
@@ -47,9 +47,12 @@ define void @f3() {
 define void @f12() {
 ; CHECK-LABEL: f12:
 ; CHECK:         .functype f12 () -> ()
+; CHECK-NEXT:    .local i32
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    call op_0_to_2
 ; CHECK-NEXT:    drop
+; CHECK-NEXT:    local.set 0
+; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    call op_1_to_0
 ; CHECK-NEXT:    # fallthrough-return
   %t0 = call {i32, i32} @op_0_to_2()
@@ -82,7 +85,8 @@ define void @f14() {
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    call op_0_to_2
 ; CHECK-NEXT:    drop
-; CHECK-NEXT:    local.tee 0
+; CHECK-NEXT:    local.set 0
+; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    call op_2_to_0
 ; CHECK-NEXT:    # fallthrough-return
@@ -96,8 +100,13 @@ define void @f14() {
 define void @f15() {
 ; CHECK-LABEL: f15:
 ; CHECK:         .functype f15 () -> ()
+; CHECK-NEXT:    .local i32, i32
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    call op_0_to_2
+; CHECK-NEXT:    local.set 1
+; CHECK-NEXT:    local.set 0
+; CHECK-NEXT:    local.get 0
+; CHECK-NEXT:    local.get 1
 ; CHECK-NEXT:    call op_2_to_0
 ; CHECK-NEXT:    # fallthrough-return
   %t0 = call {i32, i32} @op_0_to_2()
@@ -148,10 +157,13 @@ define void @f17() {
 define void @f25() {
 ; CHECK-LABEL: f25:
 ; CHECK:         .functype f25 () -> ()
+; CHECK-NEXT:    .local i32
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    call op_0_to_3
 ; CHECK-NEXT:    drop
 ; CHECK-NEXT:    drop
+; CHECK-NEXT:    local.set 0
+; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    call op_1_to_0
 ; CHECK-NEXT:    # fallthrough-return
   %t0 = call {i32, i32, i32} @op_0_to_3()
@@ -204,7 +216,8 @@ define void @f28() {
 ; CHECK-NEXT:    call op_0_to_3
 ; CHECK-NEXT:    drop
 ; CHECK-NEXT:    drop
-; CHECK-NEXT:    local.tee 0
+; CHECK-NEXT:    local.set 0
+; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    call op_2_to_0
 ; CHECK-NEXT:    # fallthrough-return
@@ -218,9 +231,14 @@ define void @f28() {
 define void @f29() {
 ; CHECK-LABEL: f29:
 ; CHECK:         .functype f29 () -> ()
+; CHECK-NEXT:    .local i32, i32
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    call op_0_to_3
 ; CHECK-NEXT:    drop
+; CHECK-NEXT:    local.set 1
+; CHECK-NEXT:    local.set 0
+; CHECK-NEXT:    local.get 0
+; CHECK-NEXT:    local.get 1
 ; CHECK-NEXT:    call op_2_to_0
 ; CHECK-NEXT:    # fallthrough-return
   %t0 = call {i32, i32, i32} @op_0_to_3()
@@ -233,12 +251,14 @@ define void @f29() {
 define void @f30() {
 ; CHECK-LABEL: f30:
 ; CHECK:         .functype f30 () -> ()
-; CHECK-NEXT:    .local i32
+; CHECK-NEXT:    .local i32, i32
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    call op_0_to_3
-; CHECK-NEXT:    local.set 0
+; CHECK-NEXT:    local.set 1
 ; CHECK-NEXT:    drop
+; CHECK-NEXT:    local.set 0
 ; CHECK-NEXT:    local.get 0
+; CHECK-NEXT:    local.get 1
 ; CHECK-NEXT:    call op_2_to_0
 ; CHECK-NEXT:    # fallthrough-return
   %t0 = call {i32, i32, i32} @op_0_to_3()
@@ -371,13 +391,15 @@ define void @f36() {
 define void @f129() {
 ; CHECK-LABEL: f129:
 ; CHECK:         .functype f129 () -> ()
-; CHECK-NEXT:    .local i32
+; CHECK-NEXT:    .local i32, i32
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    call op_0_to_2
+; CHECK-NEXT:    local.set 1
