; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN:  llc -amdgpu-scalarize-global-loads=false  -mtriple=amdgcn -mcpu=tahiti -fp-contract=on < %s | FileCheck -check-prefixes=SI,SI-STRICT %s
; RUN:  llc -amdgpu-scalarize-global-loads=false  -mtriple=amdgcn -mcpu=verde  -fp-contract=on < %s | FileCheck -check-prefixes=SI,SI-STRICT %s
; RUN:  llc -amdgpu-scalarize-global-loads=false  -mtriple=amdgcn -mcpu=tahiti -fp-contract=fast < %s | FileCheck -check-prefixes=SI,SI-CONTRACT %s
; RUN:  llc -amdgpu-scalarize-global-loads=false  -mtriple=amdgcn -mcpu=verde  -fp-contract=fast < %s | FileCheck -check-prefixes=SI,SI-CONTRACT %s
; RUN:  llc -amdgpu-scalarize-global-loads=false  -mtriple=amdgcn -mcpu=tonga -mattr=-flat-for-global  -fp-contract=on < %s | FileCheck -check-prefixes=VI,VI-STRICT %s
; RUN:  llc -amdgpu-scalarize-global-loads=false  -mtriple=amdgcn -mcpu=tonga -mattr=-flat-for-global  -fp-contract=fast < %s | FileCheck -check-prefixes=VI,VI-CONTRACT %s

