[llvm] [AArch64] Lower mathlib call ldexp into fscale when sve is enabled (PR #67552)

[via llvm-commits]

================
@@ -0,0 +1,66 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 3
+; RUN: llc -mtriple=aarch64 -mattr=+sve < %s -o - | FileCheck %s
+
+define double @testExp(double %val, i32 %a) {
+; CHECK-LABEL: testExp:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    ptrue p0.d
+; CHECK-NEXT:    // kill: def $w0 killed $w0 def $x0
+; CHECK-NEXT:    sxtw x8, w0
+; CHECK-NEXT:    // kill: def $d0 killed $d0 def $z0
+; CHECK-NEXT:    fmov d1, x8
+; CHECK-NEXT:    fscale z0.d, p0/m, z0.d, z1.d
+; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $z0
+; CHECK-NEXT:    ret
+entry:
+  %call = tail call fast double @ldexp(double %val, i32 %a)
+  ret double %call
+}
+
+declare double @ldexp(double, i32) #1
+
+define float @testExpf(float %val, i32 %a) {
+; CHECK-LABEL: testExpf:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    ptrue p0.s
+; CHECK-NEXT:    fmov s1, w0
+; CHECK-NEXT:    // kill: def $s0 killed $s0 def $z0
+; CHECK-NEXT:    fscale z0.s, p0/m, z0.s, z1.s
+; CHECK-NEXT:    // kill: def $s0 killed $s0 killed $z0
+; CHECK-NEXT:    ret
+entry:
+  %call = tail call fast float @ldexpf(float %val, i32 %a)
+  ret float %call
+}
+
+declare float @ldexpf(float, i32) #1
+
+define fp128 @testExpl(fp128 %val, i32 %a) {
+; CHECK-LABEL: testExpl:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    b ldexpl
+entry:
+  %call = tail call fast fp128 @ldexpl(fp128 %val, i32 %a)
+  ret fp128 %call
+}
+
+declare fp128 @ldexpl(fp128, i32) #1
+
+define half @testExpf16(half %val, i32 %a) {
+; CHECK-LABEL: testExpf16:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    ptrue p0.s
+; CHECK-NEXT:    fcvt s0, h0
+; CHECK-NEXT:    fmov s1, w0
+; CHECK-NEXT:    fscale z0.s, p0/m, z0.s, z1.s
+; CHECK-NEXT:    fcvt h0, s0
+; CHECK-NEXT:    ret
+entry:
+  %0 = tail call fast half @llvm.ldexp.f16.i32(half %val, i32 %a)
+  ret half %0
+}
+
+declare half @llvm.ldexp.f16.i32(half, i32) #1
+
+attributes #1 = { mustprogress nofree nosync nounwind willreturn memory(none) }
----------------
huhu233 wrote:

Done, only keep the necessary attribute.

https://github.com/llvm/llvm-project/pull/67552