 ; CHECK-NEXT:    local.set 0
-; CHECK-NEXT:    call op_1_to_0
 ; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    call op_1_to_0
+; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    call op_1_to_0
 ; CHECK-NEXT:    # fallthrough-return
   %t0 = call {i32, i32} @op_0_to_2()
   %t1 = extractvalue {i32, i32} %t0, 0
@@ -393,11 +415,12 @@ define void @f131() {
 ; CHECK-NEXT:    .local i32, i32
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    call op_0_to_2
+; CHECK-NEXT:    local.set 1
 ; CHECK-NEXT:    local.set 0
-; CHECK-NEXT:    local.tee 1
+; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    call op_1_to_0
-; CHECK-NEXT:    local.get 1
 ; CHECK-NEXT:    local.get 0
+; CHECK-NEXT:    local.get 1
 ; CHECK-NEXT:    call op_2_to_0
 ; CHECK-NEXT:    # fallthrough-return
   %t0 = call {i32, i32} @op_0_to_2()
@@ -415,11 +438,12 @@ define void @f132() {
 ; CHECK-NEXT:    .local i32, i32
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    call op_0_to_2
+; CHECK-NEXT:    local.set 1
 ; CHECK-NEXT:    local.set 0
-; CHECK-NEXT:    local.tee 1
-; CHECK-NEXT:    call op_1_to_0
 ; CHECK-NEXT:    local.get 0
+; CHECK-NEXT:    call op_1_to_0
 ; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    call op_2_to_0
 ; CHECK-NEXT:    # fallthrough-return
   %t0 = call {i32, i32} @op_0_to_2()
@@ -434,13 +458,15 @@ define void @f132() {
 define void @f133() {
 ; CHECK-LABEL: f133:
 ; CHECK:         .functype f133 () -> ()
-; CHECK-NEXT:    .local i32
+; CHECK-NEXT:    .local i32, i32
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    call op_0_to_2
+; CHECK-NEXT:    local.set 1
 ; CHECK-NEXT:    local.set 0
-; CHECK-NEXT:    call op_1_to_0
-; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    local.get 0
+; CHECK-NEXT:    call op_1_to_0
+; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 1
 ; CHECK-NEXT:    call op_2_to_0
 ; CHECK-NEXT:    # fallthrough-return
   %t0 = call {i32, i32} @op_0_to_2()
@@ -548,11 +574,12 @@ define void @f155() {
 ; CHECK-NEXT:    .local i32, i32
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    call op_0_to_2
+; CHECK-NEXT:    local.set 1
 ; CHECK-NEXT:    local.set 0
-; CHECK-NEXT:    local.tee 1
-; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    call op_2_to_0
 ; CHECK-NEXT:    local.get 0
+; CHECK-NEXT:    local.get 0
+; CHECK-NEXT:    call op_2_to_0
+; CHECK-NEXT:    local.get 1
 ; CHECK-NEXT:    call op_1_to_0
 ; CHECK-NEXT:    # fallthrough-return
   %t0 = call {i32, i32} @op_0_to_2()
@@ -570,13 +597,14 @@ define void @f159() {
 ; CHECK-NEXT:    .local i32, i32
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    call op_0_to_2
+; CHECK-NEXT:    local.set 1
 ; CHECK-NEXT:    local.set 0
-; CHECK-NEXT:    local.tee 1
-; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    call op_2_to_0
 ; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    call op_2_to_0
+; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    call op_2_to_0
 ; CHECK-NEXT:    # fallthrough-return
   %t0 = call {i32, i32} @op_0_to_2()
   %t1 = extractvalue {i32, i32} %t0, 0
@@ -594,11 +622,12 @@ define void @f167() {
 ; CHECK-NEXT:    .local i32, i32
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    call op_0_to_2
+; CHECK-NEXT:    local.set 1
 ; CHECK-NEXT:    local.set 0
