Google Git
Sign in
llvm/llvm-project/ba767d0bbbde4107700ff66ecfd97eb75d85a35d/./llvm/test/CodeGen/AMDGPU/fract-match.ll
blob: da538744fb8d2e935a52be2c06f2664cff41eaba [file]
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
; Test with buggy fract, shouldn't match
; RUN: opt -S -mtriple=amdgcn-amd-amdpal -mcpu=tahiti -amdgpu-codegenprepare %s | FileCheck -check-prefixes=IR,GFX6-IR %s
; Working fract, but no f16
; RUN: opt -S -mtriple=amdgcn-amd-amdpal -mcpu=kaveri -amdgpu-codegenprepare %s | FileCheck -check-prefixes=IR,IR-FRACT,GFX7-IR %s
; Working fract and f16 support
; RUN: opt -S -mtriple=amdgcn-amd-amdpal -mcpu=gfx900 -amdgpu-codegenprepare %s | FileCheck -check-prefixes=IR,IR-FRACT,IR-LEGALF16 %s
; RUN: opt -S -mtriple=amdgcn-amd-amdpal -mcpu=gfx1100 -amdgpu-codegenprepare %s | FileCheck -check-prefixes=IR,IR-FRACT,IR-LEGALF16 %s
; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=tahiti < %s | FileCheck -check-prefixes=GCN,GFX6 %s
; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=kaveri < %s | FileCheck -check-prefixes=GCN,GFX7 %s
; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX8 %s
; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx1100 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GCN,GFX11,GFX11-TRUE16 %s
; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx1100 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GCN,GFX11,GFX11-FAKE16 %s
; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-TRUE16 %s
; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-FAKE16 %s
; Test patterns to match v_fract_* instructions.
; Expansion as it appears in the library with the extra output for
; floor. We can fold in the nan check into the instruction, but the
; inf check must remain.
define float @safe_math_fract_f32(float %x, ptr addrspace(1) writeonly captures(none) %ip) {
; GFX6-IR-LABEL: define float @safe_math_fract_f32(
; GFX6-IR-SAME: float [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0:[0-9]+]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; GFX6-IR-NEXT: [[UNO:%.*]] = fcmp uno float [[X]], 0.000000e+00
; GFX6-IR-NEXT: [[COND:%.*]] = select i1 [[UNO]], float [[X]], float [[MIN]]
; GFX6-IR-NEXT: [[FABS:%.*]] = tail call float @llvm.fabs.f32(float [[X]])
; GFX6-IR-NEXT: [[CMPINF:%.*]] = fcmp oeq float [[FABS]], 0x7FF0000000000000
; GFX6-IR-NEXT: [[COND6:%.*]] = select i1 [[CMPINF]], float 0.000000e+00, float [[COND]]
; GFX6-IR-NEXT: store float [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; GFX6-IR-NEXT: ret float [[COND6]]
;
; IR-FRACT-LABEL: define float @safe_math_fract_f32(
; IR-FRACT-SAME: float [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0:[0-9]+]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; IR-FRACT-NEXT: [[COND:%.*]] = call float @llvm.amdgcn.fract.f32(float [[X]])
; IR-FRACT-NEXT: [[FABS:%.*]] = tail call float @llvm.fabs.f32(float [[X]])
; IR-FRACT-NEXT: [[CMPINF:%.*]] = fcmp oeq float [[FABS]], 0x7FF0000000000000
; IR-FRACT-NEXT: [[COND6:%.*]] = select i1 [[CMPINF]], float 0.000000e+00, float [[COND]]
; IR-FRACT-NEXT: store float [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; IR-FRACT-NEXT: ret float [[COND6]]
;
; GFX6-LABEL: safe_math_fract_f32:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v3, v0
; GFX6-NEXT: v_sub_f32_e32 v4, v0, v3
; GFX6-NEXT: v_min_f32_e32 v4, 0x3f7fffff, v4
