Skip to content

[WebAssembly] Optimize lowering of constant-sized memcpy and memset #163294

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
Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

[WebAssembly] Optimize lowering of constant-sized memcpy and memset #163294

wants to merge 1 commit into from

Conversation

dschuff
Copy link
Member

@dschuff dschuff commented Oct 13, 2025 •
edited
Loading

We currently emit a check that the size operand isn't zero, to avoid executing
the wasm memory.copy instruction when it would trap. But this isn't necessary
if the operand is a constant.

Fixes #163245

@dschuff
[WebAssembly] Optimize lowering of constant-sized memcpy and memset
f2e23d8 
We currently emit a check that the size operand isn't zero, to avoid executing
the wasm memory.copy instruction when it would trap. But this isn't necessary
if the operand is a constant.

Fixes 163245
@llvmbot
Copy link
Member

llvmbot commented Oct 13, 2025

@llvm/pr-subscribers-backend-webassembly

Author: Derek Schuff (dschuff)

Changes

We currently emit a check that the size operand isn't zero, to avoid executing
the wasm memory.copy instruction when it would trap. But this isn't necessary
if the operand is a constant.

Fixes 163245

Full diff: https://github.com/llvm/llvm-project/pull/163294.diff

3 Files Affected:

  • (modified) llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp (+45)
  • (modified) llvm/test/CodeGen/WebAssembly/bulk-memory.ll (+43-54)
  • (modified) llvm/test/CodeGen/WebAssembly/bulk-memory64.ll (+40-51)
diff --git a/llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp b/llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp
index 47c24fc27f1d6..f9739492e7fe3 100644
--- a/llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp
+++ b/llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp
@@ -601,6 +601,29 @@ static MachineBasicBlock *LowerMemcpy(MachineInstr &MI, DebugLoc DL,
   MachineOperand Src = MI.getOperand(3);
   MachineOperand Len = MI.getOperand(4);
 
+  // If the length is a constant, we don't actually need the check.
+  if (MachineInstr *Def = MRI.getVRegDef(Len.getReg())) {
+    if (Def->getOpcode() == WebAssembly::CONST_I32 ||
+        Def->getOpcode() == WebAssembly::CONST_I64) {
+      if (Def->getOperand(1).getImm() == 0) {
+        // A zero-length memcpy is a no-op.
+        MI.eraseFromParent();
+        return BB;
+      }
+      // A non-zero-length memcpy doesn't need a zero check.
+      unsigned MemoryCopy =
+          Int64 ? WebAssembly::MEMORY_COPY_A64 : WebAssembly::MEMORY_COPY_A32;
+      BuildMI(*BB, MI, DL, TII.get(MemoryCopy))
+          .add(DstMem)
+          .add(SrcMem)
+          .add(Dst)
+          .add(Src)
+          .add(Len);
+      MI.eraseFromParent();
+      return BB;
+    }
+  }
+
   // We're going to add an extra use to `Len` to test if it's zero; that
   // use shouldn't be a kill, even if the original use is.
   MachineOperand NoKillLen = Len;
@@ -669,6 +692,28 @@ static MachineBasicBlock *LowerMemset(MachineInstr &MI, DebugLoc DL,
   MachineOperand Val = MI.getOperand(2);
   MachineOperand Len = MI.getOperand(3);
 
+  // If the length is a constant, we don't actually need the check.
+  if (MachineInstr *Def = MRI.getVRegDef(Len.getReg())) {
+    if (Def->getOpcode() == WebAssembly::CONST_I32 ||
+        Def->getOpcode() == WebAssembly::CONST_I64) {
+      if (Def->getOperand(1).getImm() == 0) {
+        // A zero-length memset is a no-op.
+        MI.eraseFromParent();
+        return BB;
+      }
+      // A non-zero-length memset doesn't need a zero check.
+      unsigned MemoryFill =
+          Int64 ? WebAssembly::MEMORY_FILL_A64 : WebAssembly::MEMORY_FILL_A32;
+      BuildMI(*BB, MI, DL, TII.get(MemoryFill))
+          .add(Mem)
+          .add(Dst)
+          .add(Val)
+          .add(Len);
+      MI.eraseFromParent();
+      return BB;
+    }
+  }
+
   // We're going to add an extra use to `Len` to test if it's zero; that
   // use shouldn't be a kill, even if the original use is.
   MachineOperand NoKillLen = Len;
diff --git a/llvm/test/CodeGen/WebAssembly/bulk-memory.ll b/llvm/test/CodeGen/WebAssembly/bulk-memory.ll
index ae170d757a305..81c0ed39cf3c7 100644
--- a/llvm/test/CodeGen/WebAssembly/bulk-memory.ll
+++ b/llvm/test/CodeGen/WebAssembly/bulk-memory.ll
@@ -104,6 +104,31 @@ define void @memset_i32(ptr %dest, i8 %val, i32 %len) {
   ret void
 }
 
