Skip to content

Commit

Permalink
AMDGPU: Switch some tests to generated checks
Browse files Browse the repository at this point in the history
  • Loading branch information
@arsenm
arsenm committed Nov 12, 2022
1 parent d93be48 commit 5247ae9
Show file tree
Hide file tree
Showing 2 changed files with 500 additions and 85 deletions.
213 changes: 178 additions & 35 deletions llvm/test/CodeGen/AMDGPU/sint_to_fp.f64.ll
View file
@@ -1,59 +1,174 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=hawaii -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,CI %s
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=fiji -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,VI %s

declare i32 @llvm.amdgcn.workitem.id.x() nounwind readnone

; GCN-LABEL: {{^}}sint_to_fp_i32_to_f64
; GCN: v_cvt_f64_i32_e32
define amdgpu_kernel void @sint_to_fp_i32_to_f64(double addrspace(1)* %out, i32 %in) {
; CI-LABEL: sint_to_fp_i32_to_f64:
; CI: ; %bb.0:
; CI-NEXT: s_load_dword s2, s[4:5], 0x2
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: v_cvt_f64_i32_e32 v[0:1], s2
; CI-NEXT: v_mov_b32_e32 v3, s1
; CI-NEXT: v_mov_b32_e32 v2, s0
; CI-NEXT: flat_store_dwordx2 v[2:3], v[0:1]
; CI-NEXT: s_endpgm
;
; VI-LABEL: sint_to_fp_i32_to_f64:
; VI: ; %bb.0:
; VI-NEXT: s_load_dword s2, s[4:5], 0x8
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_cvt_f64_i32_e32 v[0:1], s2
; VI-NEXT: v_mov_b32_e32 v3, s1
; VI-NEXT: v_mov_b32_e32 v2, s0
; VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1]
; VI-NEXT: s_endpgm
%result = sitofp i32 %in to double
store double %result, double addrspace(1)* %out
ret void
}

; We can't fold the SGPRs into v_cndmask_b32_e64, because it already
; uses an SGPR (implicit vcc).

; GCN-LABEL: {{^}}sint_to_fp_i1_f64:
; GCN-DAG: s_cmp_eq_u32
; GCN-DAG: s_cselect_b32 s[[SSEL:[0-9]+]], 0xbff00000, 0
; GCN-DAG: v_mov_b32_e32 v[[ZERO:[0-9]+]], 0{{$}}
; GCN-DAG: v_mov_b32_e32 v[[SEL:[0-9]+]], s[[SSEL]]
; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, v[[[ZERO]]:[[SEL]]]
; GCN: s_endpgm
define amdgpu_kernel void @sint_to_fp_i1_f64(double addrspace(1)* %out, i32 %in) {
; CI-LABEL: sint_to_fp_i1_f64:
; CI: ; %bb.0:
; CI-NEXT: s_load_dword s2, s[4:5], 0x2
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0
; CI-NEXT: v_mov_b32_e32 v0, 0
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: s_cmp_eq_u32 s2, 0
; CI-NEXT: s_cselect_b32 s2, 0xbff00000, 0
; CI-NEXT: v_mov_b32_e32 v3, s1
; CI-NEXT: v_mov_b32_e32 v1, s2
; CI-NEXT: v_mov_b32_e32 v2, s0
; CI-NEXT: flat_store_dwordx2 v[2:3], v[0:1]
; CI-NEXT: s_endpgm
;
; VI-LABEL: sint_to_fp_i1_f64:
; VI: ; %bb.0:
; VI-NEXT: s_load_dword s2, s[4:5], 0x8
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0
; VI-NEXT: v_mov_b32_e32 v0, 0
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: s_cmp_eq_u32 s2, 0
; VI-NEXT: s_cselect_b32 s2, 0xbff00000, 0
; VI-NEXT: v_mov_b32_e32 v3, s1
; VI-NEXT: v_mov_b32_e32 v1, s2
; VI-NEXT: v_mov_b32_e32 v2, s0
; VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1]
; VI-NEXT: s_endpgm
%cmp = icmp eq i32 %in, 0
%fp = sitofp i1 %cmp to double
store double %fp, double addrspace(1)* %out, align 4
ret void
}

