Skip to content

[DAG] ExpandOp_NormalStore - check for bitcasted value that has legal store instead of splitting #171478

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
Closed
RKSimon wants to merge 1 commit into from
Closed

[DAG] ExpandOp_NormalStore - check for bitcasted value that has legal store instead of splitting #171478

RKSimon wants to merge 1 commit into from
+71 −143

Conversation

@RKSimon
Copy link
Collaborator

@RKSimon RKSimon commented Dec 9, 2025

DAGCombine does attempt this, but we can end up in situations where the bitcast appears after the store has been folded and we don't try again - someday we'll have better topological sorting :(

Noticed while working on some i256/i512 codegen patches that gets cast back from 256/512-bit vectors.

@RKSimon
[DAG] ExpandOp_NormalStore - check for bitcasted value that has legal…
8983aae 
… store instead of splitting

DAGCombine does attempt this, but we can end up in situations where the bitcast appears after the store has been folded and we don't try again.

Noticed while working on some i256/i512 codegen patches that gets cast back from 256/512-bit vectors.
@llvmbot llvmbot added backend:X86 llvm:SelectionDAG SelectionDAGISel as well labels Dec 9, 2025
@llvmbot
Copy link
Member

llvmbot commented Dec 9, 2025

@llvm/pr-subscribers-backend-x86

Author: Simon Pilgrim (RKSimon)

Changes

DAGCombine does attempt this, but we can end up in situations where the bitcast appears after the store has been folded and we don't try again - someday we'll have better topological sorting :(

Noticed while working on some i256/i512 codegen patches that gets cast back from 256/512-bit vectors.

Full diff: https://github.com/llvm/llvm-project/pull/171478.diff

3 Files Affected:

  • (modified) llvm/lib/CodeGen/SelectionDAG/LegalizeTypesGeneric.cpp (+10-2)
  • (modified) llvm/test/CodeGen/X86/atomic-fp.ll (+60-140)
  • (modified) llvm/test/CodeGen/X86/single_elt_vector_memory_operation.ll (+1-1)
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeTypesGeneric.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeTypesGeneric.cpp
index 88c1af20a321e..4348f0d6f0aa7 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeTypesGeneric.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeTypesGeneric.cpp
@@ -479,15 +479,23 @@ SDValue DAGTypeLegalizer::ExpandOp_NormalStore(SDNode *N, unsigned OpNo) {
 
   StoreSDNode *St = cast<StoreSDNode>(N);
   assert(!St->isAtomic() && "Atomics can not be split");
-  EVT ValueVT = St->getValue().getValueType();
-  EVT NVT = TLI.getTypeToTransformTo(*DAG.getContext(), ValueVT);
   SDValue Chain = St->getChain();
+  SDValue Value = St->getValue();
   SDValue Ptr = St->getBasePtr();
+  EVT ValueVT = Value.getValueType();
+  EVT NVT = TLI.getTypeToTransformTo(*DAG.getContext(), ValueVT);
   AAMDNodes AAInfo = St->getAAInfo();
 
   assert(NVT.isByteSized() && "Expanded type not byte sized!");
   unsigned IncrementSize = NVT.getSizeInBits() / 8;
 
+  // Storing a bitcasted value, see if the original type is a legal store.
+  // TODO: Not necessary if we had proper topological sorting of nodes.
+  if (Value.getOpcode() == ISD::BITCAST &&
+      TLI.isOperationLegal(ISD::STORE, Value.getOperand(0).getValueType()))
+    return DAG.getStore(Chain, dl, Value.getOperand(0), Ptr,
+                        St->getMemOperand());
+
   SDValue Lo, Hi;
   GetExpandedOp(St->getValue(), Lo, Hi);
 