define amdgpu_kernel void @fmuladd_f64(ptr addrspace(1) %out, ptr addrspace(1) %in1, ptr addrspace(1) %in2, ptr addrspace(1) %in3) #0 {
; SI-LABEL: fmuladd_f64:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx8 s[0:7], s[4:5], 0x9
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s2
; SI-NEXT:    s_mov_b32 s13, s3
; SI-NEXT:    s_mov_b32 s16, s4
; SI-NEXT:    s_mov_b32 s17, s5
; SI-NEXT:    s_mov_b32 s18, s10
; SI-NEXT:    s_mov_b32 s19, s11
; SI-NEXT:    s_mov_b32 s4, s6
; SI-NEXT:    s_mov_b32 s5, s7
; SI-NEXT:    s_mov_b32 s6, s10
; SI-NEXT:    s_mov_b32 s7, s11
; SI-NEXT:    buffer_load_dwordx2 v[0:1], off, s[12:15], 0
; SI-NEXT:    buffer_load_dwordx2 v[2:3], off, s[16:19], 0
; SI-NEXT:    buffer_load_dwordx2 v[4:5], off, s[4:7], 0
; SI-NEXT:    s_mov_b32 s8, s0
; SI-NEXT:    s_mov_b32 s9, s1
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_fma_f64 v[0:1], v[0:1], v[2:3], v[4:5]
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: fmuladd_f64:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx8 s[0:7], s[4:5], 0x24
; VI-NEXT:    s_mov_b32 s11, 0xf000
; VI-NEXT:    s_mov_b32 s10, -1
; VI-NEXT:    s_mov_b32 s14, s10
; VI-NEXT:    s_mov_b32 s15, s11
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s12, s2
; VI-NEXT:    s_mov_b32 s13, s3
; VI-NEXT:    s_mov_b32 s16, s4
; VI-NEXT:    s_mov_b32 s17, s5
; VI-NEXT:    s_mov_b32 s18, s10
; VI-NEXT:    s_mov_b32 s19, s11
; VI-NEXT:    s_mov_b32 s4, s6
; VI-NEXT:    s_mov_b32 s5, s7
; VI-NEXT:    s_mov_b32 s6, s10
; VI-NEXT:    s_mov_b32 s7, s11
; VI-NEXT:    buffer_load_dwordx2 v[0:1], off, s[12:15], 0
; VI-NEXT:    buffer_load_dwordx2 v[2:3], off, s[16:19], 0
; VI-NEXT:    buffer_load_dwordx2 v[4:5], off, s[4:7], 0
; VI-NEXT:    s_mov_b32 s8, s0
; VI-NEXT:    s_mov_b32 s9, s1
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_fma_f64 v[0:1], v[0:1], v[2:3], v[4:5]
; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
; VI-NEXT:    s_endpgm
  %r0 = load double, ptr addrspace(1) %in1
  %r1 = load double, ptr addrspace(1) %in2
  %r2 = load double, ptr addrspace(1) %in3
  %r3 = tail call double @llvm.fmuladd.f64(double %r0, double %r1, double %r2)
  store double %r3, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @fmul_fadd_f64(ptr addrspace(1) %out, ptr addrspace(1) %in1, ptr addrspace(1) %in2, ptr addrspace(1) %in3) #0 {
; SI-STRICT-LABEL: fmul_fadd_f64:
; SI-STRICT:       ; %bb.0:
; SI-STRICT-NEXT:    s_load_dwordx8 s[0:7], s[4:5], 0x9
; SI-STRICT-NEXT:    s_mov_b32 s11, 0xf000
; SI-STRICT-NEXT:    s_mov_b32 s10, -1
; SI-STRICT-NEXT:    s_mov_b32 s14, s10
; SI-STRICT-NEXT:    s_mov_b32 s15, s11
; SI-STRICT-NEXT:    s_waitcnt lgkmcnt(0)
; SI-STRICT-NEXT:    s_mov_b32 s12, s2
; SI-STRICT-NEXT:    s_mov_b32 s13, s3
; SI-STRICT-NEXT:    s_mov_b32 s16, s4
; SI-STRICT-NEXT:    s_mov_b32 s17, s5
; SI-STRICT-NEXT:    s_mov_b32 s18, s10
; SI-STRICT-NEXT:    s_mov_b32 s19, s11
; SI-STRICT-NEXT:    buffer_load_dwordx2 v[0:1], off, s[12:15], 0
; SI-STRICT-NEXT:    buffer_load_dwordx2 v[2:3], off, s[16:19], 0
; SI-STRICT-NEXT:    s_mov_b32 s4, s6
; SI-STRICT-NEXT:    s_mov_b32 s5, s7
; SI-STRICT-NEXT:    s_mov_b32 s6, s10
; SI-STRICT-NEXT:    s_mov_b32 s7, s11
; SI-STRICT-NEXT:    buffer_load_dwordx2 v[4:5], off, s[4:7], 0
; SI-STRICT-NEXT:    s_mov_b32 s8, s0
; SI-STRICT-NEXT:    s_mov_b32 s9, s1
; SI-STRICT-NEXT:    s_waitcnt vmcnt(1)
; SI-STRICT-NEXT:    v_mul_f64 v[0:1], v[0:1], v[2:3]
; SI-STRICT-NEXT:    s_waitcnt vmcnt(0)
; SI-STRICT-NEXT:    v_add_f64 v[0:1], v[0:1], v[4:5]
; SI-STRICT-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
; SI-STRICT-NEXT:    s_endpgm
;
; SI-CONTRACT-LABEL: fmul_fadd_f64:
; SI-CONTRACT:       ; %bb.0:
; SI-CONTRACT-NEXT:    s_load_dwordx8 s[0:7], s[4:5], 0x9
; SI-CONTRACT-NEXT:    s_mov_b32 s11, 0xf000
; SI-CONTRACT-NEXT:    s_mov_b32 s10, -1
; SI-CONTRACT-NEXT:    s_mov_b32 s14, s10
; SI-CONTRACT-NEXT:    s_mov_b32 s15, s11
; SI-CONTRACT-NEXT:    s_waitcnt lgkmcnt(0)
; SI-CONTRACT-NEXT:    s_mov_b32 s12, s2
; SI-CONTRACT-NEXT:    s_mov_b32 s13, s3
; SI-CONTRACT-NEXT:    s_mov_b32 s16, s4
; SI-CONTRACT-NEXT:    s_mov_b32 s17, s5
; SI-CONTRACT-NEXT:    s_mov_b32 s18, s10
; SI-CONTRACT-NEXT:    s_mov_b32 s19, s11
; SI-CONTRACT-NEXT:    s_mov_b32 s4, s6
; SI-CONTRACT-NEXT:    s_mov_b32 s5, s7
; SI-CONTRACT-NEXT:    s_mov_b32 s6, s10
; SI-CONTRACT-NEXT:    s_mov_b32 s7, s11
; SI-CONTRACT-NEXT:    buffer_load_dwordx2 v[0:1], off, s[12:15], 0
; SI-CONTRACT-NEXT:    buffer_load_dwordx2 v[2:3], off, s[16:19], 0
; SI-CONTRACT-NEXT:    buffer_load_dwordx2 v[4:5], off, s[4:7], 0
