test/CodeGen/AMDGPU/fadd-fma-fmul-combine.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -march=amdgcn -mattr=+fast-fmaf,+mad-mac-f32-insts -denormal-fp-math-f32=preserve-sign -enable-unsafe-fp-math -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GCN-FLUSH %s
; RUN: llc -march=amdgcn -mattr=-fast-fmaf,+mad-mac-f32-insts -denormal-fp-math-f32=preserve-sign -enable-unsafe-fp-math -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GCN-FLUSH %s

; RUN: llc -march=amdgcn -mattr=+fast-fmaf,+mad-mac-f32-insts -denormal-fp-math-f32=ieee  -enable-unsafe-fp-math -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GCN-FASTFMA %s
; RUN: llc -march=amdgcn -mattr=-fast-fmaf,+mad-mac-f32-insts -denormal-fp-math-f32=ieee -enable-unsafe-fp-math -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GCN-SLOWFMA %s

; FIXME: This should also fold when fma is actually fast if an FMA
; exists in the original program.

; (fadd (fma x, y, (fmul u, v), z) -> (fma x, y (fma u, v, z))
define amdgpu_kernel void @fast_add_fmuladd_fmul() #0 {
; GCN-FLUSH-LABEL: fast_add_fmuladd_fmul:
; GCN-FLUSH:       ; %bb.0:
; GCN-FLUSH-NEXT:    s_mov_b32 s3, 0xf000
; GCN-FLUSH-NEXT:    s_mov_b32 s2, -1
; GCN-FLUSH-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v3, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v4, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    v_mac_f32_e32 v2, v3, v4
; GCN-FLUSH-NEXT:    v_mac_f32_e32 v2, v0, v1
; GCN-FLUSH-NEXT:    buffer_store_dword v2, off, s[0:3], 0
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    s_endpgm
;
; GCN-FASTFMA-LABEL: fast_add_fmuladd_fmul:
; GCN-FASTFMA:       ; %bb.0:
; GCN-FASTFMA-NEXT:    s_mov_b32 s3, 0xf000
; GCN-FASTFMA-NEXT:    s_mov_b32 s2, -1
; GCN-FASTFMA-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v3, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v4, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    v_fma_f32 v2, v3, v4, v2
; GCN-FASTFMA-NEXT:    v_fma_f32 v0, v0, v1, v2
; GCN-FASTFMA-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    s_endpgm
;
; GCN-SLOWFMA-LABEL: fast_add_fmuladd_fmul:
; GCN-SLOWFMA:       ; %bb.0:
; GCN-SLOWFMA-NEXT:    s_mov_b32 s3, 0xf000
; GCN-SLOWFMA-NEXT:    s_mov_b32 s2, -1
; GCN-SLOWFMA-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v3, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v4, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    v_mul_f32_e32 v3, v3, v4
; GCN-SLOWFMA-NEXT:    v_mul_f32_e32 v0, v0, v1
; GCN-SLOWFMA-NEXT:    v_add_f32_e32 v0, v0, v3
; GCN-SLOWFMA-NEXT:    v_add_f32_e32 v0, v0, v2
; GCN-SLOWFMA-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    s_endpgm
  %x = load volatile float, ptr addrspace(1) undef
  %y = load volatile float, ptr addrspace(1) undef
  %z = load volatile float, ptr addrspace(1) undef
  %u = load volatile float, ptr addrspace(1) undef
  %v = load volatile float, ptr addrspace(1) undef
  %mul.u.v = fmul fast float %u, %v
  %fma = call fast float @llvm.fmuladd.f32(float %x, float %y, float %mul.u.v)
  %add = fadd fast float %fma, %z
  store volatile float %add, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @fast_sub_fmuladd_fmul() #0 {
; GCN-FLUSH-LABEL: fast_sub_fmuladd_fmul:
; GCN-FLUSH:       ; %bb.0:
; GCN-FLUSH-NEXT:    s_mov_b32 s3, 0xf000
; GCN-FLUSH-NEXT:    s_mov_b32 s2, -1
; GCN-FLUSH-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v3, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v4, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    v_mad_f32 v2, v3, v4, -v2
; GCN-FLUSH-NEXT:    v_mac_f32_e32 v2, v0, v1
; GCN-FLUSH-NEXT:    buffer_store_dword v2, off, s[0:3], 0
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    s_endpgm