diff --git a/llvm/test/CodeGen/X86/atomic-fp.ll b/llvm/test/CodeGen/X86/atomic-fp.ll
index fe79dfe39f645..be67c19dfe111 100644
--- a/llvm/test/CodeGen/X86/atomic-fp.ll
+++ b/llvm/test/CodeGen/X86/atomic-fp.ll
@@ -87,15 +87,11 @@ define dso_local void @fadd_64r(ptr %loc, double %val) nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-NOSSE-NEXT:    movl %edx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fldl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %ecx, (%esp)
+; X86-NOSSE-NEXT:    fldl (%esp)
 ; X86-NOSSE-NEXT:    faddl 12(%ebp)
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NOSSE-NEXT:    movl %ecx, (%esp)
-; X86-NOSSE-NEXT:    movl %edx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll (%eax)
 ; X86-NOSSE-NEXT:    movl %ebp, %esp
 ; X86-NOSSE-NEXT:    popl %ebp
@@ -252,15 +248,11 @@ define dso_local void @fadd_64g() nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %eax, {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %eax, (%esp)
 ; X86-NOSSE-NEXT:    fld1
-; X86-NOSSE-NEXT:    faddl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    faddl (%esp)
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl %eax, (%esp)
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll glob64
 ; X86-NOSSE-NEXT:    movl %ebp, %esp
 ; X86-NOSSE-NEXT:    popl %ebp
@@ -415,15 +407,11 @@ define dso_local void @fadd_64imm() nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %eax, {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %eax, (%esp)
 ; X86-NOSSE-NEXT:    fld1
-; X86-NOSSE-NEXT:    faddl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    faddl (%esp)
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl %eax, (%esp)
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll -559038737
 ; X86-NOSSE-NEXT:    movl %ebp, %esp
 ; X86-NOSSE-NEXT:    popl %ebp
@@ -583,15 +571,11 @@ define dso_local void @fadd_64stack() nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %eax, {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %eax, (%esp)
 ; X86-NOSSE-NEXT:    fld1
-; X86-NOSSE-NEXT:    faddl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    faddl (%esp)
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl %eax, (%esp)
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    movl %ebp, %esp
 ; X86-NOSSE-NEXT:    popl %ebp
@@ -685,15 +669,11 @@ define dso_local void @fadd_array(ptr %arg, double %arg1, i64 %arg2) nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X86-NOSSE-NEXT:    movl %esi, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %edx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fldl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %edx, (%esp)
+; X86-NOSSE-NEXT:    fldl (%esp)
 ; X86-NOSSE-NEXT:    faddl 12(%ebp)
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-NOSSE-NEXT:    movl %edx, (%esp)
-; X86-NOSSE-NEXT:    movl %esi, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll (%ecx,%eax,8)
 ; X86-NOSSE-NEXT:    leal -4(%ebp), %esp
 ; X86-NOSSE-NEXT:    popl %esi
@@ -859,15 +839,11 @@ define dso_local void @fsub_64r(ptr %loc, double %val) nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-NOSSE-NEXT:    movl %edx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fldl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %ecx, (%esp)
+; X86-NOSSE-NEXT:    fldl (%esp)
 ; X86-NOSSE-NEXT:    fsubl 12(%ebp)
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NOSSE-NEXT:    movl %ecx, (%esp)
-; X86-NOSSE-NEXT:    movl %edx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll (%eax)
 ; X86-NOSSE-NEXT:    movl %ebp, %esp
 ; X86-NOSSE-NEXT:    popl %ebp
@@ -1024,16 +1000,12 @@ define dso_local void @fsub_64g() nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %eax, {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %eax, (%esp)
 ; X86-NOSSE-NEXT:    fld1
 ; X86-NOSSE-NEXT:    fchs
-; X86-NOSSE-NEXT:    faddl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    faddl (%esp)
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl %eax, (%esp)
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll glob64
 ; X86-NOSSE-NEXT:    movl %ebp, %esp
 ; X86-NOSSE-NEXT:    popl %ebp
@@ -1190,16 +1162,12 @@ define dso_local void @fsub_64imm() nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %eax, {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %eax, (%esp)
 ; X86-NOSSE-NEXT:    fld1
 ; X86-NOSSE-NEXT:    fchs
-; X86-NOSSE-NEXT:    faddl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    faddl (%esp)
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl %eax, (%esp)
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll -559038737
 ; X86-NOSSE-NEXT:    movl %ebp, %esp
 ; X86-NOSSE-NEXT:    popl %ebp
@@ -1360,15 +1328,11 @@ define dso_local void @fsub_64stack() nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %eax, {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %eax, (%esp)
 ; X86-NOSSE-NEXT:    fld1
-; X86-NOSSE-NEXT:    fsubl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    fsubl (%esp)
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl %eax, (%esp)
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    movl %ebp, %esp
 ; X86-NOSSE-NEXT:    popl %ebp
@@ -1464,15 +1428,11 @@ define dso_local void @fsub_array(ptr %arg, double %arg1, i64 %arg2) nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X86-NOSSE-NEXT:    movl %esi, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %edx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fldl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %edx, (%esp)
+; X86-NOSSE-NEXT:    fldl (%esp)
 ; X86-NOSSE-NEXT:    fsubl 12(%ebp)
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-NOSSE-NEXT:    movl %edx, (%esp)
-; X86-NOSSE-NEXT:    movl %esi, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll (%ecx,%eax,8)
 ; X86-NOSSE-NEXT:    leal -4(%ebp), %esp
 ; X86-NOSSE-NEXT:    popl %esi
@@ -1638,15 +1598,11 @@ define dso_local void @fmul_64r(ptr %loc, double %val) nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-NOSSE-NEXT:    movl %edx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fldl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %ecx, (%esp)