+; CHECK-LABEL: memcpy_0:
+; CHECK-NEXT: .functype memcpy_0 (i32, i32) -> ()
+; CHECK-NEXT: return
+define void @memcpy_0(ptr %dest, ptr %src) {
+  call void @llvm.memcpy.p0.p0.i32(ptr %dest, ptr %src, i32 0, i1 0)
+  ret void
+}
+
+; CHECK-LABEL: memmove_0:
+; CHECK-NEXT: .functype memmove_0 (i32, i32) -> ()
+; CHECK-NEXT: return
+define void @memmove_0(ptr %dest, ptr %src) {
+  call void @llvm.memmove.p0.p0.i32(ptr %dest, ptr %src, i32 0, i1 0)
+  ret void
+}
+
+; CHECK-LABEL: memset_0:
+; NO-BULK-MEM-NOT: memory.fill
+; BULK-MEM-NEXT: .functype memset_0 (i32, i32) -> ()
+; BULK-MEM-NEXT: return
+define void @memset_0(ptr %dest, i8 %val) {
+  call void @llvm.memset.p0.i32(ptr %dest, i8 %val, i32 0, i1 0)
+  ret void
+}
+
 ; CHECK-LABEL: memcpy_1:
 ; CHECK-NEXT: .functype memcpy_1 (i32, i32) -> ()
 ; CHECK-NEXT: i32.load8_u $push[[L0:[0-9]+]]=, 0($1)
@@ -137,14 +162,8 @@ define void @memset_1(ptr %dest, i8 %val) {
 ; CHECK-LABEL: memcpy_1024:
 ; NO-BULK-MEM-NOT: memory.copy
 ; BULK-MEM-NEXT: .functype memcpy_1024 (i32, i32) -> ()
-; BULK-MEM-NEXT: block
 ; BULK-MEM-NEXT: i32.const $push[[L0:[0-9]+]]=, 1024
-; BULK-MEM-NEXT: i32.eqz $push[[L1:[0-9]+]]=, $pop[[L0]]
-; BULK-MEM-NEXT: br_if 0, $pop[[L1]]
-; BULK-MEM-NEXT: i32.const $push[[L2:[0-9]+]]=, 1024
-; BULK-MEM-NEXT: memory.copy 0, 0, $0, $1, $pop[[L2]]
-; BULK-MEM-NEXT: .LBB{{.*}}:
-; BULK-MEM-NEXT: end_block
+; BULK-MEM-NEXT: memory.copy 0, 0, $0, $1, $pop[[L0]]
 ; BULK-MEM-NEXT: return
 define void @memcpy_1024(ptr %dest, ptr %src) {
   call void @llvm.memcpy.p0.p0.i32(ptr %dest, ptr %src, i32 1024, i1 0)
@@ -154,14 +173,8 @@ define void @memcpy_1024(ptr %dest, ptr %src) {
 ; CHECK-LABEL: memmove_1024:
 ; NO-BULK-MEM-NOT: memory.copy
 ; BULK-MEM-NEXT: .functype memmove_1024 (i32, i32) -> ()
-; BULK-MEM-NEXT: block
 ; BULK-MEM-NEXT: i32.const $push[[L0:[0-9]+]]=, 1024
-; BULK-MEM-NEXT: i32.eqz $push[[L1:[0-9]+]]=, $pop[[L0]]
-; BULK-MEM-NEXT: br_if 0, $pop[[L1]]
-; BULK-MEM-NEXT: i32.const $push[[L2:[0-9]+]]=, 1024
-; BULK-MEM-NEXT: memory.copy 0, 0, $0, $1, $pop[[L2]]
-; BULK-MEM-NEXT: .LBB{{.*}}:
-; BULK-MEM-NEXT: end_block
+; BULK-MEM-NEXT: memory.copy 0, 0, $0, $1, $pop[[L0]]
 ; BULK-MEM-NEXT: return
 define void @memmove_1024(ptr %dest, ptr %src) {
   call void @llvm.memmove.p0.p0.i32(ptr %dest, ptr %src, i32 1024, i1 0)