-; CHECK-NEXT:    local.tee 1
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call op_2_to_0
 ; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    call op_2_to_0
+; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    call op_1_to_0
 ; CHECK-NEXT:    # fallthrough-return
   %t0 = call {i32, i32} @op_0_to_2()
@@ -613,13 +642,15 @@ define void @f167() {
 define void @f168() {
 ; CHECK-LABEL: f168:
 ; CHECK:         .functype f168 () -> ()
-; CHECK-NEXT:    .local i32
+; CHECK-NEXT:    .local i32, i32
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    call op_0_to_2
+; CHECK-NEXT:    local.set 1
 ; CHECK-NEXT:    local.set 0
 ; CHECK-NEXT:    local.get 0
+; CHECK-NEXT:    local.get 1
 ; CHECK-NEXT:    call op_2_to_0
-; CHECK-NEXT:    local.get 0
+; CHECK-NEXT:    local.get 1
 ; CHECK-NEXT:    call op_1_to_0
 ; CHECK-NEXT:    # fallthrough-return
   %t0 = call {i32, i32} @op_0_to_2()
@@ -637,12 +668,13 @@ define void @f171() {
 ; CHECK-NEXT:    .local i32, i32
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    call op_0_to_2
+; CHECK-NEXT:    local.set 1
 ; CHECK-NEXT:    local.set 0
-; CHECK-NEXT:    local.tee 1
 ; CHECK-NEXT:    local.get 0
+; CHECK-NEXT:    local.get 1
 ; CHECK-NEXT:    call op_2_to_0
-; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    call op_2_to_0
 ; CHECK-NEXT:    # fallthrough-return
   %t0 = call {i32, i32} @op_0_to_2()
@@ -777,14 +809,16 @@ define void @f195() {
 define void @f291() {
 ; CHECK-LABEL: f291:
 ; CHECK:         .functype f291 () -> ()
-; CHECK-NEXT:    .local i32
+; CHECK-NEXT:    .local i32, i32
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    call op_0_to_3
 ; CHECK-NEXT:    drop
+; CHECK-NEXT:    local.set 1
 ; CHECK-NEXT:    local.set 0
-; CHECK-NEXT:    call op_1_to_0
 ; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    call op_1_to_0
+; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    call op_1_to_0
 ; CHECK-NEXT:    # fallthrough-return
   %t0 = call {i32, i32, i32} @op_0_to_3()
   %t1 = extractvalue {i32, i32, i32} %t0, 0
@@ -797,14 +831,16 @@ define void @f291() {
 define void @f292() {
 ; CHECK-LABEL: f292:
 ; CHECK:         .functype f292 () -> ()
-; CHECK-NEXT:    .local i32
+; CHECK-NEXT:    .local i32, i32
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    call op_0_to_3
-; CHECK-NEXT:    local.set 0
+; CHECK-NEXT:    local.set 1
 ; CHECK-NEXT:    drop
-; CHECK-NEXT:    call op_1_to_0
+; CHECK-NEXT:    local.set 0
 ; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    call op_1_to_0
+; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    call op_1_to_0
 ; CHECK-NEXT:    # fallthrough-return
   %t0 = call {i32, i32, i32} @op_0_to_3()
   %t1 = extractvalue {i32, i32, i32} %t0, 0
@@ -821,11 +857,12 @@ define void @f294() {
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    call op_0_to_3
 ; CHECK-NEXT:    drop
+; CHECK-NEXT:    local.set 1
 ; CHECK-NEXT:    local.set 0
-; CHECK-NEXT:    local.tee 1
+; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    call op_1_to_0
-; CHECK-NEXT:    local.get 1
 ; CHECK-NEXT:    local.get 0
+; CHECK-NEXT:    local.get 1
 ; CHECK-NEXT:    call op_2_to_0
 ; CHECK-NEXT:    # fallthrough-return
   %t0 = call {i32, i32, i32} @op_0_to_3()
@@ -843,12 +880,13 @@ define void @f295() {