; GCN-LABEL: {{^}}sint_to_fp_i1_f64_load:
; GCN: v_cndmask_b32_e64 [[IRESULT:v[0-9]]], 0, -1
; GCN: v_cvt_f64_i32_e32 [[RESULT:v\[[0-9]+:[0-9]\]]], [[IRESULT]]
; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
; GCN: s_endpgm
define amdgpu_kernel void @sint_to_fp_i1_f64_load(double addrspace(1)* %out, i1 %in) {
; CI-LABEL: sint_to_fp_i1_f64_load:
; CI: ; %bb.0:
; CI-NEXT: s_load_dword s2, s[4:5], 0x2
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: s_bitcmp1_b32 s2, 0
; CI-NEXT: s_cselect_b64 s[2:3], -1, 0
; CI-NEXT: v_cndmask_b32_e64 v0, 0, -1, s[2:3]
; CI-NEXT: v_cvt_f64_i32_e32 v[0:1], v0
; CI-NEXT: v_mov_b32_e32 v3, s1
; CI-NEXT: v_mov_b32_e32 v2, s0
; CI-NEXT: flat_store_dwordx2 v[2:3], v[0:1]
; CI-NEXT: s_endpgm
;
; VI-LABEL: sint_to_fp_i1_f64_load:
; VI: ; %bb.0:
; VI-NEXT: s_load_dword s2, s[4:5], 0x8
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: s_bitcmp1_b32 s2, 0
; VI-NEXT: s_cselect_b64 s[2:3], -1, 0
; VI-NEXT: v_cndmask_b32_e64 v0, 0, -1, s[2:3]
; VI-NEXT: v_cvt_f64_i32_e32 v[0:1], v0
; VI-NEXT: v_mov_b32_e32 v3, s1
; VI-NEXT: v_mov_b32_e32 v2, s0
; VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1]
; VI-NEXT: s_endpgm
%fp = sitofp i1 %in to double
store double %fp, double addrspace(1)* %out, align 8
ret void
}

; GCN-LABEL: @s_sint_to_fp_i64_to_f64
define amdgpu_kernel void @s_sint_to_fp_i64_to_f64(double addrspace(1)* %out, i64 %in) {
; CI-LABEL: s_sint_to_fp_i64_to_f64:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x0
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: v_cvt_f64_i32_e32 v[0:1], s3
; CI-NEXT: v_cvt_f64_u32_e32 v[2:3], s2
; CI-NEXT: v_mov_b32_e32 v4, s0
; CI-NEXT: v_mov_b32_e32 v5, s1
; CI-NEXT: v_ldexp_f64 v[0:1], v[0:1], 32
; CI-NEXT: v_add_f64 v[0:1], v[0:1], v[2:3]
; CI-NEXT: flat_store_dwordx2 v[4:5], v[0:1]
; CI-NEXT: s_endpgm
;
; VI-LABEL: s_sint_to_fp_i64_to_f64:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x0
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_cvt_f64_i32_e32 v[0:1], s3
; VI-NEXT: v_cvt_f64_u32_e32 v[2:3], s2
; VI-NEXT: v_ldexp_f64 v[0:1], v[0:1], 32
; VI-NEXT: v_add_f64 v[0:1], v[0:1], v[2:3]
; VI-NEXT: v_mov_b32_e32 v2, s0
; VI-NEXT: v_mov_b32_e32 v3, s1
; VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1]
; VI-NEXT: s_endpgm
%result = sitofp i64 %in to double
store double %result, double addrspace(1)* %out
ret void
}

; GCN-LABEL: @v_sint_to_fp_i64_to_f64
; GCN: flat_load_dwordx2 v[[[LO:[0-9]+]]:[[HI:[0-9]+]]]
; GCN-DAG: v_cvt_f64_i32_e32 [[HI_CONV:v\[[0-9]+:[0-9]+\]]], v[[HI]]
; GCN-DAG: v_cvt_f64_u32_e32 [[LO_CONV:v\[[0-9]+:[0-9]+\]]], v[[LO]]
; GCN-DAG: v_ldexp_f64 [[LDEXP:v\[[0-9]+:[0-9]+\]]], [[HI_CONV]], 32
; GCN: v_add_f64 [[RESULT:v\[[0-9]+:[0-9]+\]]], [[LDEXP]], [[LO_CONV]]
; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
define amdgpu_kernel void @v_sint_to_fp_i64_to_f64(double addrspace(1)* %out, i64 addrspace(1)* %in) {
; CI-LABEL: v_sint_to_fp_i64_to_f64:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x0
; CI-NEXT: v_lshlrev_b32_e32 v0, 3, v0
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: v_mov_b32_e32 v1, s3
; CI-NEXT: v_add_i32_e32 v0, vcc, s2, v0
; CI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
; CI-NEXT: flat_load_dwordx2 v[0:1], v[0:1]
; CI-NEXT: s_waitcnt vmcnt(0)
; CI-NEXT: v_cvt_f64_i32_e32 v[1:2], v1
; CI-NEXT: v_cvt_f64_u32_e32 v[3:4], v0
; CI-NEXT: v_ldexp_f64 v[0:1], v[1:2], 32
; CI-NEXT: v_mov_b32_e32 v2, s0
; CI-NEXT: v_add_f64 v[0:1], v[0:1], v[3:4]
; CI-NEXT: v_mov_b32_e32 v3, s1
; CI-NEXT: flat_store_dwordx2 v[2:3], v[0:1]
; CI-NEXT: s_endpgm
;
; VI-LABEL: v_sint_to_fp_i64_to_f64:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x0
; VI-NEXT: v_lshlrev_b32_e32 v0, 3, v0
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v1, s3
; VI-NEXT: v_add_u32_e32 v0, vcc, s2, v0
; VI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT: flat_load_dwordx2 v[0:1], v[0:1]
; VI-NEXT: s_waitcnt vmcnt(0)
; VI-NEXT: v_cvt_f64_i32_e32 v[1:2], v1
; VI-NEXT: v_cvt_f64_u32_e32 v[3:4], v0
; VI-NEXT: v_ldexp_f64 v[1:2], v[1:2], 32
; VI-NEXT: v_add_f64 v[0:1], v[1:2], v[3:4]
; VI-NEXT: v_mov_b32_e32 v2, s0
; VI-NEXT: v_mov_b32_e32 v3, s1
; VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1]
; VI-NEXT: s_endpgm
%tid = call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
%gep = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
%val = load i64, i64 addrspace(1)* %gep, align 8
Expand All @@ -63,25 +178,53 @@ define amdgpu_kernel void @v_sint_to_fp_i64_to_f64(double addrspace(1)* %out, i6
}