@@ -171,14 +184,8 @@ define void @memmove_1024(ptr %dest, ptr %src) {
 ; CHECK-LABEL: memset_1024:
 ; NO-BULK-MEM-NOT: memory.fill
 ; BULK-MEM-NEXT: .functype memset_1024 (i32, i32) -> ()
-; BULK-MEM-NEXT: block
 ; BULK-MEM-NEXT: i32.const $push[[L0:[0-9]+]]=, 1024
-; BULK-MEM-NEXT: i32.eqz $push[[L1:[0-9]+]]=, $pop[[L0]]
-; BULK-MEM-NEXT: br_if 0, $pop[[L1]]
-; BULK-MEM-NEXT: i32.const $push[[L2:[0-9]+]]=, 1024
-; BULK-MEM-NEXT: memory.fill 0, $0, $1, $pop[[L2]]
-; BULK-MEM-NEXT: .LBB{{.*}}:
-; BULK-MEM-NEXT: end_block
+; BULK-MEM-NEXT: memory.fill 0, $0, $1, $pop[[L0]]
 ; BULK-MEM-NEXT: return
 define void @memset_1024(ptr %dest, i8 %val) {
   call void @llvm.memset.p0.i32(ptr %dest, i8 %val, i32 1024, i1 0)
@@ -201,17 +208,11 @@ define void @memset_1024(ptr %dest, i8 %val) {
 ; BULK-MEM-NEXT: .functype memcpy_alloca_src (i32) -> ()
 ; BULK-MEM-NEXT: global.get $push[[L0:[0-9]+]]=, __stack_pointer
 ; BULK-MEM-NEXT: i32.const $push[[L1:[0-9]+]]=, 112
-; BULK-MEM-NEXT: i32.sub $[[L2:[0-9]+]]=, $pop[[L0]], $pop[[L1]]
-; BULK-MEM-NEXT: block
-; BULK-MEM-NEXT: i32.const $push[[L3:[0-9]+]]=, 100
-; BULK-MEM-NEXT: i32.eqz $push[[L4:[0-9]+]]=, $pop[[L3]]
-; BULK-MEM-NEXT: br_if 0, $pop[[L4]]
-; BULK-MEM-NEXT: i32.const $push[[L5:[0-9]+]]=, 12
-; BULK-MEM-NEXT: i32.add $push[[L6:[0-9]+]]=, $[[L2]], $pop[[L5]]
-; BULK-MEM-NEXT: i32.const $push[[L7:[0-9]+]]=, 100
-; BULK-MEM-NEXT: memory.copy 0, 0, $0, $pop[[L6]], $pop[[L7]]
-; BULK-MEM-NEXT: .LBB{{.*}}:
-; BULK-MEM-NEXT: end_block
+; BULK-MEM-NEXT: i32.sub $push[[L2:[0-9]+]]=, $pop[[L0]], $pop[[L1]]
+; BULK-MEM-NEXT: i32.const $push[[L3:[0-9]+]]=, 12
+; BULK-MEM-NEXT: i32.add $push[[L4:[0-9]+]]=, $pop[[L2]], $pop[[L3]]
+; BULK-MEM-NEXT: i32.const $push[[L5:[0-9]+]]=, 100
+; BULK-MEM-NEXT: memory.copy 0, 0, $0, $pop[[L4]], $pop[[L5]]
 ; BULK-MEM-NEXT: return
 define void @memcpy_alloca_src(ptr %dst) {
   %a = alloca [100 x i8]
@@ -224,17 +225,11 @@ define void @memcpy_alloca_src(ptr %dst) {
 ; BULK-MEM-NEXT: .functype memcpy_alloca_dst (i32) -> ()
 ; BULK-MEM-NEXT: global.get $push[[L0:[0-9]+]]=, __stack_pointer
 ; BULK-MEM-NEXT: i32.const $push[[L1:[0-9]+]]=, 112
-; BULK-MEM-NEXT: i32.sub $[[L2:[0-9]+]]=, $pop[[L0]], $pop[[L1]]
-; BULK-MEM-NEXT: block
-; BULK-MEM-NEXT: i32.const $push[[L3:[0-9]+]]=, 100
