AMDGPU: Change fdiv lowering based on !fpmath metadata
If 2.5 ulp is acceptable, denormals are not required, and
isn't a reciprocal which will already be handled, replace
with a faster fdiv.
Simplify the lowering tests by using per function
subtarget features.
llvm-svn: 276051
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.fdiv.fast.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.fdiv.fast.ll
new file mode 100644
index 0000000..54d7848
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.fdiv.fast.ll
@@ -0,0 +1,18 @@
+; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck %s
+
+declare float @llvm.amdgcn.fdiv.fast(float, float) #0
+
+; CHECK-LABEL: {{^}}test_fdiv_fast:
+; CHECK: v_cndmask_b32_e32 v{{[0-9]+}}, 1.0, v{{[0-9]+}}, vcc
+; CHECK: v_mul_f32_e32
+; CHECK: v_rcp_f32_e32
+; CHECK: v_mul_f32_e32
+; CHECK: v_mul_f32_e32
+define void @test_fdiv_fast(float addrspace(1)* %out, float %a, float %b) #1 {
+ %fdiv = call float @llvm.amdgcn.fdiv.fast(float %a, float %b)
+ store float %fdiv, float addrspace(1)* %out
+ ret void
+}
+
+attributes #0 = { nounwind readnone }
+attributes #1 = { nounwind }