+; X86-NOSSE-NEXT:    fldl (%esp)
 ; X86-NOSSE-NEXT:    fmull 12(%ebp)
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NOSSE-NEXT:    movl %ecx, (%esp)
-; X86-NOSSE-NEXT:    movl %edx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll (%eax)
 ; X86-NOSSE-NEXT:    movl %ebp, %esp
 ; X86-NOSSE-NEXT:    popl %ebp
@@ -1800,15 +1756,11 @@ define dso_local void @fmul_64g() nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %eax, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fldl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %eax, (%esp)
+; X86-NOSSE-NEXT:    fldl (%esp)
 ; X86-NOSSE-NEXT:    fmuls {{\.?LCPI[0-9]+_[0-9]+}}
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl %eax, (%esp)
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll glob64
 ; X86-NOSSE-NEXT:    movl %ebp, %esp
 ; X86-NOSSE-NEXT:    popl %ebp
@@ -1963,15 +1915,11 @@ define dso_local void @fmul_64imm() nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %eax, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fldl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %eax, (%esp)
+; X86-NOSSE-NEXT:    fldl (%esp)
 ; X86-NOSSE-NEXT:    fmuls {{\.?LCPI[0-9]+_[0-9]+}}
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl %eax, (%esp)
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll -559038737
 ; X86-NOSSE-NEXT:    movl %ebp, %esp
 ; X86-NOSSE-NEXT:    popl %ebp
@@ -2131,15 +2079,11 @@ define dso_local void @fmul_64stack() nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %eax, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fldl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %eax, (%esp)
+; X86-NOSSE-NEXT:    fldl (%esp)
 ; X86-NOSSE-NEXT:    fmuls {{\.?LCPI[0-9]+_[0-9]+}}
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl %eax, (%esp)
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    movl %ebp, %esp
 ; X86-NOSSE-NEXT:    popl %ebp
@@ -2233,15 +2177,11 @@ define dso_local void @fmul_array(ptr %arg, double %arg1, i64 %arg2) nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X86-NOSSE-NEXT:    movl %esi, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %edx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fldl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %edx, (%esp)
+; X86-NOSSE-NEXT:    fldl (%esp)
 ; X86-NOSSE-NEXT:    fmull 12(%ebp)
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-NOSSE-NEXT:    movl %edx, (%esp)
-; X86-NOSSE-NEXT:    movl %esi, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll (%ecx,%eax,8)
 ; X86-NOSSE-NEXT:    leal -4(%ebp), %esp
 ; X86-NOSSE-NEXT:    popl %esi
@@ -2407,15 +2347,11 @@ define dso_local void @fdiv_64r(ptr %loc, double %val) nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-NOSSE-NEXT:    movl %edx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fldl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %ecx, (%esp)
+; X86-NOSSE-NEXT:    fldl (%esp)
 ; X86-NOSSE-NEXT:    fdivl 12(%ebp)
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NOSSE-NEXT:    movl %ecx, (%esp)
-; X86-NOSSE-NEXT:    movl %edx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll (%eax)
 ; X86-NOSSE-NEXT:    movl %ebp, %esp
 ; X86-NOSSE-NEXT:    popl %ebp
@@ -2571,15 +2507,11 @@ define dso_local void @fdiv_64g() nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %eax, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fldl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %eax, (%esp)
+; X86-NOSSE-NEXT:    fldl (%esp)
 ; X86-NOSSE-NEXT:    fdivs {{\.?LCPI[0-9]+_[0-9]+}}
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl %eax, (%esp)
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll glob64
 ; X86-NOSSE-NEXT:    movl %ebp, %esp
 ; X86-NOSSE-NEXT:    popl %ebp
@@ -2734,15 +2666,11 @@ define dso_local void @fdiv_64imm() nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %eax, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fldl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %eax, (%esp)
+; X86-NOSSE-NEXT:    fldl (%esp)
 ; X86-NOSSE-NEXT:    fdivs {{\.?LCPI[0-9]+_[0-9]+}}
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl %eax, (%esp)
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll -559038737
 ; X86-NOSSE-NEXT:    movl %ebp, %esp
 ; X86-NOSSE-NEXT:    popl %ebp
@@ -2902,15 +2830,11 @@ define dso_local void @fdiv_64stack() nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %eax, {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %eax, (%esp)
 ; X86-NOSSE-NEXT:    fld1
-; X86-NOSSE-NEXT:    fdivl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    fdivl (%esp)
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl %eax, (%esp)
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    movl %ebp, %esp
 ; X86-NOSSE-NEXT:    popl %ebp
@@ -3006,15 +2930,11 @@ define dso_local void @fdiv_array(ptr %arg, double %arg1, i64 %arg2) nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X86-NOSSE-NEXT:    movl %esi, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %edx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fldl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %edx, (%esp)
+; X86-NOSSE-NEXT:    fldl (%esp)
 ; X86-NOSSE-NEXT:    fdivl 12(%ebp)
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-NOSSE-NEXT:    movl %edx, (%esp)
-; X86-NOSSE-NEXT:    movl %esi, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll (%ecx,%eax,8)
 ; X86-NOSSE-NEXT:    leal -4(%ebp), %esp
 ; X86-NOSSE-NEXT:    popl %esi
diff --git a/llvm/test/CodeGen/X86/single_elt_vector_memory_operation.ll b/llvm/test/CodeGen/X86/single_elt_vector_memory_operation.ll
index f65461ccee23b..4e7a694f31c9c 100644
--- a/llvm/test/CodeGen/X86/single_elt_vector_memory_operation.ll
+++ b/llvm/test/CodeGen/X86/single_elt_vector_memory_operation.ll
@@ -87,8 +87,8 @@ define void @load_single_256bit_elt_vector(ptr %in, ptr %off, ptr %out) nounwind
 ; SSE-NEXT:    xorps %xmm2, %xmm2
 ; SSE-NEXT:    movaps %xmm2, 48(%rdx)
 ; SSE-NEXT:    movaps %xmm2, 32(%rdx)
-; SSE-NEXT:    movaps %xmm0, (%rdx)
 ; SSE-NEXT:    movaps %xmm1, 16(%rdx)
+; SSE-NEXT:    movaps %xmm0, (%rdx)
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: load_single_256bit_elt_vector:

@llvmbot
Copy link
Member

llvmbot commented Dec 9, 2025

@llvm/pr-subscribers-llvm-selectiondag

Author: Simon Pilgrim (RKSimon)

Changes

DAGCombine does attempt this, but we can end up in situations where the bitcast appears after the store has been folded and we don't try again - someday we'll have better topological sorting :(

Noticed while working on some i256/i512 codegen patches that gets cast back from 256/512-bit vectors.

Full diff: https://github.com/llvm/llvm-project/pull/171478.diff

3 Files Affected:

  • (modified) llvm/lib/CodeGen/SelectionDAG/LegalizeTypesGeneric.cpp (+10-2)
  • (modified) llvm/test/CodeGen/X86/atomic-fp.ll (+60-140)
  • (modified) llvm/test/CodeGen/X86/single_elt_vector_memory_operation.ll (+1-1)
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeTypesGeneric.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeTypesGeneric.cpp
index 88c1af20a321e..4348f0d6f0aa7 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeTypesGeneric.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeTypesGeneric.cpp
@@ -479,15 +479,23 @@ SDValue DAGTypeLegalizer::ExpandOp_NormalStore(SDNode *N, unsigned OpNo) {
 
   StoreSDNode *St = cast<StoreSDNode>(N);
   assert(!St->isAtomic() && "Atomics can not be split");
-  EVT ValueVT = St->getValue().getValueType();
-  EVT NVT = TLI.getTypeToTransformTo(*DAG.getContext(), ValueVT);
   SDValue Chain = St->getChain();
+  SDValue Value = St->getValue();
   SDValue Ptr = St->getBasePtr();
+  EVT ValueVT = Value.getValueType();
+  EVT NVT = TLI.getTypeToTransformTo(*DAG.getContext(), ValueVT);
   AAMDNodes AAInfo = St->getAAInfo();
 
   assert(NVT.isByteSized() && "Expanded type not byte sized!");
   unsigned IncrementSize = NVT.getSizeInBits() / 8;
 
+  // Storing a bitcasted value, see if the original type is a legal store.
+  // TODO: Not necessary if we had proper topological sorting of nodes.
+  if (Value.getOpcode() == ISD::BITCAST &&
+      TLI.isOperationLegal(ISD::STORE, Value.getOperand(0).getValueType()))
+    return DAG.getStore(Chain, dl, Value.getOperand(0), Ptr,
+                        St->getMemOperand());
+
   SDValue Lo, Hi;
   GetExpandedOp(St->getValue(), Lo, Hi);
 