;
; GCN-FASTFMA-LABEL: fast_sub_fmuladd_fmul:
; GCN-FASTFMA:       ; %bb.0:
; GCN-FASTFMA-NEXT:    s_mov_b32 s3, 0xf000
; GCN-FASTFMA-NEXT:    s_mov_b32 s2, -1
; GCN-FASTFMA-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v3, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v4, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    v_fma_f32 v2, v3, v4, -v2
; GCN-FASTFMA-NEXT:    v_fma_f32 v0, v0, v1, v2
; GCN-FASTFMA-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    s_endpgm
;
; GCN-SLOWFMA-LABEL: fast_sub_fmuladd_fmul:
; GCN-SLOWFMA:       ; %bb.0:
; GCN-SLOWFMA-NEXT:    s_mov_b32 s3, 0xf000
; GCN-SLOWFMA-NEXT:    s_mov_b32 s2, -1
; GCN-SLOWFMA-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v3, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v4, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    v_mul_f32_e32 v3, v3, v4
; GCN-SLOWFMA-NEXT:    v_mul_f32_e32 v0, v0, v1
; GCN-SLOWFMA-NEXT:    v_add_f32_e32 v0, v0, v3
; GCN-SLOWFMA-NEXT:    v_sub_f32_e32 v0, v0, v2
; GCN-SLOWFMA-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    s_endpgm
  %x = load volatile float, ptr addrspace(1) undef
  %y = load volatile float, ptr addrspace(1) undef
  %z = load volatile float, ptr addrspace(1) undef
  %u = load volatile float, ptr addrspace(1) undef
  %v = load volatile float, ptr addrspace(1) undef
  %mul.u.v = fmul fast float %u, %v
  %fma = call fast float @llvm.fmuladd.f32(float %x, float %y, float %mul.u.v)
  %add = fsub fast float %fma, %z
  store volatile float %add, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @fast_add_fmuladd_fmul_multi_use_mul() #0 {
; GCN-FLUSH-LABEL: fast_add_fmuladd_fmul_multi_use_mul:
; GCN-FLUSH:       ; %bb.0:
; GCN-FLUSH-NEXT:    s_mov_b32 s3, 0xf000
; GCN-FLUSH-NEXT:    s_mov_b32 s2, -1
; GCN-FLUSH-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v3, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v4, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    v_mul_f32_e32 v3, v3, v4
; GCN-FLUSH-NEXT:    buffer_store_dword v3, off, s[0:3], 0
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0) expcnt(0)
; GCN-FLUSH-NEXT:    v_mac_f32_e32 v3, v0, v1
; GCN-FLUSH-NEXT:    v_add_f32_e32 v0, v3, v2
; GCN-FLUSH-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    s_endpgm
;
; GCN-FASTFMA-LABEL: fast_add_fmuladd_fmul_multi_use_mul:
; GCN-FASTFMA:       ; %bb.0:
; GCN-FASTFMA-NEXT:    s_mov_b32 s3, 0xf000
; GCN-FASTFMA-NEXT:    s_mov_b32 s2, -1
; GCN-FASTFMA-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v3, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v4, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    v_mul_f32_e32 v3, v3, v4
; GCN-FASTFMA-NEXT:    buffer_store_dword v3, off, s[0:3], 0
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    v_fma_f32 v0, v0, v1, v3
; GCN-FASTFMA-NEXT:    v_add_f32_e32 v0, v0, v2
; GCN-FASTFMA-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    s_endpgm
;
; GCN-SLOWFMA-LABEL: fast_add_fmuladd_fmul_multi_use_mul:
; GCN-SLOWFMA:       ; %bb.0:
; GCN-SLOWFMA-NEXT:    s_mov_b32 s3, 0xf000
; GCN-SLOWFMA-NEXT:    s_mov_b32 s2, -1
; GCN-SLOWFMA-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v3, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v4, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    v_mul_f32_e32 v3, v3, v4
; GCN-SLOWFMA-NEXT:    v_mul_f32_e32 v0, v0, v1
; GCN-SLOWFMA-NEXT:    buffer_store_dword v3, off, s[0:3], 0
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    v_add_f32_e32 v0, v0, v3
; GCN-SLOWFMA-NEXT:    v_add_f32_e32 v0, v0, v2
; GCN-SLOWFMA-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    s_endpgm
  %x = load volatile float, ptr addrspace(1) undef