; SI-CONTRACT-NEXT:    s_mov_b32 s8, s0
; SI-CONTRACT-NEXT:    s_mov_b32 s9, s1
; SI-CONTRACT-NEXT:    s_waitcnt vmcnt(0)
; SI-CONTRACT-NEXT:    v_fma_f64 v[0:1], v[0:1], v[2:3], v[4:5]
; SI-CONTRACT-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
; SI-CONTRACT-NEXT:    s_endpgm
;
; VI-STRICT-LABEL: fmul_fadd_f64:
; VI-STRICT:       ; %bb.0:
; VI-STRICT-NEXT:    s_load_dwordx8 s[0:7], s[4:5], 0x24
; VI-STRICT-NEXT:    s_mov_b32 s11, 0xf000
; VI-STRICT-NEXT:    s_mov_b32 s10, -1
; VI-STRICT-NEXT:    s_mov_b32 s14, s10
; VI-STRICT-NEXT:    s_mov_b32 s15, s11
; VI-STRICT-NEXT:    s_waitcnt lgkmcnt(0)
; VI-STRICT-NEXT:    s_mov_b32 s12, s2
; VI-STRICT-NEXT:    s_mov_b32 s13, s3
; VI-STRICT-NEXT:    s_mov_b32 s16, s4
; VI-STRICT-NEXT:    s_mov_b32 s17, s5
; VI-STRICT-NEXT:    s_mov_b32 s18, s10
; VI-STRICT-NEXT:    s_mov_b32 s19, s11
; VI-STRICT-NEXT:    buffer_load_dwordx2 v[0:1], off, s[12:15], 0
; VI-STRICT-NEXT:    buffer_load_dwordx2 v[2:3], off, s[16:19], 0
; VI-STRICT-NEXT:    s_mov_b32 s4, s6
; VI-STRICT-NEXT:    s_mov_b32 s5, s7
; VI-STRICT-NEXT:    s_mov_b32 s6, s10
; VI-STRICT-NEXT:    s_mov_b32 s7, s11
; VI-STRICT-NEXT:    buffer_load_dwordx2 v[4:5], off, s[4:7], 0
; VI-STRICT-NEXT:    s_mov_b32 s8, s0
; VI-STRICT-NEXT:    s_mov_b32 s9, s1
; VI-STRICT-NEXT:    s_waitcnt vmcnt(1)
; VI-STRICT-NEXT:    v_mul_f64 v[0:1], v[0:1], v[2:3]
; VI-STRICT-NEXT:    s_waitcnt vmcnt(0)
; VI-STRICT-NEXT:    v_add_f64 v[0:1], v[0:1], v[4:5]
; VI-STRICT-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
; VI-STRICT-NEXT:    s_endpgm
;
; VI-CONTRACT-LABEL: fmul_fadd_f64:
; VI-CONTRACT:       ; %bb.0:
; VI-CONTRACT-NEXT:    s_load_dwordx8 s[0:7], s[4:5], 0x24
; VI-CONTRACT-NEXT:    s_mov_b32 s11, 0xf000
; VI-CONTRACT-NEXT:    s_mov_b32 s10, -1
; VI-CONTRACT-NEXT:    s_mov_b32 s14, s10
; VI-CONTRACT-NEXT:    s_mov_b32 s15, s11
; VI-CONTRACT-NEXT:    s_waitcnt lgkmcnt(0)
; VI-CONTRACT-NEXT:    s_mov_b32 s12, s2
; VI-CONTRACT-NEXT:    s_mov_b32 s13, s3
; VI-CONTRACT-NEXT:    s_mov_b32 s16, s4
; VI-CONTRACT-NEXT:    s_mov_b32 s17, s5
; VI-CONTRACT-NEXT:    s_mov_b32 s18, s10
; VI-CONTRACT-NEXT:    s_mov_b32 s19, s11
; VI-CONTRACT-NEXT:    s_mov_b32 s4, s6
; VI-CONTRACT-NEXT:    s_mov_b32 s5, s7
; VI-CONTRACT-NEXT:    s_mov_b32 s6, s10
; VI-CONTRACT-NEXT:    s_mov_b32 s7, s11
; VI-CONTRACT-NEXT:    buffer_load_dwordx2 v[0:1], off, s[12:15], 0
; VI-CONTRACT-NEXT:    buffer_load_dwordx2 v[2:3], off, s[16:19], 0
; VI-CONTRACT-NEXT:    buffer_load_dwordx2 v[4:5], off, s[4:7], 0
; VI-CONTRACT-NEXT:    s_mov_b32 s8, s0
; VI-CONTRACT-NEXT:    s_mov_b32 s9, s1
; VI-CONTRACT-NEXT:    s_waitcnt vmcnt(0)
; VI-CONTRACT-NEXT:    v_fma_f64 v[0:1], v[0:1], v[2:3], v[4:5]
; VI-CONTRACT-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
; VI-CONTRACT-NEXT:    s_endpgm
  %r0 = load double, ptr addrspace(1) %in1
  %r1 = load double, ptr addrspace(1) %in2
  %r2 = load double, ptr addrspace(1) %in3
  %tmp = fmul double %r0, %r1
  %r3 = fadd double %tmp, %r2
  store double %r3, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @fmul_fadd_contract_f64(ptr addrspace(1) %out, ptr addrspace(1) %in1, ptr addrspace(1) %in2, ptr addrspace(1) %in3) #0 {
; SI-LABEL: fmul_fadd_contract_f64:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx8 s[0:7], s[4:5], 0x9
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s2
; SI-NEXT:    s_mov_b32 s13, s3
; SI-NEXT:    s_mov_b32 s16, s4
; SI-NEXT:    s_mov_b32 s17, s5
; SI-NEXT:    s_mov_b32 s18, s10
; SI-NEXT:    s_mov_b32 s19, s11
; SI-NEXT:    s_mov_b32 s4, s6
; SI-NEXT:    s_mov_b32 s5, s7
; SI-NEXT:    s_mov_b32 s6, s10
; SI-NEXT:    s_mov_b32 s7, s11
; SI-NEXT:    buffer_load_dwordx2 v[0:1], off, s[12:15], 0
; SI-NEXT:    buffer_load_dwordx2 v[2:3], off, s[16:19], 0
; SI-NEXT:    buffer_load_dwordx2 v[4:5], off, s[4:7], 0
; SI-NEXT:    s_mov_b32 s8, s0
; SI-NEXT:    s_mov_b32 s9, s1
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_fma_f64 v[0:1], v[0:1], v[2:3], v[4:5]
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: fmul_fadd_contract_f64:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx8 s[0:7], s[4:5], 0x24
; VI-NEXT:    s_mov_b32 s11, 0xf000
; VI-NEXT:    s_mov_b32 s10, -1
; VI-NEXT:    s_mov_b32 s14, s10
; VI-NEXT:    s_mov_b32 s15, s11
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s12, s2
; VI-NEXT:    s_mov_b32 s13, s3
; VI-NEXT:    s_mov_b32 s16, s4
; VI-NEXT:    s_mov_b32 s17, s5
; VI-NEXT:    s_mov_b32 s18, s10
; VI-NEXT:    s_mov_b32 s19, s11
; VI-NEXT:    s_mov_b32 s4, s6
; VI-NEXT:    s_mov_b32 s5, s7