 ; CHECK-NEXT:    .local i32, i32
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    call op_0_to_3
-; CHECK-NEXT:    local.set 0
+; CHECK-NEXT:    local.set 1
 ; CHECK-NEXT:    drop
-; CHECK-NEXT:    local.tee 1
+; CHECK-NEXT:    local.set 0
+; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    call op_1_to_0
-; CHECK-NEXT:    local.get 1
 ; CHECK-NEXT:    local.get 0
+; CHECK-NEXT:    local.get 1
 ; CHECK-NEXT:    call op_2_to_0
 ; CHECK-NEXT:    # fallthrough-return
   %t0 = call {i32, i32, i32} @op_0_to_3()
@@ -867,11 +905,12 @@ define void @f296() {
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    call op_0_to_3
 ; CHECK-NEXT:    drop
+; CHECK-NEXT:    local.set 1
 ; CHECK-NEXT:    local.set 0
-; CHECK-NEXT:    local.tee 1
-; CHECK-NEXT:    call op_1_to_0
 ; CHECK-NEXT:    local.get 0
+; CHECK-NEXT:    call op_1_to_0
 ; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    call op_2_to_0
 ; CHECK-NEXT:    # fallthrough-return
   %t0 = call {i32, i32, i32} @op_0_to_3()
@@ -886,14 +925,16 @@ define void @f296() {
 define void @f297() {
 ; CHECK-LABEL: f297:
 ; CHECK:         .functype f297 () -> ()
-; CHECK-NEXT:    .local i32
+; CHECK-NEXT:    .local i32, i32
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    call op_0_to_3
 ; CHECK-NEXT:    drop
+; CHECK-NEXT:    local.set 1
 ; CHECK-NEXT:    local.set 0
-; CHECK-NEXT:    call op_1_to_0
-; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    local.get 0
+; CHECK-NEXT:    call op_1_to_0
+; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 1
 ; CHECK-NEXT:    call op_2_to_0
 ; CHECK-NEXT:    # fallthrough-return
   %t0 = call {i32, i32, i32} @op_0_to_3()
@@ -908,14 +949,16 @@ define void @f297() {
 define void @f298() {
 ; CHECK-LABEL: f298:
 ; CHECK:         .functype f298 () -> ()
-; CHECK-NEXT:    .local i32, i32
+; CHECK-NEXT:    .local i32, i32, i32
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    call op_0_to_3
+; CHECK-NEXT:    local.set 2
 ; CHECK-NEXT:    local.set 1
 ; CHECK-NEXT:    local.set 0
-; CHECK-NEXT:    call op_1_to_0
 ; CHECK-NEXT:    local.get 0
+; CHECK-NEXT:    call op_1_to_0
 ; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 2
 ; CHECK-NEXT:    call op_2_to_0
 ; CHECK-NEXT:    # fallthrough-return
   %t0 = call {i32, i32, i32} @op_0_to_3()
@@ -933,12 +976,13 @@ define void @f299() {
 ; CHECK-NEXT:    .local i32, i32
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    call op_0_to_3
-; CHECK-NEXT:    local.set 0
+; CHECK-NEXT:    local.set 1
 ; CHECK-NEXT:    drop
-; CHECK-NEXT:    local.tee 1
-; CHECK-NEXT:    call op_1_to_0
+; CHECK-NEXT:    local.set 0
 ; CHECK-NEXT:    local.get 0
+; CHECK-NEXT:    call op_1_to_0
 ; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    call op_2_to_0
 ; CHECK-NEXT:    # fallthrough-return
   %t0 = call {i32, i32, i32} @op_0_to_3()
@@ -953,14 +997,16 @@ define void @f299() {
 define void @f300() {
 ; CHECK-LABEL: f300:
 ; CHECK:         .functype f300 () -> ()
-; CHECK-NEXT:    .local i32, i32
+; CHECK-NEXT:    .local i32, i32, i32
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    call op_0_to_3
+; CHECK-NEXT:    local.set 2
 ; CHECK-NEXT:    local.set 1
 ; CHECK-NEXT:    local.set 0
+; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    call op_1_to_0
+; CHECK-NEXT:    local.get 2
 ; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    call op_2_to_0
 ; CHECK-NEXT:    # fallthrough-return
   %t0 = call {i32, i32, i32} @op_0_to_3()
@@ -975,14 +1021,16 @@ define void @f300() {
 define void @f301() {
 ; CHECK-LABEL: f301:
 ; CHECK:         .functype f301 () -> ()
-; CHECK-NEXT:    .local i32
+; CHECK-NEXT:    .local i32, i32
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    call op_0_to_3
-; CHECK-NEXT:    local.set 0
+; CHECK-NEXT:    local.set 1
 ; CHECK-NEXT:    drop
-; CHECK-NEXT:    call op_1_to_0
-; CHECK-NEXT:    local.get 0
+; CHECK-NEXT:    local.set 0
 ; CHECK-NEXT:    local.get 0
+; CHECK-NEXT:    call op_1_to_0
+; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 1
 ; CHECK-NEXT:    call op_2_to_0
 ; CHECK-NEXT:    # fallthrough-return
   %t0 = call {i32, i32, i32} @op_0_to_3()
@@ -1473,11 +1521,12 @@ define void @f327() {
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    call op_0_to_3
 ; CHECK-NEXT:    drop
+; CHECK-NEXT:    local.set 1
 ; CHECK-NEXT:    local.set 0
-; CHECK-NEXT:    local.tee 1
-; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    call op_2_to_0
 ; CHECK-NEXT:    local.get 0
+; CHECK-NEXT:    local.get 0
+; CHECK-NEXT:    call op_2_to_0
+; CHECK-NEXT:    local.get 1
 ; CHECK-NEXT:    call op_1_to_0
 ; CHECK-NEXT:    # fallthrough-return
   %t0 = call {i32, i32, i32} @op_0_to_3()
@@ -1495,12 +1544,13 @@ define void @f328() {
 ; CHECK-NEXT:    .local i32, i32
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    call op_0_to_3
-; CHECK-NEXT:    local.set 0
+; CHECK-NEXT:    local.set 1
 ; CHECK-NEXT:    drop
-; CHECK-NEXT:    local.tee 1
-; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    call op_2_to_0
+; CHECK-NEXT:    local.set 0
+; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    local.get 0
+; CHECK-NEXT:    call op_2_to_0
+; CHECK-NEXT:    local.get 1
 ; CHECK-NEXT:    call op_1_to_0
 ; CHECK-NEXT:    # fallthrough-return
   %t0 = call {i32, i32, i32} @op_0_to_3()
@@ -1519,13 +1569,14 @@ define void @f333() {
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    call op_0_to_3
 ; CHECK-NEXT:    drop
+; CHECK-NEXT:    local.set 1
 ; CHECK-NEXT:    local.set 0
-; CHECK-NEXT:    local.tee 1
-; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    call op_2_to_0
 ; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    call op_2_to_0
+; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    call op_2_to_0
 ; CHECK-NEXT:    # fallthrough-return
   %t0 = call {i32, i32, i32} @op_0_to_3()
   %t1 = extractvalue {i32, i32, i32} %t0, 0
@@ -1543,13 +1594,14 @@ define void @f334() {
 ; CHECK-NEXT:    .local i32, i32, i32
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    call op_0_to_3
+; CHECK-NEXT:    local.set 2
 ; CHECK-NEXT:    local.set 1
 ; CHECK-NEXT:    local.set 0
-; CHECK-NEXT:    local.tee 2
-; CHECK-NEXT:    local.get 2
-; CHECK-NEXT:    call op_2_to_0
 ; CHECK-NEXT:    local.get 0
+; CHECK-NEXT:    local.get 0
+; CHECK-NEXT:    call op_2_to_0
 ; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 2
 ; CHECK-NEXT:    call op_2_to_0
 ; CHECK-NEXT:    # fallthrough-return
   %t0 = call {i32, i32, i32} @op_0_to_3()
@@ -1568,13 +1620,14 @@ define void @f336() {
 ; CHECK-NEXT:    .local i32, i32, i32...
[truncated]