diff --git a/llvm/test/CodeGen/X86/atomic-fp.ll b/llvm/test/CodeGen/X86/atomic-fp.ll
index fe79dfe39f645..be67c19dfe111 100644
--- a/llvm/test/CodeGen/X86/atomic-fp.ll
+++ b/llvm/test/CodeGen/X86/atomic-fp.ll
@@ -87,15 +87,11 @@ define dso_local void @fadd_64r(ptr %loc, double %val) nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-NOSSE-NEXT:    movl %edx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fldl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %ecx, (%esp)
+; X86-NOSSE-NEXT:    fldl (%esp)
 ; X86-NOSSE-NEXT:    faddl 12(%ebp)
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NOSSE-NEXT:    movl %ecx, (%esp)
-; X86-NOSSE-NEXT:    movl %edx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll (%eax)
 ; X86-NOSSE-NEXT:    movl %ebp, %esp
 ; X86-NOSSE-NEXT:    popl %ebp
@@ -252,15 +248,11 @@ define dso_local void @fadd_64g() nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %eax, {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %eax, (%esp)
 ; X86-NOSSE-NEXT:    fld1
-; X86-NOSSE-NEXT:    faddl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    faddl (%esp)
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl %eax, (%esp)
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll glob64
 ; X86-NOSSE-NEXT:    movl %ebp, %esp
 ; X86-NOSSE-NEXT:    popl %ebp
@@ -415,15 +407,11 @@ define dso_local void @fadd_64imm() nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %eax, {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %eax, (%esp)
 ; X86-NOSSE-NEXT:    fld1
-; X86-NOSSE-NEXT:    faddl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    faddl (%esp)
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl %eax, (%esp)
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll -559038737
 ; X86-NOSSE-NEXT:    movl %ebp, %esp
 ; X86-NOSSE-NEXT:    popl %ebp
@@ -583,15 +571,11 @@ define dso_local void @fadd_64stack() nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %eax, {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %eax, (%esp)
 ; X86-NOSSE-NEXT:    fld1
-; X86-NOSSE-NEXT:    faddl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    faddl (%esp)
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl %eax, (%esp)
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    movl %ebp, %esp
 ; X86-NOSSE-NEXT:    popl %ebp
@@ -685,15 +669,11 @@ define dso_local void @fadd_array(ptr %arg, double %arg1, i64 %arg2) nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X86-NOSSE-NEXT:    movl %esi, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %edx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fldl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %edx, (%esp)
+; X86-NOSSE-NEXT:    fldl (%esp)
 ; X86-NOSSE-NEXT:    faddl 12(%ebp)
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-NOSSE-NEXT:    movl %edx, (%esp)
-; X86-NOSSE-NEXT:    movl %esi, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll (%ecx,%eax,8)
 ; X86-NOSSE-NEXT:    leal -4(%ebp), %esp
 ; X86-NOSSE-NEXT:    popl %esi
@@ -859,15 +839,11 @@ define dso_local void @fsub_64r(ptr %loc, double %val) nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-NOSSE-NEXT:    movl %edx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fldl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %ecx, (%esp)
+; X86-NOSSE-NEXT:    fldl (%esp)
 ; X86-NOSSE-NEXT:    fsubl 12(%ebp)
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NOSSE-NEXT:    movl %ecx, (%esp)
-; X86-NOSSE-NEXT:    movl %edx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll (%eax)
 ; X86-NOSSE-NEXT:    movl %ebp, %esp
 ; X86-NOSSE-NEXT:    popl %ebp
@@ -1024,16 +1000,12 @@ define dso_local void @fsub_64g() nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %eax, {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %eax, (%esp)
 ; X86-NOSSE-NEXT:    fld1
 ; X86-NOSSE-NEXT:    fchs
-; X86-NOSSE-NEXT:    faddl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    faddl (%esp)
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl %eax, (%esp)
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll glob64
 ; X86-NOSSE-NEXT:    movl %ebp, %esp
 ; X86-NOSSE-NEXT:    popl %ebp