  %y = load volatile float, ptr addrspace(1) undef
  %z = load volatile float, ptr addrspace(1) undef
  %u = load volatile float, ptr addrspace(1) undef
  %v = load volatile float, ptr addrspace(1) undef
  %mul.u.v = fmul fast float %u, %v
  store volatile float %mul.u.v, ptr addrspace(1) undef
  %fma = call fast float @llvm.fmuladd.f32(float %x, float %y, float %mul.u.v)
  %add = fadd fast float %fma, %z
  store volatile float %add, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @fast_add_fmuladd_fmul_multi_use_mul_commute() #0 {
; GCN-FLUSH-LABEL: fast_add_fmuladd_fmul_multi_use_mul_commute:
; GCN-FLUSH:       ; %bb.0:
; GCN-FLUSH-NEXT:    s_mov_b32 s3, 0xf000
; GCN-FLUSH-NEXT:    s_mov_b32 s2, -1
; GCN-FLUSH-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v3, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v4, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    v_mul_f32_e32 v3, v3, v4
; GCN-FLUSH-NEXT:    buffer_store_dword v3, off, s[0:3], 0
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0) expcnt(0)
; GCN-FLUSH-NEXT:    v_mac_f32_e32 v3, v0, v1
; GCN-FLUSH-NEXT:    v_add_f32_e32 v0, v2, v3
; GCN-FLUSH-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    s_endpgm
;
; GCN-FASTFMA-LABEL: fast_add_fmuladd_fmul_multi_use_mul_commute:
; GCN-FASTFMA:       ; %bb.0:
; GCN-FASTFMA-NEXT:    s_mov_b32 s3, 0xf000
; GCN-FASTFMA-NEXT:    s_mov_b32 s2, -1
; GCN-FASTFMA-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v3, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v4, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    v_mul_f32_e32 v3, v3, v4
; GCN-FASTFMA-NEXT:    buffer_store_dword v3, off, s[0:3], 0
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    v_fma_f32 v0, v0, v1, v3
; GCN-FASTFMA-NEXT:    v_add_f32_e32 v0, v2, v0
; GCN-FASTFMA-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    s_endpgm
;
; GCN-SLOWFMA-LABEL: fast_add_fmuladd_fmul_multi_use_mul_commute:
; GCN-SLOWFMA:       ; %bb.0:
; GCN-SLOWFMA-NEXT:    s_mov_b32 s3, 0xf000
; GCN-SLOWFMA-NEXT:    s_mov_b32 s2, -1
; GCN-SLOWFMA-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v3, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v4, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    v_mul_f32_e32 v3, v3, v4
; GCN-SLOWFMA-NEXT:    v_mul_f32_e32 v0, v0, v1
; GCN-SLOWFMA-NEXT:    buffer_store_dword v3, off, s[0:3], 0
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    v_add_f32_e32 v0, v0, v3
; GCN-SLOWFMA-NEXT:    v_add_f32_e32 v0, v2, v0
; GCN-SLOWFMA-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    s_endpgm
  %x = load volatile float, ptr addrspace(1) undef
  %y = load volatile float, ptr addrspace(1) undef
  %z = load volatile float, ptr addrspace(1) undef
  %u = load volatile float, ptr addrspace(1) undef
  %v = load volatile float, ptr addrspace(1) undef
  %mul.u.v = fmul fast float %u, %v
  store volatile float %mul.u.v, ptr addrspace(1) undef
  %fma = call fast float @llvm.fmuladd.f32(float %x, float %y, float %mul.u.v)
  %add = fadd fast float %z, %fma
  store volatile float %add, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @fast_add_fmuladd_fmul_multi_use_fmuladd() #0 {
; GCN-FLUSH-LABEL: fast_add_fmuladd_fmul_multi_use_fmuladd:
; GCN-FLUSH:       ; %bb.0:
; GCN-FLUSH-NEXT:    s_mov_b32 s3, 0xf000
; GCN-FLUSH-NEXT:    s_mov_b32 s2, -1
; GCN-FLUSH-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v3, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v4, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    v_mul_f32_e32 v3, v3, v4
; GCN-FLUSH-NEXT:    v_mac_f32_e32 v3, v0, v1
; GCN-FLUSH-NEXT:    buffer_store_dword v3, off, s[0:3], 0