-; BULK-MEM-NEXT: i32.eqz $push[[L4:[0-9]+]]=, $pop[[L3]]
-; BULK-MEM-NEXT: br_if 0, $pop[[L4]]
-; BULK-MEM-NEXT: i32.const $push[[L5:[0-9]+]]=, 12
-; BULK-MEM-NEXT: i32.add $push[[L6:[0-9]+]]=, $[[L2]], $pop[[L5]]
-; BULK-MEM-NEXT: i32.const $push[[L7:[0-9]+]]=, 100
-; BULK-MEM-NEXT: memory.copy 0, 0, $pop[[L6]], $0, $pop[[L7]]
-; BULK-MEM-NEXT: .LBB{{.*}}:
-; BULK-MEM-NEXT: end_block
+; BULK-MEM-NEXT: i32.sub $push[[L2:[0-9]+]]=, $pop[[L0]], $pop[[L1]]
+; BULK-MEM-NEXT: i32.const $push[[L4:[0-9]+]]=, 12
+; BULK-MEM-NEXT: i32.add $push[[L5:[0-9]+]]=, $pop[[L2]], $pop[[L4]]
+; BULK-MEM-NEXT: i32.const $push[[L6:[0-9]+]]=, 100
+; BULK-MEM-NEXT: memory.copy 0, 0, $pop[[L5]], $0, $pop[[L6]]
 ; BULK-MEM-NEXT: return
 define void @memcpy_alloca_dst(ptr %src) {
   %a = alloca [100 x i8]
@@ -247,17 +242,11 @@ define void @memcpy_alloca_dst(ptr %src) {
 ; BULK-MEM-NEXT: .functype memset_alloca (i32) -> ()
 ; BULK-MEM-NEXT: global.get $push[[L0:[0-9]+]]=, __stack_pointer
 ; BULK-MEM-NEXT: i32.const $push[[L1:[0-9]+]]=, 112
-; BULK-MEM-NEXT: i32.sub $1=, $pop[[L0]], $pop[[L1]]
-; BULK-MEM-NEXT: block
-; BULK-MEM-NEXT: i32.const $push[[L2:[0-9]+]]=, 100
-; BULK-MEM-NEXT: i32.eqz $push[[L3:[0-9]+]]=, $pop[[L2]]
-; BULK-MEM-NEXT: br_if 0, $pop[[L3]]
-; BULK-MEM-NEXT: i32.const $push[[L4:[0-9]+]]=, 12
-; BULK-MEM-NEXT: i32.add $push[[L5:[0-9]+]]=, $1, $pop[[L4]]
-; BULK-MEM-NEXT: i32.const $push[[L6:[0-9]+]]=, 100
-; BULK-MEM-NEXT: memory.fill 0, $pop[[L5]], $0, $pop[[L6]]
-; BULK-MEM-NEXT: .LBB{{.*}}:
-; BULK-MEM-NEXT: end_block
+; BULK-MEM-NEXT: i32.sub $push[[L2:[0-9]+]]=, $pop[[L0]], $pop[[L1]]
+; BULK-MEM-NEXT: i32.const $push[[L3:[0-9]+]]=, 12
+; BULK-MEM-NEXT: i32.add $push[[L4:[0-9]+]]=, $pop[[L2]], $pop[[L3]]
+; BULK-MEM-NEXT: i32.const $push[[L5:[0-9]+]]=, 100
+; BULK-MEM-NEXT: memory.fill 0, $pop[[L4]], $0, $pop[[L5]]
 ; BULK-MEM-NEXT: return
 define void @memset_alloca(i8 %val) {
   %a = alloca [100 x i8]
diff --git a/llvm/test/CodeGen/WebAssembly/bulk-memory64.ll b/llvm/test/CodeGen/WebAssembly/bulk-memory64.ll
index 0cf8493a995f9..d0206a399a9b2 100644
--- a/llvm/test/CodeGen/WebAssembly/bulk-memory64.ll
+++ b/llvm/test/CodeGen/WebAssembly/bulk-memory64.ll
@@ -110,6 +110,31 @@ define void @memset_i32(ptr %dest, i8 %val, i64 %len) {
   ret void
 }
 