; VI-NEXT:    s_mov_b32 s6, s10
; VI-NEXT:    s_mov_b32 s7, s11
; VI-NEXT:    buffer_load_dwordx2 v[0:1], off, s[12:15], 0
; VI-NEXT:    buffer_load_dwordx2 v[2:3], off, s[16:19], 0
; VI-NEXT:    buffer_load_dwordx2 v[4:5], off, s[4:7], 0
; VI-NEXT:    s_mov_b32 s8, s0
; VI-NEXT:    s_mov_b32 s9, s1
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_fma_f64 v[0:1], v[0:1], v[2:3], v[4:5]
; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
; VI-NEXT:    s_endpgm
  %r0 = load double, ptr addrspace(1) %in1
  %r1 = load double, ptr addrspace(1) %in2
  %r2 = load double, ptr addrspace(1) %in3
  %tmp = fmul contract double %r0, %r1
  %r3 = fadd contract double %tmp, %r2
  store double %r3, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @fadd_a_a_b_f64(ptr addrspace(1) %out, ptr addrspace(1) %in1, ptr addrspace(1) %in2) #0 {
; SI-STRICT-LABEL: fadd_a_a_b_f64:
; SI-STRICT:       ; %bb.0:
; SI-STRICT-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x9
; SI-STRICT-NEXT:    s_mov_b32 s3, 0xf000
; SI-STRICT-NEXT:    s_mov_b32 s2, 0
; SI-STRICT-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; SI-STRICT-NEXT:    v_mov_b32_e32 v1, 0
; SI-STRICT-NEXT:    s_waitcnt lgkmcnt(0)
; SI-STRICT-NEXT:    buffer_load_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64 glc
; SI-STRICT-NEXT:    s_waitcnt vmcnt(0)
; SI-STRICT-NEXT:    buffer_load_dwordx2 v[4:5], v[0:1], s[0:3], 0 addr64 offset:8 glc
; SI-STRICT-NEXT:    s_waitcnt vmcnt(0)
; SI-STRICT-NEXT:    v_add_f64 v[2:3], v[2:3], v[2:3]
; SI-STRICT-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
; SI-STRICT-NEXT:    buffer_store_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64
; SI-STRICT-NEXT:    s_endpgm
;
; SI-CONTRACT-LABEL: fadd_a_a_b_f64:
; SI-CONTRACT:       ; %bb.0:
; SI-CONTRACT-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x9
; SI-CONTRACT-NEXT:    s_mov_b32 s3, 0xf000
; SI-CONTRACT-NEXT:    s_mov_b32 s2, 0
; SI-CONTRACT-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; SI-CONTRACT-NEXT:    v_mov_b32_e32 v1, 0
; SI-CONTRACT-NEXT:    s_waitcnt lgkmcnt(0)
; SI-CONTRACT-NEXT:    buffer_load_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64 glc
; SI-CONTRACT-NEXT:    s_waitcnt vmcnt(0)
; SI-CONTRACT-NEXT:    buffer_load_dwordx2 v[4:5], v[0:1], s[0:3], 0 addr64 offset:8 glc
; SI-CONTRACT-NEXT:    s_waitcnt vmcnt(0)
; SI-CONTRACT-NEXT:    v_fma_f64 v[2:3], v[2:3], 2.0, v[4:5]
; SI-CONTRACT-NEXT:    buffer_store_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64
; SI-CONTRACT-NEXT:    s_endpgm
;
; VI-STRICT-LABEL: fadd_a_a_b_f64:
; VI-STRICT:       ; %bb.0:
; VI-STRICT-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; VI-STRICT-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; VI-STRICT-NEXT:    s_waitcnt lgkmcnt(0)
; VI-STRICT-NEXT:    v_mov_b32_e32 v1, s1
; VI-STRICT-NEXT:    v_add_u32_e32 v0, vcc, s0, v0
; VI-STRICT-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-STRICT-NEXT:    flat_load_dwordx2 v[2:3], v[0:1] glc
; VI-STRICT-NEXT:    s_waitcnt vmcnt(0)
; VI-STRICT-NEXT:    v_add_u32_e32 v4, vcc, 8, v0
; VI-STRICT-NEXT:    v_addc_u32_e32 v5, vcc, 0, v1, vcc
; VI-STRICT-NEXT:    flat_load_dwordx2 v[4:5], v[4:5] glc
; VI-STRICT-NEXT:    s_waitcnt vmcnt(0)
; VI-STRICT-NEXT:    v_add_f64 v[2:3], v[2:3], v[2:3]
; VI-STRICT-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
; VI-STRICT-NEXT:    flat_store_dwordx2 v[0:1], v[2:3]
; VI-STRICT-NEXT:    s_endpgm
;
; VI-CONTRACT-LABEL: fadd_a_a_b_f64:
; VI-CONTRACT:       ; %bb.0:
; VI-CONTRACT-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; VI-CONTRACT-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; VI-CONTRACT-NEXT:    s_waitcnt lgkmcnt(0)
; VI-CONTRACT-NEXT:    v_mov_b32_e32 v1, s1
; VI-CONTRACT-NEXT:    v_add_u32_e32 v0, vcc, s0, v0
; VI-CONTRACT-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-CONTRACT-NEXT:    v_add_u32_e32 v2, vcc, 8, v0
; VI-CONTRACT-NEXT:    v_addc_u32_e32 v3, vcc, 0, v1, vcc
; VI-CONTRACT-NEXT:    flat_load_dwordx2 v[4:5], v[0:1] glc
; VI-CONTRACT-NEXT:    s_waitcnt vmcnt(0)
; VI-CONTRACT-NEXT:    flat_load_dwordx2 v[2:3], v[2:3] glc
; VI-CONTRACT-NEXT:    s_waitcnt vmcnt(0)
; VI-CONTRACT-NEXT:    v_fma_f64 v[2:3], v[4:5], 2.0, v[2:3]
; VI-CONTRACT-NEXT:    flat_store_dwordx2 v[0:1], v[2:3]
; VI-CONTRACT-NEXT:    s_endpgm
  %tid = call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
  %gep.0 = getelementptr double, ptr addrspace(1) %out, i32 %tid
  %gep.1 = getelementptr double, ptr addrspace(1) %gep.0, i32 1
  %gep.out = getelementptr double, ptr addrspace(1) %out, i32 %tid