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    v_add_f32_e32 v0, v3, v2
; GCN-FLUSH-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    s_endpgm
;
; GCN-FASTFMA-LABEL: fast_add_fmuladd_fmul_multi_use_fmuladd:
; GCN-FASTFMA:       ; %bb.0:
; GCN-FASTFMA-NEXT:    s_mov_b32 s3, 0xf000
; GCN-FASTFMA-NEXT:    s_mov_b32 s2, -1
; GCN-FASTFMA-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v3, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v4, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    v_mul_f32_e32 v3, v3, v4
; GCN-FASTFMA-NEXT:    v_fma_f32 v0, v0, v1, v3
; GCN-FASTFMA-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0) expcnt(0)
; GCN-FASTFMA-NEXT:    v_add_f32_e32 v0, v0, v2
; GCN-FASTFMA-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    s_endpgm
;
; GCN-SLOWFMA-LABEL: fast_add_fmuladd_fmul_multi_use_fmuladd:
; GCN-SLOWFMA:       ; %bb.0:
; GCN-SLOWFMA-NEXT:    s_mov_b32 s3, 0xf000
; GCN-SLOWFMA-NEXT:    s_mov_b32 s2, -1
; GCN-SLOWFMA-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v3, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v4, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    v_mul_f32_e32 v3, v3, v4
; GCN-SLOWFMA-NEXT:    v_mul_f32_e32 v0, v0, v1
; GCN-SLOWFMA-NEXT:    v_add_f32_e32 v0, v0, v3
; GCN-SLOWFMA-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0) expcnt(0)
; GCN-SLOWFMA-NEXT:    v_add_f32_e32 v0, v0, v2
; GCN-SLOWFMA-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    s_endpgm
  %x = load volatile float, ptr addrspace(1) undef
  %y = load volatile float, ptr addrspace(1) undef
  %z = load volatile float, ptr addrspace(1) undef
  %u = load volatile float, ptr addrspace(1) undef
  %v = load volatile float, ptr addrspace(1) undef
  %mul.u.v = fmul fast float %u, %v
  %fma = call fast float @llvm.fmuladd.f32(float %x, float %y, float %mul.u.v)
  store volatile float %fma, ptr addrspace(1) undef
  %add = fadd fast float %fma, %z
  store volatile float %add, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @fast_add_fmuladd_fmul_multi_use_fmuladd_commute() #0 {
; GCN-FLUSH-LABEL: fast_add_fmuladd_fmul_multi_use_fmuladd_commute:
; GCN-FLUSH:       ; %bb.0:
; GCN-FLUSH-NEXT:    s_mov_b32 s3, 0xf000
; GCN-FLUSH-NEXT:    s_mov_b32 s2, -1
; GCN-FLUSH-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v3, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v4, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    v_mul_f32_e32 v3, v3, v4
; GCN-FLUSH-NEXT:    v_mac_f32_e32 v3, v0, v1
; GCN-FLUSH-NEXT:    buffer_store_dword v3, off, s[0:3], 0
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    v_add_f32_e32 v0, v2, v3
; GCN-FLUSH-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    s_endpgm
;
; GCN-FASTFMA-LABEL: fast_add_fmuladd_fmul_multi_use_fmuladd_commute:
; GCN-FASTFMA:       ; %bb.0:
; GCN-FASTFMA-NEXT:    s_mov_b32 s3, 0xf000
; GCN-FASTFMA-NEXT:    s_mov_b32 s2, -1
; GCN-FASTFMA-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v3, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v4, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    v_mul_f32_e32 v3, v3, v4
; GCN-FASTFMA-NEXT:    v_fma_f32 v0, v0, v1, v3
; GCN-FASTFMA-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0) expcnt(0)
; GCN-FASTFMA-NEXT:    v_add_f32_e32 v0, v2, v0
; GCN-FASTFMA-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    s_endpgm
;
; GCN-SLOWFMA-LABEL: fast_add_fmuladd_fmul_multi_use_fmuladd_commute:
; GCN-SLOWFMA:       ; %bb.0:
; GCN-SLOWFMA-NEXT:    s_mov_b32 s3, 0xf000
; GCN-SLOWFMA-NEXT:    s_mov_b32 s2, -1
; GCN-SLOWFMA-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v3, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v4, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    v_mul_f32_e32 v3, v3, v4