+; CHECK-LABEL: memcpy_0:
+; CHECK-NEXT: .functype memcpy_0 (i64, i64) -> ()
+; CHECK-NEXT: return
+define void @memcpy_0(ptr %dest, ptr %src) {
+  call void @llvm.memcpy.p0.p0.i64(ptr %dest, ptr %src, i64 0, i1 0)
+  ret void
+}
+
+; CHECK-LABEL: memmove_0:
+; CHECK-NEXT: .functype memmove_0 (i64, i64) -> ()
+; CHECK-NEXT: return
+define void @memmove_0(ptr %dest, ptr %src) {
+  call void @llvm.memmove.p0.p0.i64(ptr %dest, ptr %src, i64 0, i1 0)
+  ret void
+}
+
+; CHECK-LABEL: memset_0:
+; NO-BULK-MEM-NOT: memory.fill
+; BULK-MEM-NEXT: .functype memset_0 (i64, i32) -> ()
+; BULK-MEM-NEXT: return
+define void @memset_0(ptr %dest, i8 %val) {
+  call void @llvm.memset.p0.i64(ptr %dest, i8 %val, i64 0, i1 0)
+  ret void
+}
+
 ; CHECK-LABEL: memcpy_1:
 ; CHECK-NEXT: .functype memcpy_1 (i64, i64) -> ()
 ; CHECK-NEXT: i32.load8_u $push[[L0:[0-9]+]]=, 0($1)
@@ -143,14 +168,8 @@ define void @memset_1(ptr %dest, i8 %val) {
 ; CHECK-LABEL: memcpy_1024:
 ; NO-BULK-MEM-NOT: memory.copy
 ; BULK-MEM-NEXT: .functype memcpy_1024 (i64, i64) -> ()
-; BULK-MEM-NEXT: block
-; BULK-MEM-NEXT: i64.const	$push[[L1:[0-9]+]]=, 1024
-; BULK-MEM-NEXT: i64.eqz 	$push0=, $pop[[L1]]
-; BULK-MEM-NEXT: br_if   	0, $pop0
 ; BULK-MEM-NEXT: i64.const	$push[[L0:[0-9]+]]=, 1024
 ; BULK-MEM-NEXT: memory.copy	0, 0, $0, $1, $pop[[L0]]
-; BULK-MEM-NEXT: .LBB{{.*}}:
-; BULK-MEM-NEXT: end_block
 ; BULK-MEM-NEXT: return
 define void @memcpy_1024(ptr %dest, ptr %src) {
   call void @llvm.memcpy.p0.p0.i64(ptr %dest, ptr %src, i64 1024, i1 0)
@@ -160,14 +179,8 @@ define void @memcpy_1024(ptr %dest, ptr %src) {
 ; CHECK-LABEL: memmove_1024:
 ; NO-BULK-MEM-NOT: memory.copy
 ; BULK-MEM-NEXT: .functype memmove_1024 (i64, i64) -> ()
-; BULK-MEM-NEXT: block
-; BULK-MEM-NEXT: i64.const	$push[[L1:[0-9]+]]=, 1024
-; BULK-MEM-NEXT: i64.eqz 	$push0=, $pop[[L1]]
-; BULK-MEM-NEXT: br_if   	0, $pop0
 ; BULK-MEM-NEXT: i64.const	$push[[L0:[0-9]+]]=, 1024
 ; BULK-MEM-NEXT: memory.copy	0, 0, $0, $1, $pop[[L0]]
-; BULK-MEM-NEXT: .LBB{{.*}}:
-; BULK-MEM-NEXT: end_block
 ; BULK-MEM-NEXT: return
 define void @memmove_1024(ptr %dest, ptr %src) {
   call void @llvm.memmove.p0.p0.i64(ptr %dest, ptr %src, i64 1024, i1 0)
@@ -177,14 +190,8 @@ define void @memmove_1024(ptr %dest, ptr %src) {
 ; CHECK-LABEL: memset_1024:
 ; NO-BULK-MEM-NOT: memory.fill
 ; BULK-MEM-NEXT: .functype memset_1024 (i64, i32) -> ()
-; BULK-MEM-NEXT: block