  %r0 = load volatile double, ptr addrspace(1) %gep.0
  %r1 = load volatile double, ptr addrspace(1) %gep.1

  %add.0 = fadd double %r0, %r0
  %add.1 = fadd double %add.0, %r1
  store double %add.1, ptr addrspace(1) %gep.out
  ret void
}

define amdgpu_kernel void @fadd_b_a_a_f64(ptr addrspace(1) %out, ptr addrspace(1) %in1, ptr addrspace(1) %in2) #0 {
; SI-STRICT-LABEL: fadd_b_a_a_f64:
; SI-STRICT:       ; %bb.0:
; SI-STRICT-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x9
; SI-STRICT-NEXT:    s_mov_b32 s3, 0xf000
; SI-STRICT-NEXT:    s_mov_b32 s2, 0
; SI-STRICT-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; SI-STRICT-NEXT:    v_mov_b32_e32 v1, 0
; SI-STRICT-NEXT:    s_waitcnt lgkmcnt(0)
; SI-STRICT-NEXT:    buffer_load_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64 glc
; SI-STRICT-NEXT:    s_waitcnt vmcnt(0)
; SI-STRICT-NEXT:    buffer_load_dwordx2 v[4:5], v[0:1], s[0:3], 0 addr64 offset:8 glc
; SI-STRICT-NEXT:    s_waitcnt vmcnt(0)
; SI-STRICT-NEXT:    v_add_f64 v[2:3], v[2:3], v[2:3]
; SI-STRICT-NEXT:    v_add_f64 v[2:3], v[4:5], v[2:3]
; SI-STRICT-NEXT:    buffer_store_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64
; SI-STRICT-NEXT:    s_endpgm
;
; SI-CONTRACT-LABEL: fadd_b_a_a_f64:
; SI-CONTRACT:       ; %bb.0:
; SI-CONTRACT-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x9
; SI-CONTRACT-NEXT:    s_mov_b32 s3, 0xf000
; SI-CONTRACT-NEXT:    s_mov_b32 s2, 0
; SI-CONTRACT-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; SI-CONTRACT-NEXT:    v_mov_b32_e32 v1, 0
; SI-CONTRACT-NEXT:    s_waitcnt lgkmcnt(0)
; SI-CONTRACT-NEXT:    buffer_load_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64 glc
; SI-CONTRACT-NEXT:    s_waitcnt vmcnt(0)
; SI-CONTRACT-NEXT:    buffer_load_dwordx2 v[4:5], v[0:1], s[0:3], 0 addr64 offset:8 glc
; SI-CONTRACT-NEXT:    s_waitcnt vmcnt(0)
; SI-CONTRACT-NEXT:    v_fma_f64 v[2:3], v[2:3], 2.0, v[4:5]
; SI-CONTRACT-NEXT:    buffer_store_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64
; SI-CONTRACT-NEXT:    s_endpgm
;
; VI-STRICT-LABEL: fadd_b_a_a_f64:
; VI-STRICT:       ; %bb.0:
; VI-STRICT-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; VI-STRICT-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; VI-STRICT-NEXT:    s_waitcnt lgkmcnt(0)
; VI-STRICT-NEXT:    v_mov_b32_e32 v1, s1
; VI-STRICT-NEXT:    v_add_u32_e32 v0, vcc, s0, v0
; VI-STRICT-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-STRICT-NEXT:    flat_load_dwordx2 v[2:3], v[0:1] glc
; VI-STRICT-NEXT:    s_waitcnt vmcnt(0)
; VI-STRICT-NEXT:    v_add_u32_e32 v4, vcc, 8, v0
; VI-STRICT-NEXT:    v_addc_u32_e32 v5, vcc, 0, v1, vcc
; VI-STRICT-NEXT:    flat_load_dwordx2 v[4:5], v[4:5] glc
; VI-STRICT-NEXT:    s_waitcnt vmcnt(0)
; VI-STRICT-NEXT:    v_add_f64 v[2:3], v[2:3], v[2:3]
; VI-STRICT-NEXT:    v_add_f64 v[2:3], v[4:5], v[2:3]
; VI-STRICT-NEXT:    flat_store_dwordx2 v[0:1], v[2:3]
; VI-STRICT-NEXT:    s_endpgm
;
; VI-CONTRACT-LABEL: fadd_b_a_a_f64:
; VI-CONTRACT:       ; %bb.0:
; VI-CONTRACT-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; VI-CONTRACT-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; VI-CONTRACT-NEXT:    s_waitcnt lgkmcnt(0)
; VI-CONTRACT-NEXT:    v_mov_b32_e32 v1, s1
; VI-CONTRACT-NEXT:    v_add_u32_e32 v0, vcc, s0, v0
; VI-CONTRACT-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-CONTRACT-NEXT:    v_add_u32_e32 v2, vcc, 8, v0
; VI-CONTRACT-NEXT:    v_addc_u32_e32 v3, vcc, 0, v1, vcc
; VI-CONTRACT-NEXT:    flat_load_dwordx2 v[4:5], v[0:1] glc
; VI-CONTRACT-NEXT:    s_waitcnt vmcnt(0)
; VI-CONTRACT-NEXT:    flat_load_dwordx2 v[2:3], v[2:3] glc
; VI-CONTRACT-NEXT:    s_waitcnt vmcnt(0)
; VI-CONTRACT-NEXT:    v_fma_f64 v[2:3], v[4:5], 2.0, v[2:3]
; VI-CONTRACT-NEXT:    flat_store_dwordx2 v[0:1], v[2:3]
; VI-CONTRACT-NEXT:    s_endpgm
  %tid = call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
  %gep.0 = getelementptr double, ptr addrspace(1) %out, i32 %tid
  %gep.1 = getelementptr double, ptr addrspace(1) %gep.0, i32 1
  %gep.out = getelementptr double, ptr addrspace(1) %out, i32 %tid