; GCN-SLOWFMA-NEXT:    v_mul_f32_e32 v0, v0, v1
; GCN-SLOWFMA-NEXT:    v_add_f32_e32 v0, v0, v3
; GCN-SLOWFMA-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0) expcnt(0)
; GCN-SLOWFMA-NEXT:    v_add_f32_e32 v0, v2, v0
; GCN-SLOWFMA-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    s_endpgm
  %x = load volatile float, ptr addrspace(1) undef
  %y = load volatile float, ptr addrspace(1) undef
  %z = load volatile float, ptr addrspace(1) undef
  %u = load volatile float, ptr addrspace(1) undef
  %v = load volatile float, ptr addrspace(1) undef
  %mul.u.v = fmul fast float %u, %v
  %fma = call fast float @llvm.fmuladd.f32(float %x, float %y, float %mul.u.v)
  store volatile float %fma, ptr addrspace(1) undef
  %add = fadd fast float %z, %fma
  store volatile float %add, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @fast_sub_fmuladd_fmul_multi_use_mul() #0 {
; GCN-FLUSH-LABEL: fast_sub_fmuladd_fmul_multi_use_mul:
; GCN-FLUSH:       ; %bb.0:
; GCN-FLUSH-NEXT:    s_mov_b32 s3, 0xf000
; GCN-FLUSH-NEXT:    s_mov_b32 s2, -1
; GCN-FLUSH-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v3, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v4, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    v_mul_f32_e32 v3, v3, v4
; GCN-FLUSH-NEXT:    v_mad_f32 v0, v0, v1, v3
; GCN-FLUSH-NEXT:    v_sub_f32_e32 v0, v0, v2
; GCN-FLUSH-NEXT:    buffer_store_dword v3, off, s[0:3], 0
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    s_endpgm
;
; GCN-FASTFMA-LABEL: fast_sub_fmuladd_fmul_multi_use_mul:
; GCN-FASTFMA:       ; %bb.0:
; GCN-FASTFMA-NEXT:    s_mov_b32 s3, 0xf000
; GCN-FASTFMA-NEXT:    s_mov_b32 s2, -1
; GCN-FASTFMA-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v3, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v4, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    v_mul_f32_e32 v3, v3, v4
; GCN-FASTFMA-NEXT:    v_fma_f32 v0, v0, v1, v3
; GCN-FASTFMA-NEXT:    v_sub_f32_e32 v0, v0, v2
; GCN-FASTFMA-NEXT:    buffer_store_dword v3, off, s[0:3], 0
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    s_endpgm
;
; GCN-SLOWFMA-LABEL: fast_sub_fmuladd_fmul_multi_use_mul:
; GCN-SLOWFMA:       ; %bb.0:
; GCN-SLOWFMA-NEXT:    s_mov_b32 s3, 0xf000
; GCN-SLOWFMA-NEXT:    s_mov_b32 s2, -1
; GCN-SLOWFMA-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v3, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v4, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    v_mul_f32_e32 v3, v3, v4
; GCN-SLOWFMA-NEXT:    v_mul_f32_e32 v0, v0, v1
; GCN-SLOWFMA-NEXT:    v_add_f32_e32 v0, v0, v3
; GCN-SLOWFMA-NEXT:    v_sub_f32_e32 v0, v0, v2
; GCN-SLOWFMA-NEXT:    buffer_store_dword v3, off, s[0:3], 0
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    s_endpgm
  %x = load volatile float, ptr addrspace(1) undef
  %y = load volatile float, ptr addrspace(1) undef
  %z = load volatile float, ptr addrspace(1) undef
  %u = load volatile float, ptr addrspace(1) undef
  %v = load volatile float, ptr addrspace(1) undef
  %mul.u.v = fmul fast float %u, %v
  %fma = call fast float @llvm.fmuladd.f32(float %x, float %y, float %mul.u.v)
  %sub = fsub fast float %fma, %z
  store volatile float %mul.u.v, ptr addrspace(1) undef
  store volatile float %sub, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @fast_sub_fmuladd_fmul_multi_use_fmuladd_lhs() #0 {
; GCN-FLUSH-LABEL: fast_sub_fmuladd_fmul_multi_use_fmuladd_lhs:
; GCN-FLUSH:       ; %bb.0:
; GCN-FLUSH-NEXT:    s_mov_b32 s3, 0xf000
; GCN-FLUSH-NEXT:    s_mov_b32 s2, -1
; GCN-FLUSH-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v3, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v4, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    v_mul_f32_e32 v3, v3, v4