alexcrichton · 2026-06-08T23:42:46Z

@dschuff would you be able to merge this for me?

QuantumSegfault · 2026-06-08T23:44:05Z

I got it.

alexcrichton · 2026-06-08T23:44:48Z

Thanks!

llvm-ci · 2026-06-08T23:48:50Z

LLVM Buildbot has detected a new failure on builder release-noassertions-warnings running on google-integrate-b2 while building llvm at step 6 "build-default".

Full details are available at: https://lab.llvm.org/buildbot/#/builders/228/builds/1332

Here is the relevant piece of the build log for the reference

Step 6 (build-default) failure: cmake (failure)
...
39.354 [4526/34/4027] Building WasmSSAOps.h.inc...
39.355 [4525/34/4028] Building SymbolInterfaces.cpp.inc...
39.355 [4524/34/4029] Building SymbolInterfaces.h.inc...
39.360 [4523/34/4030] Building RegionKindInterface.cpp.inc...
39.366 [4522/34/4031] Building RegionKindInterface.h.inc...
39.367 [4521/34/4032] Building XeGPUTransformOps.cpp.inc...
39.367 [4520/34/4033] Building QuantStorageTypeInterface.cpp.inc...
39.369 [4519/34/4034] Building SymbolInterfacesAttrInterface.cpp.inc...
39.375 [4518/34/4035] Building XeGPUTransformOps.h.inc...
39.378 [4517/34/4036] Building CXX object lib/Target/WebAssembly/CMakeFiles/LLVMWebAssemblyCodeGen.dir/WebAssemblyRegStackify.cpp.o
FAILED: lib/Target/WebAssembly/CMakeFiles/LLVMWebAssemblyCodeGen.dir/WebAssemblyRegStackify.cpp.o 
CCACHE_CPP2=yes CCACHE_HASHDIR=yes CCACHE_SLOPPINESS=pch_defines,time_macros /usr/bin/ccache /usr/bin/c++ -D_GLIBCXX_USE_CXX11_ABI=1 -D_GNU_SOURCE -D__STDC_CONSTANT_MACROS -D__STDC_FORMAT_MACROS -D__STDC_LIMIT_MACROS -I/b/release-noassertions-warnings-build/build/lib/Target/WebAssembly -I/b/release-noassertions-warnings-build/llvm-project/llvm/lib/Target/WebAssembly -I/b/release-noassertions-warnings-build/build/include -I/b/release-noassertions-warnings-build/llvm-project/llvm/include -fPIC -fno-semantic-interposition -fvisibility-inlines-hidden -Werror -Werror=date-time -Werror=unguarded-availability-new -Wall -Wextra -Wno-unused-parameter -Wwrite-strings -Wcast-qual -Wmissing-field-initializers -pedantic -Wno-long-long -Wc++98-compat-extra-semi -Wimplicit-fallthrough -Wcovered-switch-default -Wno-noexcept-type -Wnon-virtual-dtor -Wdelete-non-virtual-dtor -Wsuggest-override -Wstring-conversion -Wno-pass-failed -Wmisleading-indentation -Wctad-maybe-unsupported -fdiagnostics-color -ffunction-sections -fdata-sections -Xclang -fno-pch-timestamp -O3 -DNDEBUG -std=c++17 -fvisibility=hidden -fno-exceptions -funwind-tables -fno-rtti -Winvalid-pch -Xclang -include-pch -Xclang /b/release-noassertions-warnings-build/build/lib/CodeGen/CMakeFiles/LLVMCodeGen.dir/cmake_pch.hxx.pch -Xclang -include -Xclang /b/release-noassertions-warnings-build/build/lib/CodeGen/CMakeFiles/LLVMCodeGen.dir/cmake_pch.hxx -MD -MT lib/Target/WebAssembly/CMakeFiles/LLVMWebAssemblyCodeGen.dir/WebAssemblyRegStackify.cpp.o -MF lib/Target/WebAssembly/CMakeFiles/LLVMWebAssemblyCodeGen.dir/WebAssemblyRegStackify.cpp.o.d -o lib/Target/WebAssembly/CMakeFiles/LLVMWebAssemblyCodeGen.dir/WebAssemblyRegStackify.cpp.o -c /b/release-noassertions-warnings-build/llvm-project/llvm/lib/Target/WebAssembly/WebAssemblyRegStackify.cpp
/b/release-noassertions-warnings-build/llvm-project/llvm/lib/Target/WebAssembly/WebAssemblyRegStackify.cpp:354:23: error: unused variable 'UseI' [-Werror,-Wunused-variable]
  354 |   const MachineInstr *UseI = Use->getParent();
      |                       ^~~~
1 error generated.
39.378 [4517/33/4037] Building SymbolInterfacesAttrInterface.h.inc...
39.380 [4517/32/4038] Building QuantStorageTypeInterface.h.inc...
39.380 [4517/31/4039] Building XeGPUEnums.cpp.inc...
39.381 [4517/30/4040] Building OpAsmAttrInterface.cpp.inc...
39.385 [4517/29/4041] Building X86.h.inc...
39.388 [4517/28/4042] Building X86Types.cpp.inc...
39.388 [4517/27/4043] Building OpAsmAttrInterface.h.inc...
39.390 [4517/26/4044] Building OpAsmOpInterface.cpp.inc...
39.390 [4517/25/4045] Linking CXX static library lib/libLLVMAnalysis.a
39.391 [4517/24/4046] Building XeGPUEnums.h.inc...
39.394 [4517/23/4047] Building AlignmentAttrInterface.cpp.inc...
39.394 [4517/22/4048] Building OpAsmOpInterface.h.inc...
39.396 [4517/21/4049] Building SPIRVAttrUtils.inc...
39.399 [4517/20/4050] Building OpAsmTypeInterface.h.inc...
39.400 [4517/19/4051] Building AlignmentAttrInterface.h.inc...
39.405 [4517/18/4052] Building CallInterfaces.cpp.inc...
39.406 [4517/17/4053] Building OpAsmTypeInterface.cpp.inc...
39.573 [4517/16/4054] Building X86GenFastISel.inc...
40.408 [4517/15/4055] Building X86GenGlobalISel.inc...
41.935 [4517/14/4056] Building X86GenDAGISel.inc...
42.377 [4517/13/4057] Building RISCVGenDAGISel.inc...
42.681 [4517/12/4058] Building RISCVGenInstrInfo.inc...
44.579 [4517/11/4059] Building RISCVGenGlobalISel.inc...
45.318 [4517/10/4060] Building AMDGPUGenCallingConv.inc...
45.498 [4517/9/4061] Building X86GenSubtargetInfo.inc...
46.318 [4517/8/4062] Building AMDGPUGenAsmWriter.inc...
46.489 [4517/7/4063] Building X86GenInstrInfo.inc...
50.823 [4517/6/4064] Building AMDGPUGenDAGISel.inc...
52.553 [4517/5/4065] Building AMDGPUGenGlobalISel.inc...
53.227 [4517/4/4066] Building AMDGPUGenInstrInfo.inc...
57.229 [4517/3/4067] Building AMDGPUGenAsmMatcher.inc...
59.803 [4517/2/4068] Building AMDGPUGenRegisterInfo.inc...
61.729 [4517/1/4069] Building AMDGPUGenRegisterBank.inc...