-; BULK-MEM-NEXT: i64.const	$push[[L1:[0-9]+]]=, 1024
-; BULK-MEM-NEXT: i64.eqz 	$push0=, $pop[[L1]]
-; BULK-MEM-NEXT: br_if   	0, $pop0
 ; BULK-MEM-NEXT: i64.const	$push[[L0:[0-9]+]]=, 1024
 ; BULK-MEM-NEXT: memory.fill	0, $0, $1, $pop[[L0]]
-; BULK-MEM-NEXT: .LBB{{.*}}:
-; BULK-MEM-NEXT: end_block
 ; BULK-MEM-NEXT: return
 define void @memset_1024(ptr %dest, i8 %val) {
   call void @llvm.memset.p0.i64(ptr %dest, i8 %val, i64 1024, i1 0)
@@ -207,17 +214,11 @@ define void @memset_1024(ptr %dest, i8 %val) {
 ; BULK-MEM-NEXT: .functype memcpy_alloca_src (i64) -> ()
 ; BULK-MEM-NEXT: global.get	$push[[L1:[0-9]+]]=, __stack_pointer
 ; BULK-MEM-NEXT: i64.const	$push[[L0:[0-9]+]]=, 112
-; BULK-MEM-NEXT: i64.sub 	$[[L2:[0-9]+]]=, $pop[[L1]], $pop[[L0]]
-; BULK-MEM-NEXT: block
-; BULK-MEM-NEXT: i64.const	$push[[L3:[0-9]+]]=, 100
-; BULK-MEM-NEXT: i64.eqz 	$push[[L4:[0-9]+]]=, $pop[[L3]]
-; BULK-MEM-NEXT: br_if   	0, $pop[[L4]]
-; BULK-MEM-NEXT: i64.const	$push[[L5:[0-9]+]]=, 12
-; BULK-MEM-NEXT: i64.add 	$push[[L6:[0-9]+]]=, $[[L2]], $pop[[L5]]
-; BULK-MEM-NEXT: i64.const	$push[[L7:[0-9]+]]=, 100
-; BULK-MEM-NEXT: memory.copy	0, 0, $0, $pop[[L6]], $pop[[L7]]
-; BULK-MEM-NEXT: .LBB{{.*}}:
-; BULK-MEM-NEXT: end_block
+; BULK-MEM-NEXT: i64.sub 	$push[[L2:[0-9]+]]=, $pop[[L1]], $pop[[L0]]
+; BULK-MEM-NEXT: i64.const	$push[[L3:[0-9]+]]=, 12
+; BULK-MEM-NEXT: i64.add 	$push[[L4:[0-9]+]]=, $pop[[L2]], $pop[[L3]]
+; BULK-MEM-NEXT: i64.const	$push[[L5:[0-9]+]]=, 100
+; BULK-MEM-NEXT: memory.copy	0, 0, $0, $pop[[L4]], $pop[[L5]]
 ; BULK-MEM-NEXT: return
 define void @memcpy_alloca_src(ptr %dst) {
   %a = alloca [100 x i8]
@@ -230,17 +231,11 @@ define void @memcpy_alloca_src(ptr %dst) {
 ; BULK-MEM-NEXT: .functype memcpy_alloca_dst (i64) -> ()
 ; BULK-MEM-NEXT: global.get	$push[[L1:[0-9]+]]=, __stack_pointer
 ; BULK-MEM-NEXT: i64.const	$push[[L0:[0-9]+]]=, 112
-; BULK-MEM-NEXT: i64.sub 	$[[L2:[0-9]+]]=, $pop[[L1]], $pop[[L0]]
-; BULK-MEM-NEXT: block
-; BULK-MEM-NEXT: i64.const	$push[[L3:[0-9]+]]=, 100
-; BULK-MEM-NEXT: i64.eqz 	$push[[L4:[0-9]+]]=, $pop[[L3]]
-; BULK-MEM-NEXT: br_if   	0, $pop[[L4]]
-; BULK-MEM-NEXT: i64.const	$push[[L5:[0-9]+]]=, 12
-; BULK-MEM-NEXT: i64.add 	$push[[L6:[0-9]+]]=, $[[L2]], $pop[[L5]]
-; BULK-MEM-NEXT: i64.const	$push[[L7:[0-9]+]]=, 100
-; BULK-MEM-NEXT: memory.copy	0, 0, $pop[[L6]], $0, $pop[[L7]]
-; BULK-MEM-NEXT: .LBB{{.*}}:
-; BULK-MEM-NEXT: end_block