@@ -1190,16 +1162,12 @@ define dso_local void @fsub_64imm() nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %eax, {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %eax, (%esp)
 ; X86-NOSSE-NEXT:    fld1
 ; X86-NOSSE-NEXT:    fchs
-; X86-NOSSE-NEXT:    faddl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    faddl (%esp)
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl %eax, (%esp)
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll -559038737
 ; X86-NOSSE-NEXT:    movl %ebp, %esp
 ; X86-NOSSE-NEXT:    popl %ebp
@@ -1360,15 +1328,11 @@ define dso_local void @fsub_64stack() nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %eax, {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %eax, (%esp)
 ; X86-NOSSE-NEXT:    fld1
-; X86-NOSSE-NEXT:    fsubl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    fsubl (%esp)
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl %eax, (%esp)
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    movl %ebp, %esp
 ; X86-NOSSE-NEXT:    popl %ebp
@@ -1464,15 +1428,11 @@ define dso_local void @fsub_array(ptr %arg, double %arg1, i64 %arg2) nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X86-NOSSE-NEXT:    movl %esi, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %edx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fldl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %edx, (%esp)
+; X86-NOSSE-NEXT:    fldl (%esp)
 ; X86-NOSSE-NEXT:    fsubl 12(%ebp)
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-NOSSE-NEXT:    movl %edx, (%esp)
-; X86-NOSSE-NEXT:    movl %esi, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll (%ecx,%eax,8)
 ; X86-NOSSE-NEXT:    leal -4(%ebp), %esp
 ; X86-NOSSE-NEXT:    popl %esi
@@ -1638,15 +1598,11 @@ define dso_local void @fmul_64r(ptr %loc, double %val) nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-NOSSE-NEXT:    movl %edx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fldl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %ecx, (%esp)
+; X86-NOSSE-NEXT:    fldl (%esp)
 ; X86-NOSSE-NEXT:    fmull 12(%ebp)
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NOSSE-NEXT:    movl %ecx, (%esp)
-; X86-NOSSE-NEXT:    movl %edx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll (%eax)
 ; X86-NOSSE-NEXT:    movl %ebp, %esp
 ; X86-NOSSE-NEXT:    popl %ebp
@@ -1800,15 +1756,11 @@ define dso_local void @fmul_64g() nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %eax, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fldl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %eax, (%esp)
+; X86-NOSSE-NEXT:    fldl (%esp)
 ; X86-NOSSE-NEXT:    fmuls {{\.?LCPI[0-9]+_[0-9]+}}
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl %eax, (%esp)
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll glob64
 ; X86-NOSSE-NEXT:    movl %ebp, %esp
 ; X86-NOSSE-NEXT:    popl %ebp
@@ -1963,15 +1915,11 @@ define dso_local void @fmul_64imm() nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %eax, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fldl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %eax, (%esp)
+; X86-NOSSE-NEXT:    fldl (%esp)
 ; X86-NOSSE-NEXT:    fmuls {{\.?LCPI[0-9]+_[0-9]+}}
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl %eax, (%esp)
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll -559038737
 ; X86-NOSSE-NEXT:    movl %ebp, %esp
 ; X86-NOSSE-NEXT:    popl %ebp
@@ -2131,15 +2079,11 @@ define dso_local void @fmul_64stack() nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %eax, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fldl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %eax, (%esp)
+; X86-NOSSE-NEXT:    fldl (%esp)
 ; X86-NOSSE-NEXT:    fmuls {{\.?LCPI[0-9]+_[0-9]+}}
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl %eax, (%esp)
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    movl %ebp, %esp
 ; X86-NOSSE-NEXT:    popl %ebp
@@ -2233,15 +2177,11 @@ define dso_local void @fmul_array(ptr %arg, double %arg1, i64 %arg2) nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X86-NOSSE-NEXT:    movl %esi, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %edx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fldl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %edx, (%esp)