; GCN-FLUSH-NEXT:    v_mac_f32_e32 v3, v0, v1
; GCN-FLUSH-NEXT:    v_sub_f32_e32 v0, v3, v2
; GCN-FLUSH-NEXT:    buffer_store_dword v3, off, s[0:3], 0
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    s_endpgm
;
; GCN-FASTFMA-LABEL: fast_sub_fmuladd_fmul_multi_use_fmuladd_lhs:
; GCN-FASTFMA:       ; %bb.0:
; GCN-FASTFMA-NEXT:    s_mov_b32 s3, 0xf000
; GCN-FASTFMA-NEXT:    s_mov_b32 s2, -1
; GCN-FASTFMA-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v3, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v4, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    v_mul_f32_e32 v3, v3, v4
; GCN-FASTFMA-NEXT:    v_fma_f32 v0, v0, v1, v3
; GCN-FASTFMA-NEXT:    v_sub_f32_e32 v1, v0, v2
; GCN-FASTFMA-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_store_dword v1, off, s[0:3], 0
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    s_endpgm
;
; GCN-SLOWFMA-LABEL: fast_sub_fmuladd_fmul_multi_use_fmuladd_lhs:
; GCN-SLOWFMA:       ; %bb.0:
; GCN-SLOWFMA-NEXT:    s_mov_b32 s3, 0xf000
; GCN-SLOWFMA-NEXT:    s_mov_b32 s2, -1
; GCN-SLOWFMA-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v3, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v4, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    v_mul_f32_e32 v3, v3, v4
; GCN-SLOWFMA-NEXT:    v_mul_f32_e32 v0, v0, v1
; GCN-SLOWFMA-NEXT:    v_add_f32_e32 v0, v0, v3
; GCN-SLOWFMA-NEXT:    v_sub_f32_e32 v1, v0, v2
; GCN-SLOWFMA-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_store_dword v1, off, s[0:3], 0
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    s_endpgm
  %x = load volatile float, ptr addrspace(1) undef
  %y = load volatile float, ptr addrspace(1) undef
  %z = load volatile float, ptr addrspace(1) undef
  %u = load volatile float, ptr addrspace(1) undef
  %v = load volatile float, ptr addrspace(1) undef
  %mul.u.v = fmul fast float %u, %v
  %fma = call fast float @llvm.fmuladd.f32(float %x, float %y, float %mul.u.v)
  %add = fsub fast float %fma, %z
  store volatile float %fma, ptr addrspace(1) undef
  store volatile float %add, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @fast_sub_fmuladd_fmul_multi_use_fmuladd_rhs() #0 {
; GCN-FLUSH-LABEL: fast_sub_fmuladd_fmul_multi_use_fmuladd_rhs:
; GCN-FLUSH:       ; %bb.0:
; GCN-FLUSH-NEXT:    s_mov_b32 s3, 0xf000
; GCN-FLUSH-NEXT:    s_mov_b32 s2, -1
; GCN-FLUSH-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v3, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v4, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    v_mul_f32_e32 v3, v3, v4
; GCN-FLUSH-NEXT:    v_mac_f32_e32 v3, v0, v1
; GCN-FLUSH-NEXT:    v_sub_f32_e32 v0, v2, v3
; GCN-FLUSH-NEXT:    buffer_store_dword v3, off, s[0:3], 0
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    s_endpgm
;
; GCN-FASTFMA-LABEL: fast_sub_fmuladd_fmul_multi_use_fmuladd_rhs:
; GCN-FASTFMA:       ; %bb.0:
; GCN-FASTFMA-NEXT:    s_mov_b32 s3, 0xf000
; GCN-FASTFMA-NEXT:    s_mov_b32 s2, -1
; GCN-FASTFMA-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v3, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v4, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    v_mul_f32_e32 v3, v3, v4
; GCN-FASTFMA-NEXT:    v_fma_f32 v0, v0, v1, v3
; GCN-FASTFMA-NEXT:    v_sub_f32_e32 v1, v2, v0
; GCN-FASTFMA-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_store_dword v1, off, s[0:3], 0
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    s_endpgm
;
; GCN-SLOWFMA-LABEL: fast_sub_fmuladd_fmul_multi_use_fmuladd_rhs:
; GCN-SLOWFMA:       ; %bb.0:
; GCN-SLOWFMA-NEXT:    s_mov_b32 s3, 0xf000
; GCN-SLOWFMA-NEXT:    s_mov_b32 s2, -1
