llvm/test/CodeGen/AMDGPU/load-constant-f64.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
; RUN: llc -mtriple=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GFX6-NOHSA %s
; RUN: llc -mtriple=amdgcn-amdhsa -mcpu=kaveri -verify-machineinstrs < %s | FileCheck -check-prefix=GFX7-HSA %s
; RUN: llc -mtriple=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=GFX8-NOHSA %s
; RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -verify-machineinstrs < %s | FileCheck -check-prefix=GFX12 %s

; FUNC-LABEL: {{^}}constant_load_f64:
define amdgpu_kernel void @constant_load_f64(ptr addrspace(1) %out, ptr addrspace(4) %in) #0 {
; GFX6-NOHSA-LABEL: constant_load_f64:
; GFX6-NOHSA:       ; %bb.0:
; GFX6-NOHSA-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
; GFX6-NOHSA-NEXT:    s_waitcnt lgkmcnt(0)
; GFX6-NOHSA-NEXT:    s_load_dwordx2 s[4:5], s[2:3], 0x0
; GFX6-NOHSA-NEXT:    s_mov_b32 s3, 0xf000
; GFX6-NOHSA-NEXT:    s_mov_b32 s2, -1
; GFX6-NOHSA-NEXT:    s_waitcnt lgkmcnt(0)
; GFX6-NOHSA-NEXT:    v_mov_b32_e32 v0, s4
; GFX6-NOHSA-NEXT:    v_mov_b32_e32 v1, s5
; GFX6-NOHSA-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; GFX6-NOHSA-NEXT:    s_endpgm
;
; GFX7-HSA-LABEL: constant_load_f64:
; GFX7-HSA:       ; %bb.0:
; GFX7-HSA-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; GFX7-HSA-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-HSA-NEXT:    s_load_dwordx2 s[2:3], s[2:3], 0x0
; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s0
; GFX7-HSA-NEXT:    v_mov_b32_e32 v1, s1
; GFX7-HSA-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s2
; GFX7-HSA-NEXT:    v_mov_b32_e32 v3, s3
; GFX7-HSA-NEXT:    flat_store_dwordx2 v[0:1], v[2:3]
; GFX7-HSA-NEXT:    s_endpgm
;
; GFX8-NOHSA-LABEL: constant_load_f64:
; GFX8-NOHSA:       ; %bb.0:
; GFX8-NOHSA-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX8-NOHSA-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NOHSA-NEXT:    s_load_dwordx2 s[2:3], s[2:3], 0x0
; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v0, s0
; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v1, s1
; GFX8-NOHSA-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v2, s2
; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v3, s3
; GFX8-NOHSA-NEXT:    flat_store_dwordx2 v[0:1], v[2:3]
; GFX8-NOHSA-NEXT:    s_endpgm
;
; GFX12-LABEL: constant_load_f64:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    s_load_b64 s[2:3], s[2:3], 0x0
; GFX12-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-NEXT:    s_endpgm
  %ld = load double, ptr addrspace(4) %in
  store double %ld, ptr addrspace(1) %out
  ret void
}

attributes #0 = { nounwind }

; Tests whether a load-chain of 8 constants of 64bit each gets vectorized into a wider load.