; FIXME: bfe and sext on VI+
; GCN-LABEL: {{^}}s_sint_to_fp_i8_to_f64:
; GCN: s_load_dword [[VAL:s[0-9]+]]
; CI-NOT: bfe
; CI: s_sext_i32_i8 [[SEXT:s[0-9]+]], [[VAL]]

; VI: s_bfe_i32 [[BFE:s[0-9]+]], [[VAL]], 0x80000
; VI: s_sext_i32_i16 [[SEXT:s[0-9]+]], [[BFE]]

; GCN: v_cvt_f64_i32_e32 v{{\[[0-9]+:[0-9]+\]}}, [[SEXT]]
define amdgpu_kernel void @s_sint_to_fp_i8_to_f64(double addrspace(1)* %out, i8 %in) {
; CI-LABEL: s_sint_to_fp_i8_to_f64:
; CI: ; %bb.0:
; CI-NEXT: s_load_dword s2, s[4:5], 0x2
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: s_sext_i32_i8 s2, s2
; CI-NEXT: v_cvt_f64_i32_e32 v[0:1], s2
; CI-NEXT: v_mov_b32_e32 v3, s1
; CI-NEXT: v_mov_b32_e32 v2, s0
; CI-NEXT: flat_store_dwordx2 v[2:3], v[0:1]
; CI-NEXT: s_endpgm
;
; VI-LABEL: s_sint_to_fp_i8_to_f64:
; VI: ; %bb.0:
; VI-NEXT: s_load_dword s2, s[4:5], 0x8
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: s_bfe_i32 s2, s2, 0x80000
; VI-NEXT: s_sext_i32_i16 s2, s2
; VI-NEXT: v_cvt_f64_i32_e32 v[0:1], s2
; VI-NEXT: v_mov_b32_e32 v3, s1
; VI-NEXT: v_mov_b32_e32 v2, s0
; VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1]
; VI-NEXT: s_endpgm
%fp = sitofp i8 %in to double
store double %fp, double addrspace(1)* %out
ret void
}

; GCN-LABEL: {{^}}v_sint_to_fp_i8_to_f64:
; GCN: v_bfe_i32 [[SEXT:v[0-9]+]]
; GCN: v_cvt_f64_i32_e32 v{{\[[0-9]+:[0-9]+\]}}, [[SEXT]]
define double @v_sint_to_fp_i8_to_f64(i8 %in) {
; CI-LABEL: v_sint_to_fp_i8_to_f64:
; CI: ; %bb.0:
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CI-NEXT: v_bfe_i32 v0, v0, 0, 8
; CI-NEXT: v_cvt_f64_i32_e32 v[0:1], v0
; CI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: v_sint_to_fp_i8_to_f64:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_bfe_i32 v0, v0, 0, 8
; VI-NEXT: v_bfe_i32 v0, v0, 0, 16
; VI-NEXT: v_cvt_f64_i32_e32 v[0:1], v0
; VI-NEXT: s_setpc_b64 s[30:31]
%fp = sitofp i8 %in to double
ret double %fp
}
;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
; GCN: {{.*}}

0 comments on commit 5247ae9

Please sign in to comment.