[llvm] [SDAG][X86] Promote FMODF to double on 32-bit Windows (PR #130636)

[via llvm-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-backend-x86

Author: Benjamin Maxwell (MacDue)

<details>
<summary>Changes</summary>

On 32-bit MSVC `modff` is not a defined symbol -- only `modf` (`modff` is an inline function). Promoting FMODF to double in this case ensures we end up calling `modf` -- matching the behaviour of the CRT headers.

---
Full diff: https://github.com/llvm/llvm-project/pull/130636.diff


2 Files Affected:

- (modified) llvm/lib/Target/X86/X86ISelLowering.cpp (+3-1) 
- (added) llvm/test/CodeGen/X86/llvm.modf-win32.ll (+40) 


``````````diff
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index ccd7f2418fcd1..3e02162020302 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -2634,7 +2634,9 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
           ISD::FSIN,   ISD::STRICT_FSIN,
           ISD::FSINH,  ISD::STRICT_FSINH,
           ISD::FTAN,   ISD::STRICT_FTAN,
-          ISD::FTANH,  ISD::STRICT_FTANH})
+          ISD::FTANH,  ISD::STRICT_FTANH,
+          // TODO: Add ISD:::STRICT_FMODF too once implemented.
+          ISD::FMODF})
       if (isOperationExpand(Op, MVT::f32))
         setOperationAction(Op, MVT::f32, Promote);
   // clang-format on
diff --git a/llvm/test/CodeGen/X86/llvm.modf-win32.ll b/llvm/test/CodeGen/X86/llvm.modf-win32.ll
new file mode 100644
index 0000000000000..70ce773dda482
--- /dev/null
+++ b/llvm/test/CodeGen/X86/llvm.modf-win32.ll
@@ -0,0 +1,40 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=i386-pc-win32 < %s | FileCheck -check-prefix=WIN32 %s
+; RUN: llc -mtriple=x86_64-unknown-unknown < %s | FileCheck -check-prefixes=X64 %s
+
+; On 32-bit windows this should be promoted to a call to modf (not modff).
+define { float, float } @test_modf_f32(float %a) {
+; WIN32-LABEL: test_modf_f32:
+; WIN32:       # %bb.0:
+; WIN32-NEXT:    pushl %ebp
+; WIN32-NEXT:    movl %esp, %ebp
+; WIN32-NEXT:    andl $-8, %esp
+; WIN32-NEXT:    subl $32, %esp
+; WIN32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; WIN32-NEXT:    movl %eax, {{[0-9]+}}(%esp)
+; WIN32-NEXT:    flds 8(%ebp)
+; WIN32-NEXT:    fstpl (%esp)
+; WIN32-NEXT:    calll _modf
+; WIN32-NEXT:    fstps {{[0-9]+}}(%esp)
+; WIN32-NEXT:    fldl {{[0-9]+}}(%esp)
+; WIN32-NEXT:    fstps {{[0-9]+}}(%esp)
+; WIN32-NEXT:    flds {{[0-9]+}}(%esp)
+; WIN32-NEXT:    flds {{[0-9]+}}(%esp)
+; WIN32-NEXT:    fxch %st(1)
+; WIN32-NEXT:    movl %ebp, %esp
+; WIN32-NEXT:    popl %ebp
+; WIN32-NEXT:    retl
+;
+; X64-LABEL: test_modf_f32:
+; X64:       # %bb.0:
+; X64-NEXT:    pushq %rax
+; X64-NEXT:    .cfi_def_cfa_offset 16
+; X64-NEXT:    leaq {{[0-9]+}}(%rsp), %rdi
+; X64-NEXT:    callq modff at PLT
+; X64-NEXT:    movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; X64-NEXT:    popq %rax
+; X64-NEXT:    .cfi_def_cfa_offset 8
+; X64-NEXT:    retq
+  %result = call { float, float } @llvm.modf.f32(float %a)
+  ret { float, float } %result
+}

``````````

</details>


https://github.com/llvm/llvm-project/pull/130636