; GCN-SLOWFMA-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v3, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v4, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    v_mul_f32_e32 v3, v3, v4
; GCN-SLOWFMA-NEXT:    v_mul_f32_e32 v0, v0, v1
; GCN-SLOWFMA-NEXT:    v_add_f32_e32 v0, v0, v3
; GCN-SLOWFMA-NEXT:    v_sub_f32_e32 v1, v2, v0
; GCN-SLOWFMA-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_store_dword v1, off, s[0:3], 0
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    s_endpgm
  %x = load volatile float, ptr addrspace(1) undef
  %y = load volatile float, ptr addrspace(1) undef
  %z = load volatile float, ptr addrspace(1) undef
  %u = load volatile float, ptr addrspace(1) undef
  %v = load volatile float, ptr addrspace(1) undef
  %mul.u.v = fmul fast float %u, %v
  %fma = call fast float @llvm.fmuladd.f32(float %x, float %y, float %mul.u.v)
  %add = fsub fast float %z, %fma
  store volatile float %fma, ptr addrspace(1) undef
  store volatile float %add, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @fast_sub_fmuladd_fpext_fmul_multi_use_fmuladd_lhs() #0 {
; GCN-FLUSH-LABEL: fast_sub_fmuladd_fpext_fmul_multi_use_fmuladd_lhs:
; GCN-FLUSH:       ; %bb.0:
; GCN-FLUSH-NEXT:    s_mov_b32 s3, 0xf000
; GCN-FLUSH-NEXT:    s_mov_b32 s2, -1
; GCN-FLUSH-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_ushort v3, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_ushort v4, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    v_cvt_f32_f16_e32 v3, v3
; GCN-FLUSH-NEXT:    v_cvt_f32_f16_e32 v4, v4
; GCN-FLUSH-NEXT:    v_mul_f32_e32 v3, v3, v4
; GCN-FLUSH-NEXT:    v_mac_f32_e32 v3, v0, v1
; GCN-FLUSH-NEXT:    v_sub_f32_e32 v0, v3, v2
; GCN-FLUSH-NEXT:    buffer_store_dword v3, off, s[0:3], 0
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    s_endpgm
;
; GCN-FASTFMA-LABEL: fast_sub_fmuladd_fpext_fmul_multi_use_fmuladd_lhs:
; GCN-FASTFMA:       ; %bb.0:
; GCN-FASTFMA-NEXT:    s_mov_b32 s3, 0xf000
; GCN-FASTFMA-NEXT:    s_mov_b32 s2, -1
; GCN-FASTFMA-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_ushort v3, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_ushort v4, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    v_cvt_f32_f16_e32 v3, v3
; GCN-FASTFMA-NEXT:    v_cvt_f32_f16_e32 v4, v4
; GCN-FASTFMA-NEXT:    v_mul_f32_e32 v3, v3, v4
; GCN-FASTFMA-NEXT:    v_fma_f32 v0, v0, v1, v3
; GCN-FASTFMA-NEXT:    v_sub_f32_e32 v1, v0, v2
; GCN-FASTFMA-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_store_dword v1, off, s[0:3], 0
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    s_endpgm
;
; GCN-SLOWFMA-LABEL: fast_sub_fmuladd_fpext_fmul_multi_use_fmuladd_lhs:
; GCN-SLOWFMA:       ; %bb.0:
; GCN-SLOWFMA-NEXT:    s_mov_b32 s3, 0xf000
; GCN-SLOWFMA-NEXT:    s_mov_b32 s2, -1
; GCN-SLOWFMA-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_ushort v3, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_ushort v4, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    v_cvt_f32_f16_e32 v3, v3
; GCN-SLOWFMA-NEXT:    v_cvt_f32_f16_e32 v4, v4
; GCN-SLOWFMA-NEXT:    v_mul_f32_e32 v0, v0, v1
; GCN-SLOWFMA-NEXT:    v_mul_f32_e32 v1, v3, v4
; GCN-SLOWFMA-NEXT:    v_add_f32_e32 v0, v0, v1
; GCN-SLOWFMA-NEXT:    v_sub_f32_e32 v1, v0, v2
; GCN-SLOWFMA-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_store_dword v1, off, s[0:3], 0
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    s_endpgm
  %x = load volatile float, ptr addrspace(1) undef
  %y = load volatile float, ptr addrspace(1) undef
  %z = load volatile float, ptr addrspace(1) undef
  %u = load volatile half, ptr addrspace(1) undef
  %v = load volatile half, ptr addrspace(1) undef
  %mul.u.v.half = fmul fast half %u, %v
