[llvm-branch-commits] [llvm] DAG: Remove special case ldexp libcall recognition (PR #241467)

Matt Arsenault via llvm-branch-commits llvm-branch-commits at lists.llvm.org
Wed Nov 11 12:07:03 PST 2026


================
@@ -9,15 +9,17 @@ define float @call_ldexpf(float %a, i32 %b) nounwind {
 ;
 ; CHECK-WIN-LABEL: call_ldexpf:
 ; CHECK-WIN:       # %bb.0:
-; CHECK-WIN-NEXT:    subl $8, %esp
-; CHECK-WIN-NEXT:    flds {{[0-9]+}}(%esp)
+; CHECK-WIN-NEXT:    subl $16, %esp
 ; CHECK-WIN-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; CHECK-WIN-NEXT:    movl %eax, {{[0-9]+}}(%esp)
-; CHECK-WIN-NEXT:    fstps (%esp)
-; CHECK-WIN-NEXT:    calll _ldexpf
-; CHECK-WIN-NEXT:    addl $8, %esp
+; CHECK-WIN-NEXT:    flds {{[0-9]+}}(%esp)
+; CHECK-WIN-NEXT:    fstpl (%esp)
+; CHECK-WIN-NEXT:    calll _ldexp
+; CHECK-WIN-NEXT:    fstps {{[0-9]+}}(%esp)
+; CHECK-WIN-NEXT:    flds {{[0-9]+}}(%esp)
+; CHECK-WIN-NEXT:    addl $16, %esp
 ; CHECK-WIN-NEXT:    retl
-  %result = call float @ldexpf(float %a, i32 %b)
+  %result = call float @llvm.ldexp.f32.i32(float %a, i32 %b)
----------------
arsenm wrote:

Windows doesn't have the float version so it needs to promote to double 

https://github.com/llvm/llvm-project/pull/241467


More information about the llvm-branch-commits mailing list