define amdgpu_kernel void @constant_load_2v4f64(ptr addrspace(4) noalias nocapture readonly %weights, ptr addrspace(1) noalias nocapture %out_ptr) {
; GFX6-NOHSA-LABEL: constant_load_2v4f64:
; GFX6-NOHSA:       ; %bb.0: ; %entry
; GFX6-NOHSA-NEXT:    s_load_dwordx4 s[16:19], s[4:5], 0x9
; GFX6-NOHSA-NEXT:    s_waitcnt lgkmcnt(0)
; GFX6-NOHSA-NEXT:    s_load_dwordx2 s[24:25], s[18:19], 0x0
; GFX6-NOHSA-NEXT:    s_load_dwordx16 s[0:15], s[16:17], 0x0
; GFX6-NOHSA-NEXT:    s_mov_b32 s23, 0xf000
; GFX6-NOHSA-NEXT:    s_mov_b32 s22, -1
; GFX6-NOHSA-NEXT:    s_mov_b32 s20, s18
; GFX6-NOHSA-NEXT:    s_mov_b32 s21, s19
; GFX6-NOHSA-NEXT:    s_waitcnt lgkmcnt(0)
; GFX6-NOHSA-NEXT:    v_mov_b32_e32 v0, s24
; GFX6-NOHSA-NEXT:    v_mov_b32_e32 v1, s25
; GFX6-NOHSA-NEXT:    v_add_f64 v[0:1], s[0:1], v[0:1]
; GFX6-NOHSA-NEXT:    v_add_f64 v[0:1], s[2:3], v[0:1]
; GFX6-NOHSA-NEXT:    v_add_f64 v[0:1], s[4:5], v[0:1]
; GFX6-NOHSA-NEXT:    v_add_f64 v[0:1], s[6:7], v[0:1]
; GFX6-NOHSA-NEXT:    v_add_f64 v[0:1], s[8:9], v[0:1]
; GFX6-NOHSA-NEXT:    v_add_f64 v[0:1], s[10:11], v[0:1]
; GFX6-NOHSA-NEXT:    v_add_f64 v[0:1], s[12:13], v[0:1]
; GFX6-NOHSA-NEXT:    v_add_f64 v[0:1], s[14:15], v[0:1]
; GFX6-NOHSA-NEXT:    buffer_store_dwordx2 v[0:1], off, s[20:23], 0
; GFX6-NOHSA-NEXT:    s_endpgm
;
; GFX7-HSA-LABEL: constant_load_2v4f64:
; GFX7-HSA:       ; %bb.0: ; %entry
; GFX7-HSA-NEXT:    s_load_dwordx4 s[16:19], s[8:9], 0x0
; GFX7-HSA-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-HSA-NEXT:    s_load_dwordx2 s[20:21], s[18:19], 0x0
; GFX7-HSA-NEXT:    s_load_dwordx16 s[0:15], s[16:17], 0x0
; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s18
; GFX7-HSA-NEXT:    v_mov_b32_e32 v3, s19
; GFX7-HSA-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s20
; GFX7-HSA-NEXT:    v_mov_b32_e32 v1, s21
; GFX7-HSA-NEXT:    v_add_f64 v[0:1], s[0:1], v[0:1]
; GFX7-HSA-NEXT:    v_add_f64 v[0:1], s[2:3], v[0:1]
; GFX7-HSA-NEXT:    v_add_f64 v[0:1], s[4:5], v[0:1]
; GFX7-HSA-NEXT:    v_add_f64 v[0:1], s[6:7], v[0:1]
; GFX7-HSA-NEXT:    v_add_f64 v[0:1], s[8:9], v[0:1]
; GFX7-HSA-NEXT:    v_add_f64 v[0:1], s[10:11], v[0:1]
; GFX7-HSA-NEXT:    v_add_f64 v[0:1], s[12:13], v[0:1]
; GFX7-HSA-NEXT:    v_add_f64 v[0:1], s[14:15], v[0:1]
; GFX7-HSA-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
; GFX7-HSA-NEXT:    s_endpgm
;
; GFX8-NOHSA-LABEL: constant_load_2v4f64:
; GFX8-NOHSA:       ; %bb.0: ; %entry
; GFX8-NOHSA-NEXT:    s_load_dwordx4 s[16:19], s[4:5], 0x24
; GFX8-NOHSA-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NOHSA-NEXT:    s_load_dwordx2 s[20:21], s[18:19], 0x0
; GFX8-NOHSA-NEXT:    s_load_dwordx16 s[0:15], s[16:17], 0x0
; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v2, s18
; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v3, s19
; GFX8-NOHSA-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v0, s20
; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v1, s21
; GFX8-NOHSA-NEXT:    v_add_f64 v[0:1], s[0:1], v[0:1]
