Commit d657321a authored by Olivier Sallenave's avatar Olivier Sallenave
Browse files

Check that the TLI callback enableAggressiveFMAFusion has the desired effect on FMA folding. 

llvm-svn: 225987
parent 0181812b

llvm/test/CodeGen/NVPTX/fma-assoc.ll

0 → 100644
+25 −0
Original line number Diff line number Diff line 
; RUN: llc < %s -march=nvptx -mcpu=sm_20 -fp-contract=fast | FileCheck %s



define ptx_device float @t1_f32(float %x, float %y, float %z,

                                float %u, float %v) {

; CHECK: fma.rn.f32 %f{{[0-9]+}}, %f{{[0-9]+}}, %f{{[0-9]+}}, %f{{[0-9]+}};

; CHECK: fma.rn.f32 %f{{[0-9]+}}, %f{{[0-9]+}}, %f{{[0-9]+}}, %f{{[0-9]+}};

; CHECK: ret;

  %a = fmul float %x, %y

  %b = fmul float %u, %v

  %c = fadd float %a, %b

  %d = fadd float %c, %z

  ret float %d

}



define ptx_device double @t1_f64(double %x, double %y, double %z,

                                 double %u, double %v) {

; CHECK: fma.rn.f64 %fd{{[0-9]+}}, %fd{{[0-9]+}}, %fd{{[0-9]+}}, %fd{{[0-9]+}};

; CHECK: fma.rn.f64 %fd{{[0-9]+}}, %fd{{[0-9]+}}, %fd{{[0-9]+}}, %fd{{[0-9]+}};

; CHECK: ret;

  %a = fmul double %x, %y

  %b = fmul double %u, %v

  %c = fadd double %a, %b

  %d = fadd double %c, %z

  ret double %d

}

llvm/test/CodeGen/NVPTX/fma.ll

+25 −0
Original line number Diff line number Diff line 
; RUN: llc < %s -march=nvptx -mcpu=sm_20 -fp-contract=fast | FileCheck %s



declare float @dummy_f32(float, float) #0

declare double @dummy_f64(double, double) #0



define ptx_device float @t1_f32(float %x, float %y, float %z) {

; CHECK: fma.rn.f32 %f{{[0-9]+}}, %f{{[0-9]+}}, %f{{[0-9]+}}, %f{{[0-9]+}};

; CHECK: ret;
@@ -8,6 +11,17 @@ define ptx_device float @t1_f32(float %x, float %y, float %z) { 
  ret float %b

}



define ptx_device float @t2_f32(float %x, float %y, float %z, float %w) {

; CHECK: fma.rn.f32 %f{{[0-9]+}}, %f{{[0-9]+}}, %f{{[0-9]+}}, %f{{[0-9]+}};

; CHECK: fma.rn.f32 %f{{[0-9]+}}, %f{{[0-9]+}}, %f{{[0-9]+}}, %f{{[0-9]+}};

; CHECK: ret;

  %a = fmul float %x, %y

  %b = fadd float %a, %z

  %c = fadd float %a, %w

  %d = call float @dummy_f32(float %b, float %c)

  ret float %d

}



define ptx_device double @t1_f64(double %x, double %y, double %z) {

; CHECK: fma.rn.f64 %fd{{[0-9]+}}, %fd{{[0-9]+}}, %fd{{[0-9]+}}, %fd{{[0-9]+}};

; CHECK: ret;
@@ -15,3 +29,14 @@ define ptx_device double @t1_f64(double %x, double %y, double %z) { 
  %b = fadd double %a, %z

  ret double %b

}



define ptx_device double @t2_f64(double %x, double %y, double %z, double %w) {

; CHECK: fma.rn.f64 %fd{{[0-9]+}}, %fd{{[0-9]+}}, %fd{{[0-9]+}}, %fd{{[0-9]+}};

; CHECK: fma.rn.f64 %fd{{[0-9]+}}, %fd{{[0-9]+}}, %fd{{[0-9]+}}, %fd{{[0-9]+}};

; CHECK: ret;

  %a = fmul double %x, %y

  %b = fadd double %a, %z

  %c = fadd double %a, %w

  %d = call double @dummy_f64(double %b, double %c)

  ret double %d

}