QuantumSegfault · 2026-06-08T23:52:06Z

@alexcrichton

For future reference, CI treats all warnings as errors. There must be no warnings when compiling.

Can you open a patch PR real quick, or would you prefer this reverted for now?

alexcrichton · 2026-06-08T23:56:12Z

Sorry about that! Posted a fix to #202464 and I don't think I saw this in development since it's a release-only unused variable (I was compiling with assertions on)

QuantumSegfault · 2026-06-09T00:05:00Z

Sorry about that! Posted a fix to #202464 and I don't think I saw this in development since it's a release-only unused variable (I was compiling with assertions on)

Ah, yeah. I had the same problem once. It's a shame the normal CI doesn't catch that. I guess the CI is already expensive enough. But a compile dry-run to check for would-be warnings in release could be helpful. But I suppose, pretty rare to come across. Not worth burdening every PR.

Fixes a mistake from llvm#200429.

Fixes a mistake from #200429.

This commit updates the `WebAssemblyRegStackify.cpp` pass to specifically exclude attempting to stackify the first def of a multi-def instruction. As the previous comments indicate this is possible to do in some situations, but the current logic is incomplete and has led to miscompilations such as llvm#98323 and llvm#199910. One option would be to make the logic more robust, but in lieu of that in the meantime the change here is to completely disable stackification in these situations. This provides at least a "known working" base to build on later and fixes the known regressions around this. Closes llvm#98323 Closes llvm#199910

Fixes a mistake from llvm#200429.