; GFX8-NOHSA-NEXT:    v_add_f64 v[0:1], s[2:3], v[0:1]
; GFX8-NOHSA-NEXT:    v_add_f64 v[0:1], s[4:5], v[0:1]
; GFX8-NOHSA-NEXT:    v_add_f64 v[0:1], s[6:7], v[0:1]
; GFX8-NOHSA-NEXT:    v_add_f64 v[0:1], s[8:9], v[0:1]
; GFX8-NOHSA-NEXT:    v_add_f64 v[0:1], s[10:11], v[0:1]
; GFX8-NOHSA-NEXT:    v_add_f64 v[0:1], s[12:13], v[0:1]
; GFX8-NOHSA-NEXT:    v_add_f64 v[0:1], s[14:15], v[0:1]
; GFX8-NOHSA-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
; GFX8-NOHSA-NEXT:    s_endpgm
;
; GFX12-LABEL: constant_load_2v4f64:
; GFX12:       ; %bb.0: ; %entry
; GFX12-NEXT:    s_load_b128 s[16:19], s[4:5], 0x24
; GFX12-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    s_load_b64 s[20:21], s[18:19], 0x0
; GFX12-NEXT:    s_load_b512 s[0:15], s[16:17], 0x0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_add_f64_e64 v[0:1], s[0:1], s[20:21]
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-NEXT:    v_add_f64_e32 v[0:1], s[2:3], v[0:1]
; GFX12-NEXT:    v_add_f64_e32 v[0:1], s[4:5], v[0:1]
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-NEXT:    v_add_f64_e32 v[0:1], s[6:7], v[0:1]
; GFX12-NEXT:    v_add_f64_e32 v[0:1], s[8:9], v[0:1]
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-NEXT:    v_add_f64_e32 v[0:1], s[10:11], v[0:1]
; GFX12-NEXT:    v_add_f64_e32 v[0:1], s[12:13], v[0:1]
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_add_f64_e32 v[0:1], s[14:15], v[0:1]
; GFX12-NEXT:    global_store_b64 v2, v[0:1], s[18:19]
; GFX12-NEXT:    s_endpgm
entry:
  %out_ptr.promoted = load double, ptr addrspace(1) %out_ptr, align 4
  %tmp = load double, ptr addrspace(4) %weights, align 4
  %add = fadd double %tmp, %out_ptr.promoted
  %arrayidx.1 = getelementptr inbounds double, ptr addrspace(4) %weights, i64 1
  %tmp1 = load double, ptr addrspace(4) %arrayidx.1, align 4
  %add.1 = fadd double %tmp1, %add
  %arrayidx.2 = getelementptr inbounds double, ptr addrspace(4) %weights, i64 2
  %tmp2 = load double, ptr addrspace(4) %arrayidx.2, align 4
  %add.2 = fadd double %tmp2, %add.1
  %arrayidx.3 = getelementptr inbounds double, ptr addrspace(4) %weights, i64 3
  %tmp3 = load double, ptr addrspace(4) %arrayidx.3, align 4
  %add.3 = fadd double %tmp3, %add.2
  %arrayidx.4 = getelementptr inbounds double, ptr addrspace(4) %weights, i64 4
  %tmp4 = load double, ptr addrspace(4) %arrayidx.4, align 4
  %add.4 = fadd double %tmp4, %add.3
  %arrayidx.5 = getelementptr inbounds double, ptr addrspace(4) %weights, i64 5
  %tmp5 = load double, ptr addrspace(4) %arrayidx.5, align 4
  %add.5 = fadd double %tmp5, %add.4
  %arrayidx.6 = getelementptr inbounds double, ptr addrspace(4) %weights, i64 6
  %tmp6 = load double, ptr addrspace(4) %arrayidx.6, align 4
  %add.6 = fadd double %tmp6, %add.5
  %arrayidx.7 = getelementptr inbounds double, ptr addrspace(4) %weights, i64 7
  %tmp7 = load double, ptr addrspace(4) %arrayidx.7, align 4
  %add.7 = fadd double %tmp7, %add.6
  store double %add.7, ptr addrspace(1) %out_ptr, align 4
  ret void
}