  %r0 = load volatile double, ptr addrspace(1) %gep.0
  %r1 = load volatile double, ptr addrspace(1) %gep.1

  %add.0 = fadd double %r0, %r0
  %add.1 = fadd double %r1, %add.0
  store double %add.1, ptr addrspace(1) %gep.out
  ret void
}

define amdgpu_kernel void @mad_sub_f64(ptr addrspace(1) noalias nocapture %out, ptr addrspace(1) noalias nocapture readonly %ptr) #1 {
; SI-STRICT-LABEL: mad_sub_f64:
; SI-STRICT:       ; %bb.0:
; SI-STRICT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
; SI-STRICT-NEXT:    s_mov_b32 s7, 0xf000
; SI-STRICT-NEXT:    s_mov_b32 s6, 0
; SI-STRICT-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; SI-STRICT-NEXT:    v_mov_b32_e32 v1, 0
; SI-STRICT-NEXT:    s_waitcnt lgkmcnt(0)
; SI-STRICT-NEXT:    s_mov_b64 s[4:5], s[2:3]
; SI-STRICT-NEXT:    buffer_load_dwordx2 v[2:3], v[0:1], s[4:7], 0 addr64 glc
; SI-STRICT-NEXT:    s_waitcnt vmcnt(0)
; SI-STRICT-NEXT:    buffer_load_dwordx2 v[4:5], v[0:1], s[4:7], 0 addr64 offset:8 glc
; SI-STRICT-NEXT:    s_waitcnt vmcnt(0)
; SI-STRICT-NEXT:    buffer_load_dwordx2 v[6:7], v[0:1], s[4:7], 0 addr64 offset:16 glc
; SI-STRICT-NEXT:    s_waitcnt vmcnt(0)
; SI-STRICT-NEXT:    s_mov_b64 s[2:3], s[6:7]
; SI-STRICT-NEXT:    v_mul_f64 v[2:3], v[2:3], v[4:5]
; SI-STRICT-NEXT:    v_add_f64 v[2:3], v[2:3], -v[6:7]
; SI-STRICT-NEXT:    buffer_store_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64
; SI-STRICT-NEXT:    s_endpgm
;
; SI-CONTRACT-LABEL: mad_sub_f64:
; SI-CONTRACT:       ; %bb.0:
; SI-CONTRACT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
; SI-CONTRACT-NEXT:    s_mov_b32 s7, 0xf000
; SI-CONTRACT-NEXT:    s_mov_b32 s6, 0
; SI-CONTRACT-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; SI-CONTRACT-NEXT:    v_mov_b32_e32 v1, 0
; SI-CONTRACT-NEXT:    s_waitcnt lgkmcnt(0)
; SI-CONTRACT-NEXT:    s_mov_b64 s[4:5], s[2:3]
; SI-CONTRACT-NEXT:    buffer_load_dwordx2 v[2:3], v[0:1], s[4:7], 0 addr64 glc
; SI-CONTRACT-NEXT:    s_waitcnt vmcnt(0)
; SI-CONTRACT-NEXT:    buffer_load_dwordx2 v[4:5], v[0:1], s[4:7], 0 addr64 offset:8 glc
; SI-CONTRACT-NEXT:    s_waitcnt vmcnt(0)
; SI-CONTRACT-NEXT:    buffer_load_dwordx2 v[6:7], v[0:1], s[4:7], 0 addr64 offset:16 glc
; SI-CONTRACT-NEXT:    s_waitcnt vmcnt(0)
; SI-CONTRACT-NEXT:    s_mov_b64 s[2:3], s[6:7]
; SI-CONTRACT-NEXT:    v_fma_f64 v[2:3], v[2:3], v[4:5], -v[6:7]
; SI-CONTRACT-NEXT:    buffer_store_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64
; SI-CONTRACT-NEXT:    s_endpgm
;
; VI-STRICT-LABEL: mad_sub_f64:
; VI-STRICT:       ; %bb.0:
; VI-STRICT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; VI-STRICT-NEXT:    v_lshlrev_b32_e32 v6, 3, v0
; VI-STRICT-NEXT:    s_waitcnt lgkmcnt(0)
; VI-STRICT-NEXT:    v_mov_b32_e32 v1, s3
; VI-STRICT-NEXT:    v_add_u32_e32 v0, vcc, s2, v6
; VI-STRICT-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-STRICT-NEXT:    v_add_u32_e32 v2, vcc, 8, v0
; VI-STRICT-NEXT:    v_addc_u32_e32 v3, vcc, 0, v1, vcc
; VI-STRICT-NEXT:    flat_load_dwordx2 v[4:5], v[0:1] glc
; VI-STRICT-NEXT:    s_waitcnt vmcnt(0)
; VI-STRICT-NEXT:    flat_load_dwordx2 v[2:3], v[2:3] glc
; VI-STRICT-NEXT:    s_waitcnt vmcnt(0)
; VI-STRICT-NEXT:    v_add_u32_e32 v0, vcc, 16, v0
; VI-STRICT-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-STRICT-NEXT:    flat_load_dwordx2 v[0:1], v[0:1] glc
; VI-STRICT-NEXT:    s_waitcnt vmcnt(0)
; VI-STRICT-NEXT:    v_mul_f64 v[2:3], v[4:5], v[2:3]
; VI-STRICT-NEXT:    v_add_f64 v[0:1], v[2:3], -v[0:1]
; VI-STRICT-NEXT:    v_mov_b32_e32 v3, s1
; VI-STRICT-NEXT:    v_add_u32_e32 v2, vcc, s0, v6
; VI-STRICT-NEXT:    v_addc_u32_e32 v3, vcc, 0, v3, vcc
; VI-STRICT-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
; VI-STRICT-NEXT:    s_endpgm
;
; VI-CONTRACT-LABEL: mad_sub_f64:
; VI-CONTRACT:       ; %bb.0:
; VI-CONTRACT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; VI-CONTRACT-NEXT:    v_lshlrev_b32_e32 v6, 3, v0
; VI-CONTRACT-NEXT:    s_waitcnt lgkmcnt(0)
; VI-CONTRACT-NEXT:    v_mov_b32_e32 v1, s3
; VI-CONTRACT-NEXT:    v_add_u32_e32 v0, vcc, s2, v6
; VI-CONTRACT-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-CONTRACT-NEXT:    v_add_u32_e32 v2, vcc, 8, v0
; VI-CONTRACT-NEXT:    v_addc_u32_e32 v3, vcc, 0, v1, vcc
; VI-CONTRACT-NEXT:    v_add_u32_e32 v4, vcc, 16, v0
; VI-CONTRACT-NEXT:    v_addc_u32_e32 v5, vcc, 0, v1, vcc
; VI-CONTRACT-NEXT:    flat_load_dwordx2 v[0:1], v[0:1] glc
; VI-CONTRACT-NEXT:    s_waitcnt vmcnt(0)
; VI-CONTRACT-NEXT:    flat_load_dwordx2 v[2:3], v[2:3] glc
; VI-CONTRACT-NEXT:    s_waitcnt vmcnt(0)
; VI-CONTRACT-NEXT:    flat_load_dwordx2 v[4:5], v[4:5] glc
; VI-CONTRACT-NEXT:    s_waitcnt vmcnt(0)
; VI-CONTRACT-NEXT:    v_fma_f64 v[0:1], v[0:1], v[2:3], -v[4:5]
; VI-CONTRACT-NEXT:    v_mov_b32_e32 v3, s1
; VI-CONTRACT-NEXT:    v_add_u32_e32 v2, vcc, s0, v6
; VI-CONTRACT-NEXT:    v_addc_u32_e32 v3, vcc, 0, v3, vcc
; VI-CONTRACT-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
; VI-CONTRACT-NEXT:    s_endpgm
  %tid = tail call i32 @llvm.amdgcn.workitem.id.x() #0
  %tid.ext = sext i32 %tid to i64
  %gep0 = getelementptr double, ptr addrspace(1) %ptr, i64 %tid.ext
  %add1 = add i64 %tid.ext, 1
  %gep1 = getelementptr double, ptr addrspace(1) %ptr, i64 %add1
  %add2 = add i64 %tid.ext, 2
  %gep2 = getelementptr double, ptr addrspace(1) %ptr, i64 %add2
  %outgep = getelementptr double, ptr addrspace(1) %out, i64 %tid.ext
  %a = load volatile double, ptr addrspace(1) %gep0, align 8
  %b = load volatile double, ptr addrspace(1) %gep1, align 8
  %c = load volatile double, ptr addrspace(1) %gep2, align 8
  %mul = fmul double %a, %b
  %sub = fsub double %mul, %c
  store double %sub, ptr addrspace(1) %outgep, align 8
  ret void
}