Backport: [WebAssembly] Don't stackify multi-def instructions (llvm#200429)

This commit updates the `WebAssemblyRegStackify.cpp` pass to specifically exclude attempting to stackify the first def of a multi-def instruction. As the previous comments indicate this is possible to do in some situations, but the current logic is incomplete and has led to miscompilations such as llvm#98323 and llvm#199910. One option would be to make the logic more robust, but in lieu of that in the meantime the change here is to completely disable stackification in these situations. This provides at least a "known working" base to build on later and fixes the known regressions around this. Closes llvm#98323 Closes llvm#199910 Backport to LLVM 21.1.2. Adapt the multivalue.ll REGS check to the older WebAssembly register-printer syntax. (cherry picked from commit b472674)

[WebAssembly] Don't stackify multi-def instructions (llvm#200429)

Adds the WebAssemblyRegStackify.cpp change from wasix-org/llvm-project a3d94d9 (backport of llvm/llvm-project#200429) to wasix-llvm.patch. The pass could stackify the first def of a multi-def instruction, which miscompiled overflow checks built on i64.mul_wide_s: the overflow flag came back set for products that fit, e.g. 3 * 7 reporting overflow while returning 21. Reachable from Rust via i64::overflowing_mul when built with -Ctarget-feature=+wide-arithmetic, and from C through __builtin_mul_overflow. Only the source hunk is taken; the commit's test-file updates are left out, matching the rest of this patch. Refs ECO-421, ECO-426. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

alexcrichton mentioned this pull request May 29, 2026

[WebAssembly] signed wide multiply overflow check miscompiled #199910

Closed

llvmorg-github-actions Bot added the backend:WebAssembly label May 29, 2026

dschuff approved these changes May 29, 2026

View reviewed changes

alexcrichton mentioned this pull request May 29, 2026

Define a new "wasm-multivalue" calling convention WebAssembly/tool-conventions#268

Open

QuantumSegfault merged commit b472674 into llvm:main Jun 8, 2026
12 checks passed

alexcrichton deleted the wasm-disable-stackify-multi-ret branch June 8, 2026 23:44

alexcrichton mentioned this pull request Jun 8, 2026

[WebAssembly] Fix unused variable in #200429 #202464

Merged

alexcrichton added a commit to alexcrichton/llvm-project that referenced this pull request Jun 9, 2026

[WebAssembly] Fix unused variable

084885b

Fixes a mistake from llvm#200429.

QuantumSegfault pushed a commit that referenced this pull request Jun 9, 2026

[WebAssembly] Fix unused variable in #200429 (#202464)

048e967

Fixes a mistake from #200429.

carlobertolli pushed a commit to carlobertolli/llvm-project that referenced this pull request Jun 11, 2026

[WebAssembly] Fix unused variable in llvm#200429 (llvm#202464)

788066d

Fixes a mistake from llvm#200429.

kilyanni mentioned this pull request Aug 5, 2026

Backport: [WebAssembly] Don't stackify multi-def instructions (#200429) wasix-org/llvm-project#2

Merged

Arshia001 added a commit to wasix-org/llvm-project that referenced this pull request Aug 6, 2026

Merge pull request #2 from wasix-org/fix/wide-arithmetic-overflow

98c00cb

Backport: [WebAssembly] Don't stackify multi-def instructions (llvm#200429)

Arshia001 added a commit to wasix-org/llvm-project that referenced this pull request Aug 6, 2026

Merge pull request #4 from wasix-org/fix/wide-arithmetic-overflow

a3d94d9

[WebAssembly] Don't stackify multi-def instructions (llvm#200429)

Arshia001 added a commit to wasix-org/llvm-project that referenced this pull request Aug 6, 2026

Merge pull request #3 from wasix-org/fix/wide-arithmetic-overflow-22

3f3e05c

[WebAssembly] Don't stackify multi-def instructions (llvm#200429)

Navigation Menu

Uh oh!

[WebAssembly] Don't stackify multi-def instructions - #200429