  %mul.u.v = fpext half %mul.u.v.half to float
  %fma = call fast float @llvm.fmuladd.f32(float %x, float %y, float %mul.u.v)
  %add = fsub fast float %fma, %z
  store volatile float %fma, ptr addrspace(1) undef
  store volatile float %add, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @fast_sub_fmuladd_fpext_fmul_multi_use_fmuladd_rhs() #0 {
; GCN-FLUSH-LABEL: fast_sub_fmuladd_fpext_fmul_multi_use_fmuladd_rhs:
; GCN-FLUSH:       ; %bb.0:
; GCN-FLUSH-NEXT:    s_mov_b32 s3, 0xf000
; GCN-FLUSH-NEXT:    s_mov_b32 s2, -1
; GCN-FLUSH-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_ushort v3, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_load_ushort v4, off, s[0:3], 0 glc
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    v_cvt_f32_f16_e32 v3, v3
; GCN-FLUSH-NEXT:    v_cvt_f32_f16_e32 v4, v4
; GCN-FLUSH-NEXT:    v_mul_f32_e32 v3, v3, v4
; GCN-FLUSH-NEXT:    v_mac_f32_e32 v3, v0, v1
; GCN-FLUSH-NEXT:    v_sub_f32_e32 v0, v2, v3
; GCN-FLUSH-NEXT:    buffer_store_dword v3, off, s[0:3], 0
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-FLUSH-NEXT:    s_waitcnt vmcnt(0)
; GCN-FLUSH-NEXT:    s_endpgm
;
; GCN-FASTFMA-LABEL: fast_sub_fmuladd_fpext_fmul_multi_use_fmuladd_rhs:
; GCN-FASTFMA:       ; %bb.0:
; GCN-FASTFMA-NEXT:    s_mov_b32 s3, 0xf000
; GCN-FASTFMA-NEXT:    s_mov_b32 s2, -1
; GCN-FASTFMA-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_ushort v3, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_load_ushort v4, off, s[0:3], 0 glc
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    v_cvt_f32_f16_e32 v3, v3
; GCN-FASTFMA-NEXT:    v_cvt_f32_f16_e32 v4, v4
; GCN-FASTFMA-NEXT:    v_mul_f32_e32 v3, v3, v4
; GCN-FASTFMA-NEXT:    v_fma_f32 v0, v0, v1, v3
; GCN-FASTFMA-NEXT:    v_sub_f32_e32 v1, v2, v0
; GCN-FASTFMA-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    buffer_store_dword v1, off, s[0:3], 0
; GCN-FASTFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-FASTFMA-NEXT:    s_endpgm
;
; GCN-SLOWFMA-LABEL: fast_sub_fmuladd_fpext_fmul_multi_use_fmuladd_rhs:
; GCN-SLOWFMA:       ; %bb.0:
; GCN-SLOWFMA-NEXT:    s_mov_b32 s3, 0xf000
; GCN-SLOWFMA-NEXT:    s_mov_b32 s2, -1
; GCN-SLOWFMA-NEXT:    buffer_load_dword v0, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_ushort v3, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_load_ushort v4, off, s[0:3], 0 glc
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    v_cvt_f32_f16_e32 v3, v3
; GCN-SLOWFMA-NEXT:    v_cvt_f32_f16_e32 v4, v4
; GCN-SLOWFMA-NEXT:    v_mul_f32_e32 v0, v0, v1
; GCN-SLOWFMA-NEXT:    v_mul_f32_e32 v1, v3, v4
; GCN-SLOWFMA-NEXT:    v_add_f32_e32 v0, v0, v1
; GCN-SLOWFMA-NEXT:    v_sub_f32_e32 v1, v2, v0
; GCN-SLOWFMA-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    buffer_store_dword v1, off, s[0:3], 0
; GCN-SLOWFMA-NEXT:    s_waitcnt vmcnt(0)
; GCN-SLOWFMA-NEXT:    s_endpgm
  %x = load volatile float, ptr addrspace(1) undef
  %y = load volatile float, ptr addrspace(1) undef
  %z = load volatile float, ptr addrspace(1) undef
  %u = load volatile half, ptr addrspace(1) undef
  %v = load volatile half, ptr addrspace(1) undef
  %mul.u.v.half = fmul fast half %u, %v
  %mul.u.v = fpext half %mul.u.v.half to float
  %fma = call fast float @llvm.fmuladd.f32(float %x, float %y, float %mul.u.v)
  %add = fsub fast float %z, %fma
  store volatile float %fma, ptr addrspace(1) undef
  store volatile float %add, ptr addrspace(1) undef
  ret void
}

declare float @llvm.fma.f32(float, float, float) #1
declare float @llvm.fmuladd.f32(float, float, float) #1

attributes #0 = { nounwind }
attributes #1 = { nounwind readnone }
;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
; GCN: {{.*}}