define amdgpu_kernel void @fadd_a_a_b_f64_fast_add0(ptr addrspace(1) %out, ptr addrspace(1) %in1, ptr addrspace(1) %in2) #0 {
; SI-STRICT-LABEL: fadd_a_a_b_f64_fast_add0:
; SI-STRICT:       ; %bb.0:
; SI-STRICT-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x9
; SI-STRICT-NEXT:    s_mov_b32 s3, 0xf000
; SI-STRICT-NEXT:    s_mov_b32 s2, 0
; SI-STRICT-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; SI-STRICT-NEXT:    v_mov_b32_e32 v1, 0
; SI-STRICT-NEXT:    s_waitcnt lgkmcnt(0)
; SI-STRICT-NEXT:    buffer_load_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64 glc
; SI-STRICT-NEXT:    s_waitcnt vmcnt(0)
; SI-STRICT-NEXT:    buffer_load_dwordx2 v[4:5], v[0:1], s[0:3], 0 addr64 offset:8 glc
; SI-STRICT-NEXT:    s_waitcnt vmcnt(0)
; SI-STRICT-NEXT:    v_add_f64 v[2:3], v[2:3], v[2:3]
; SI-STRICT-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
; SI-STRICT-NEXT:    buffer_store_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64
; SI-STRICT-NEXT:    s_endpgm
;
; SI-CONTRACT-LABEL: fadd_a_a_b_f64_fast_add0:
; SI-CONTRACT:       ; %bb.0:
; SI-CONTRACT-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x9
; SI-CONTRACT-NEXT:    s_mov_b32 s3, 0xf000
; SI-CONTRACT-NEXT:    s_mov_b32 s2, 0
; SI-CONTRACT-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; SI-CONTRACT-NEXT:    v_mov_b32_e32 v1, 0
; SI-CONTRACT-NEXT:    s_waitcnt lgkmcnt(0)
; SI-CONTRACT-NEXT:    buffer_load_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64 glc
; SI-CONTRACT-NEXT:    s_waitcnt vmcnt(0)
; SI-CONTRACT-NEXT:    buffer_load_dwordx2 v[4:5], v[0:1], s[0:3], 0 addr64 offset:8 glc
; SI-CONTRACT-NEXT:    s_waitcnt vmcnt(0)
; SI-CONTRACT-NEXT:    v_fma_f64 v[2:3], v[2:3], 2.0, v[4:5]
; SI-CONTRACT-NEXT:    buffer_store_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64
; SI-CONTRACT-NEXT:    s_endpgm
;
; VI-STRICT-LABEL: fadd_a_a_b_f64_fast_add0:
; VI-STRICT:       ; %bb.0:
; VI-STRICT-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; VI-STRICT-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; VI-STRICT-NEXT:    s_waitcnt lgkmcnt(0)
; VI-STRICT-NEXT:    v_mov_b32_e32 v1, s1
; VI-STRICT-NEXT:    v_add_u32_e32 v0, vcc, s0, v0
; VI-STRICT-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-STRICT-NEXT:    flat_load_dwordx2 v[2:3], v[0:1] glc
; VI-STRICT-NEXT:    s_waitcnt vmcnt(0)
; VI-STRICT-NEXT:    v_add_u32_e32 v4, vcc, 8, v0
; VI-STRICT-NEXT:    v_addc_u32_e32 v5, vcc, 0, v1, vcc
; VI-STRICT-NEXT:    flat_load_dwordx2 v[4:5], v[4:5] glc
; VI-STRICT-NEXT:    s_waitcnt vmcnt(0)
; VI-STRICT-NEXT:    v_add_f64 v[2:3], v[2:3], v[2:3]
; VI-STRICT-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
; VI-STRICT-NEXT:    flat_store_dwordx2 v[0:1], v[2:3]
; VI-STRICT-NEXT:    s_endpgm
;
; VI-CONTRACT-LABEL: fadd_a_a_b_f64_fast_add0:
; VI-CONTRACT:       ; %bb.0:
; VI-CONTRACT-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; VI-CONTRACT-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; VI-CONTRACT-NEXT:    s_waitcnt lgkmcnt(0)
; VI-CONTRACT-NEXT:    v_mov_b32_e32 v1, s1
; VI-CONTRACT-NEXT:    v_add_u32_e32 v0, vcc, s0, v0
; VI-CONTRACT-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-CONTRACT-NEXT:    v_add_u32_e32 v2, vcc, 8, v0
; VI-CONTRACT-NEXT:    v_addc_u32_e32 v3, vcc, 0, v1, vcc
; VI-CONTRACT-NEXT:    flat_load_dwordx2 v[4:5], v[0:1] glc
; VI-CONTRACT-NEXT:    s_waitcnt vmcnt(0)
; VI-CONTRACT-NEXT:    flat_load_dwordx2 v[2:3], v[2:3] glc
; VI-CONTRACT-NEXT:    s_waitcnt vmcnt(0)
; VI-CONTRACT-NEXT:    v_fma_f64 v[2:3], v[4:5], 2.0, v[2:3]
; VI-CONTRACT-NEXT:    flat_store_dwordx2 v[0:1], v[2:3]
; VI-CONTRACT-NEXT:    s_endpgm
  %tid = call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
  %gep.0 = getelementptr double, ptr addrspace(1) %out, i32 %tid
  %gep.1 = getelementptr double, ptr addrspace(1) %gep.0, i32 1
  %gep.out = getelementptr double, ptr addrspace(1) %out, i32 %tid

  %r0 = load volatile double, ptr addrspace(1) %gep.0
  %r1 = load volatile double, ptr addrspace(1) %gep.1

  %add.0 = fadd fast double %r0, %r0
  %add.1 = fadd double %add.0, %r1
  store double %add.1, ptr addrspace(1) %gep.out
  ret void
}