+; X86-NOSSE-NEXT:    fldl (%esp)
 ; X86-NOSSE-NEXT:    fmull 12(%ebp)
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-NOSSE-NEXT:    movl %edx, (%esp)
-; X86-NOSSE-NEXT:    movl %esi, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll (%ecx,%eax,8)
 ; X86-NOSSE-NEXT:    leal -4(%ebp), %esp
 ; X86-NOSSE-NEXT:    popl %esi
@@ -2407,15 +2347,11 @@ define dso_local void @fdiv_64r(ptr %loc, double %val) nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-NOSSE-NEXT:    movl %edx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fldl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %ecx, (%esp)
+; X86-NOSSE-NEXT:    fldl (%esp)
 ; X86-NOSSE-NEXT:    fdivl 12(%ebp)
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NOSSE-NEXT:    movl %ecx, (%esp)
-; X86-NOSSE-NEXT:    movl %edx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll (%eax)
 ; X86-NOSSE-NEXT:    movl %ebp, %esp
 ; X86-NOSSE-NEXT:    popl %ebp
@@ -2571,15 +2507,11 @@ define dso_local void @fdiv_64g() nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %eax, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fldl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %eax, (%esp)
+; X86-NOSSE-NEXT:    fldl (%esp)
 ; X86-NOSSE-NEXT:    fdivs {{\.?LCPI[0-9]+_[0-9]+}}
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl %eax, (%esp)
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll glob64
 ; X86-NOSSE-NEXT:    movl %ebp, %esp
 ; X86-NOSSE-NEXT:    popl %ebp
@@ -2734,15 +2666,11 @@ define dso_local void @fdiv_64imm() nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %eax, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fldl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %eax, (%esp)
+; X86-NOSSE-NEXT:    fldl (%esp)
 ; X86-NOSSE-NEXT:    fdivs {{\.?LCPI[0-9]+_[0-9]+}}
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl %eax, (%esp)
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll -559038737
 ; X86-NOSSE-NEXT:    movl %ebp, %esp
 ; X86-NOSSE-NEXT:    popl %ebp
@@ -2902,15 +2830,11 @@ define dso_local void @fdiv_64stack() nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %eax, {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %eax, (%esp)
 ; X86-NOSSE-NEXT:    fld1
-; X86-NOSSE-NEXT:    fdivl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    fdivl (%esp)
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl %eax, (%esp)
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    movl %ebp, %esp
 ; X86-NOSSE-NEXT:    popl %ebp
@@ -3006,15 +2930,11 @@ define dso_local void @fdiv_array(ptr %arg, double %arg1, i64 %arg2) nounwind {
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X86-NOSSE-NEXT:    movl %esi, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %edx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fldl {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl %edx, (%esp)
+; X86-NOSSE-NEXT:    fldl (%esp)
 ; X86-NOSSE-NEXT:    fdivl 12(%ebp)
 ; X86-NOSSE-NEXT:    fstpl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-NOSSE-NEXT:    movl %edx, (%esp)
-; X86-NOSSE-NEXT:    movl %esi, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    fildll (%esp)
+; X86-NOSSE-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X86-NOSSE-NEXT:    fistpll (%ecx,%eax,8)
 ; X86-NOSSE-NEXT:    leal -4(%ebp), %esp
 ; X86-NOSSE-NEXT:    popl %esi
diff --git a/llvm/test/CodeGen/X86/single_elt_vector_memory_operation.ll b/llvm/test/CodeGen/X86/single_elt_vector_memory_operation.ll
index f65461ccee23b..4e7a694f31c9c 100644
--- a/llvm/test/CodeGen/X86/single_elt_vector_memory_operation.ll
+++ b/llvm/test/CodeGen/X86/single_elt_vector_memory_operation.ll
@@ -87,8 +87,8 @@ define void @load_single_256bit_elt_vector(ptr %in, ptr %off, ptr %out) nounwind
 ; SSE-NEXT:    xorps %xmm2, %xmm2
 ; SSE-NEXT:    movaps %xmm2, 48(%rdx)
 ; SSE-NEXT:    movaps %xmm2, 32(%rdx)
-; SSE-NEXT:    movaps %xmm0, (%rdx)
 ; SSE-NEXT:    movaps %xmm1, 16(%rdx)
+; SSE-NEXT:    movaps %xmm0, (%rdx)
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: load_single_256bit_elt_vector:

arsenm
arsenm reviewed Dec 9, 2025
View reviewed changes
llvm/lib/CodeGen/SelectionDAG/LegalizeTypesGeneric.cpp
// Storing a bitcasted value, see if the original type is a legal store.
// TODO: Not necessary if we had proper topological sorting of nodes.
if (Value.getOpcode() == ISD::BITCAST &&
TLI.isOperationLegal(ISD::STORE, Value.getOperand(0).getValueType()))
Copy link
Contributor

@arsenm arsenm Dec 9, 2025

Choose a reason for hiding this comment

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

The state of memory instruction legality queries is bad. Does this need to worry about alignment, address space, and everything else?

Copy link
Collaborator Author

@RKSimon RKSimon Dec 9, 2025

Choose a reason for hiding this comment

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

We can add a isStoreBitCastBeneficial check as well similar to what DAGCombiner::visitSTORE does - which should do a allowsMemoryAccess check?

topperc
topperc reviewed Dec 9, 2025
View reviewed changes
llvm/test/CodeGen/X86/atomic-fp.ll
; X86-NOSSE-NEXT: movl %edx, {{[0-9]+}}(%esp)
; X86-NOSSE-NEXT: movl %ecx, {{[0-9]+}}(%esp)
; X86-NOSSE-NEXT: fldl {{[0-9]+}}(%esp)
; X86-NOSSE-NEXT: movl %ecx, (%esp)
Copy link
Collaborator

@topperc topperc Dec 9, 2025

Choose a reason for hiding this comment

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

This looks it should be handled in the custom lowering for i64 AtomicStore. Or we need to support f64 AtomicStore without a bitcast. We're spilling to the stack, but we don't need to.

Copy link
Collaborator Author

@RKSimon RKSimon Dec 10, 2025

Choose a reason for hiding this comment

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

Cheers, I'll take another look

RKSimon added a commit to RKSimon/llvm-project that referenced this pull request Dec 10, 2025
@RKSimon
[X86] LowerATOMIC_STORE - on 32-bit targets see if i64 values were or…
dcd7483 
…iginally legal f64 values that we can store directly.

Based off feedback from llvm#171478
@RKSimon RKSimon mentioned this pull request Dec 10, 2025
Merged
RKSimon added a commit that referenced this pull request Dec 11, 2025
@RKSimon
[X86] LowerATOMIC_STORE - on 32-bit targets see if i64 values were or…
6573f62 
…iginally legal f64 values that we can store directly. (#171602)

Based off feedback from #171478
@RKSimon RKSimon closed this Dec 12, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@arsenm arsenm arsenm left review comments

@topperc topperc topperc left review comments

@phoebewang phoebewang Awaiting requested review from phoebewang

Assignees

No one assigned

Labels

backend:X86 llvm:SelectionDAG SelectionDAGISel as well

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

4 participants

@RKSimon @llvmbot @arsenm @topperc