+; BULK-MEM-NEXT: i64.sub 	$push[[L2:[0-9]+]]=, $pop[[L1]], $pop[[L0]]
+; BULK-MEM-NEXT: i64.const	$push[[L3:[0-9]+]]=, 12
+; BULK-MEM-NEXT: i64.add 	$push[[L4:[0-9]+]]=, $pop[[L2]], $pop[[L3]]
+; BULK-MEM-NEXT: i64.const	$push[[L5:[0-9]+]]=, 100
+; BULK-MEM-NEXT: memory.copy	0, 0, $pop[[L4]], $0, $pop[[L5]]
 ; BULK-MEM-NEXT: return
 define void @memcpy_alloca_dst(ptr %src) {
   %a = alloca [100 x i8]
@@ -253,17 +248,11 @@ define void @memcpy_alloca_dst(ptr %src) {
 ; BULK-MEM-NEXT: .functype memset_alloca (i32) -> ()
 ; BULK-MEM-NEXT: global.get	$push[[L1:[0-9]+]]=, __stack_pointer
 ; BULK-MEM-NEXT: i64.const	$push[[L0:[0-9]+]]=, 112
-; BULK-MEM-NEXT: i64.sub 	$1=, $pop[[L1]], $pop[[L0]]
-; BULK-MEM-NEXT: block
-; BULK-MEM-NEXT: i64.const	$push[[L2:[0-9]+]]=, 100
-; BULK-MEM-NEXT: i64.eqz 	$push[[L3:[0-9]+]]=, $pop[[L2]]
-; BULK-MEM-NEXT: br_if   	0, $pop[[L3]]
-; BULK-MEM-NEXT: i64.const	$push[[L4:[0-9]+]]=, 12
-; BULK-MEM-NEXT: i64.add 	$push[[L5:[0-9]+]]=, $1, $pop[[L4]]
-; BULK-MEM-NEXT: i64.const	$push[[L6:[0-9]+]]=, 100
-; BULK-MEM-NEXT: memory.fill	0, $pop[[L5]], $0, $pop[[L6]]
-; BULK-MEM-NEXT: .LBB{{.*}}:
-; BULK-MEM-NEXT: end_block
+; BULK-MEM-NEXT: i64.sub 	$push[[L2:[0-9]+]]=, $pop[[L1]], $pop[[L0]]
+; BULK-MEM-NEXT: i64.const	$push[[L3:[0-9]+]]=, 12
+; BULK-MEM-NEXT: i64.add 	$push[[L4:[0-9]+]]=, $pop[[L2]], $pop[[L3]]
+; BULK-MEM-NEXT: i64.const	$push[[L5:[0-9]+]]=, 100
+; BULK-MEM-NEXT: memory.fill	0, $pop[[L4]], $0, $pop[[L5]]
 ; BULK-MEM-NEXT: return
 define void @memset_alloca(i8 %val) {
   %a = alloca [100 x i8]

aheejin
aheejin approved these changes Oct 14, 2025
View reviewed changes
llvm/test/CodeGen/WebAssembly/bulk-memory.ll
Comment on lines +229 to +232
; BULK-MEM-NEXT: i32.const $push[[L4:[0-9]+]]=, 12
; BULK-MEM-NEXT: i32.add $push[[L5:[0-9]+]]=, $pop[[L2]], $pop[[L4]]
; BULK-MEM-NEXT: i32.const $push[[L6:[0-9]+]]=, 100
; BULK-MEM-NEXT: memory.copy 0, 0, $pop[[L5]], $0, $pop[[L6]]
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No L3?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@aheejin aheejin aheejin approved these changes

@sunfishcode sunfishcode Awaiting requested review from sunfishcode

Assignees

No one assigned

Labels

backend:WebAssembly

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

[WebAseembly] Zero-length checks for constant-sized bulk memory ops

3 participants

@dschuff @llvmbot @aheejin