define amdgpu_kernel void @fadd_a_a_b_f64_fast_add1(ptr addrspace(1) %out, ptr addrspace(1) %in1, ptr addrspace(1) %in2) #0 {
; SI-STRICT-LABEL: fadd_a_a_b_f64_fast_add1:
; SI-STRICT:       ; %bb.0:
; SI-STRICT-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x9
; SI-STRICT-NEXT:    s_mov_b32 s3, 0xf000
; SI-STRICT-NEXT:    s_mov_b32 s2, 0
; SI-STRICT-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; SI-STRICT-NEXT:    v_mov_b32_e32 v1, 0
; SI-STRICT-NEXT:    s_waitcnt lgkmcnt(0)
; SI-STRICT-NEXT:    buffer_load_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64 glc
; SI-STRICT-NEXT:    s_waitcnt vmcnt(0)
; SI-STRICT-NEXT:    buffer_load_dwordx2 v[4:5], v[0:1], s[0:3], 0 addr64 offset:8 glc
; SI-STRICT-NEXT:    s_waitcnt vmcnt(0)
; SI-STRICT-NEXT:    v_add_f64 v[2:3], v[2:3], v[2:3]
; SI-STRICT-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
; SI-STRICT-NEXT:    buffer_store_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64
; SI-STRICT-NEXT:    s_endpgm
;
; SI-CONTRACT-LABEL: fadd_a_a_b_f64_fast_add1:
; SI-CONTRACT:       ; %bb.0:
; SI-CONTRACT-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x9
; SI-CONTRACT-NEXT:    s_mov_b32 s3, 0xf000
; SI-CONTRACT-NEXT:    s_mov_b32 s2, 0
; SI-CONTRACT-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; SI-CONTRACT-NEXT:    v_mov_b32_e32 v1, 0
; SI-CONTRACT-NEXT:    s_waitcnt lgkmcnt(0)
; SI-CONTRACT-NEXT:    buffer_load_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64 glc
; SI-CONTRACT-NEXT:    s_waitcnt vmcnt(0)
; SI-CONTRACT-NEXT:    buffer_load_dwordx2 v[4:5], v[0:1], s[0:3], 0 addr64 offset:8 glc
; SI-CONTRACT-NEXT:    s_waitcnt vmcnt(0)
; SI-CONTRACT-NEXT:    v_fma_f64 v[2:3], v[2:3], 2.0, v[4:5]
; SI-CONTRACT-NEXT:    buffer_store_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64
; SI-CONTRACT-NEXT:    s_endpgm
;
; VI-STRICT-LABEL: fadd_a_a_b_f64_fast_add1:
; VI-STRICT:       ; %bb.0:
; VI-STRICT-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; VI-STRICT-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; VI-STRICT-NEXT:    s_waitcnt lgkmcnt(0)
; VI-STRICT-NEXT:    v_mov_b32_e32 v1, s1
; VI-STRICT-NEXT:    v_add_u32_e32 v0, vcc, s0, v0
; VI-STRICT-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-STRICT-NEXT:    flat_load_dwordx2 v[2:3], v[0:1] glc
; VI-STRICT-NEXT:    s_waitcnt vmcnt(0)
; VI-STRICT-NEXT:    v_add_u32_e32 v4, vcc, 8, v0
; VI-STRICT-NEXT:    v_addc_u32_e32 v5, vcc, 0, v1, vcc
; VI-STRICT-NEXT:    flat_load_dwordx2 v[4:5], v[4:5] glc
; VI-STRICT-NEXT:    s_waitcnt vmcnt(0)
; VI-STRICT-NEXT:    v_add_f64 v[2:3], v[2:3], v[2:3]
; VI-STRICT-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
; VI-STRICT-NEXT:    flat_store_dwordx2 v[0:1], v[2:3]
; VI-STRICT-NEXT:    s_endpgm
;
; VI-CONTRACT-LABEL: fadd_a_a_b_f64_fast_add1:
; VI-CONTRACT:       ; %bb.0:
; VI-CONTRACT-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; VI-CONTRACT-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; VI-CONTRACT-NEXT:    s_waitcnt lgkmcnt(0)
; VI-CONTRACT-NEXT:    v_mov_b32_e32 v1, s1
; VI-CONTRACT-NEXT:    v_add_u32_e32 v0, vcc, s0, v0
; VI-CONTRACT-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-CONTRACT-NEXT:    v_add_u32_e32 v2, vcc, 8, v0
; VI-CONTRACT-NEXT:    v_addc_u32_e32 v3, vcc, 0, v1, vcc
; VI-CONTRACT-NEXT:    flat_load_dwordx2 v[4:5], v[0:1] glc
; VI-CONTRACT-NEXT:    s_waitcnt vmcnt(0)
; VI-CONTRACT-NEXT:    flat_load_dwordx2 v[2:3], v[2:3] glc
; VI-CONTRACT-NEXT:    s_waitcnt vmcnt(0)
; VI-CONTRACT-NEXT:    v_fma_f64 v[2:3], v[4:5], 2.0, v[2:3]
; VI-CONTRACT-NEXT:    flat_store_dwordx2 v[0:1], v[2:3]
; VI-CONTRACT-NEXT:    s_endpgm
  %tid = call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
  %gep.0 = getelementptr double, ptr addrspace(1) %out, i32 %tid
  %gep.1 = getelementptr double, ptr addrspace(1) %gep.0, i32 1
  %gep.out = getelementptr double, ptr addrspace(1) %out, i32 %tid

  %r0 = load volatile double, ptr addrspace(1) %gep.0
  %r1 = load volatile double, ptr addrspace(1) %gep.1

  %add.0 = fadd double %r0, %r0
  %add.1 = fadd fast double %add.0, %r1
  store double %add.1, ptr addrspace(1) %gep.out
  ret void
}

define amdgpu_kernel void @fadd_a_a_b_f64_fast(ptr addrspace(1) %out, ptr addrspace(1) %in1, ptr addrspace(1) %in2) #0 {
; SI-LABEL: fadd_a_a_b_f64_fast:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x9
; SI-NEXT:    s_mov_b32 s3, 0xf000
; SI-NEXT:    s_mov_b32 s2, 0
; SI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; SI-NEXT:    v_mov_b32_e32 v1, 0
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    buffer_load_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64 glc
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    buffer_load_dwordx2 v[4:5], v[0:1], s[0:3], 0 addr64 offset:8 glc
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_fma_f64 v[2:3], v[2:3], 2.0, v[4:5]
; SI-NEXT:    buffer_store_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64
; SI-NEXT:    s_endpgm
;
; VI-LABEL: fadd_a_a_b_f64_fast:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; VI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    v_add_u32_e32 v0, vcc, s0, v0
; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT:    v_add_u32_e32 v2, vcc, 8, v0
; VI-NEXT:    v_addc_u32_e32 v3, vcc, 0, v1, vcc
; VI-NEXT:    flat_load_dwordx2 v[4:5], v[0:1] glc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    flat_load_dwordx2 v[2:3], v[2:3] glc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_fma_f64 v[2:3], v[4:5], 2.0, v[2:3]
; VI-NEXT:    flat_store_dwordx2 v[0:1], v[2:3]
; VI-NEXT:    s_endpgm
  %tid = call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
  %gep.0 = getelementptr double, ptr addrspace(1) %out, i32 %tid
  %gep.1 = getelementptr double, ptr addrspace(1) %gep.0, i32 1
  %gep.out = getelementptr double, ptr addrspace(1) %out, i32 %tid

  %r0 = load volatile double, ptr addrspace(1) %gep.0
  %r1 = load volatile double, ptr addrspace(1) %gep.1

  %add.0 = fadd fast double %r0, %r0
  %add.1 = fadd fast double %add.0, %r1
  store double %add.1, ptr addrspace(1) %gep.out
  ret void
}

declare i32 @llvm.amdgcn.workitem.id.x() #1
declare double @llvm.fmuladd.f64(double, double, double) #1

attributes #0 = { nounwind }
attributes #1 = { nounwind readnone }