; GFX6-NEXT: v_cmp_u_f32_e32 vcc, v0, v0
; GFX6-NEXT: s_mov_b32 s8, 0x7f800000
; GFX6-NEXT: s_mov_b32 s6, 0
; GFX6-NEXT: v_cndmask_b32_e32 v4, v4, v0, vcc
; GFX6-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s8
; GFX6-NEXT: s_mov_b32 s7, 0xf000
; GFX6-NEXT: s_mov_b32 s4, s6
; GFX6-NEXT: s_mov_b32 s5, s6
; GFX6-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc
; GFX6-NEXT: buffer_store_dword v3, v[1:2], s[4:7], 0 addr64
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: safe_math_fract_f32:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: s_mov_b32 s8, 0x7f800000
; GFX7-NEXT: s_mov_b32 s6, 0
; GFX7-NEXT: v_fract_f32_e32 v4, v0
; GFX7-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s8
; GFX7-NEXT: s_mov_b32 s7, 0xf000
; GFX7-NEXT: s_mov_b32 s4, s6
; GFX7-NEXT: s_mov_b32 s5, s6
; GFX7-NEXT: v_floor_f32_e32 v3, v0
; GFX7-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc
; GFX7-NEXT: buffer_store_dword v3, v[1:2], s[4:7], 0 addr64
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: safe_math_fract_f32:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: s_mov_b32 s4, 0x7f800000
; GFX8-NEXT: v_fract_f32_e32 v4, v0
; GFX8-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX8-NEXT: v_floor_f32_e32 v3, v0
; GFX8-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc
; GFX8-NEXT: global_store_dword v[1:2], v3, off
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: safe_math_fract_f32:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v3, v0
; GFX11-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX11-NEXT: v_floor_f32_e32 v4, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_3)
; GFX11-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc_lo
; GFX11-NEXT: global_store_b32 v[1:2], v4, off
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: safe_math_fract_f32:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v3, v0
; GFX12-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX12-NEXT: v_floor_f32_e32 v4, v0
; GFX12-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_3)
; GFX12-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc_lo
; GFX12-NEXT: global_store_b32 v[1:2], v4, off
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = tail call float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
%uno = fcmp uno float %x, 0.000000e+00
%cond = select i1 %uno, float %x, float %min
%fabs = tail call float @llvm.fabs.f32(float %x)
%cmpinf = fcmp oeq float %fabs, 0x7FF0000000000000
%cond6 = select i1 %cmpinf, float 0.000000e+00, float %cond
store float %floor, ptr addrspace(1) %ip, align 4
ret float %cond6
}
define <3 x float> @safe_math_fract_v3f32(<3 x float> %x, ptr addrspace(1) writeonly captures(none) %ip) {
; GFX6-LABEL: safe_math_fract_v3f32:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v7, v2
; GFX6-NEXT: v_floor_f32_e32 v5, v0
; GFX6-NEXT: v_sub_f32_e32 v8, v2, v7
; GFX6-NEXT: v_sub_f32_e32 v9, v0, v5
; GFX6-NEXT: v_min_f32_e32 v8, 0x3f7fffff, v8
; GFX6-NEXT: v_cmp_u_f32_e32 vcc, v2, v2
; GFX6-NEXT: v_min_f32_e32 v9, 0x3f7fffff, v9
; GFX6-NEXT: v_cndmask_b32_e32 v8, v8, v2, vcc
; GFX6-NEXT: v_cmp_u_f32_e32 vcc, v0, v0
; GFX6-NEXT: v_mov_b32_e32 v10, 0x204
; GFX6-NEXT: v_cndmask_b32_e32 v9, v9, v0, vcc
; GFX6-NEXT: v_cmp_class_f32_e32 vcc, v0, v10
; GFX6-NEXT: s_mov_b32 s6, 0
; GFX6-NEXT: v_cndmask_b32_e64 v0, v9, 0, vcc
; GFX6-NEXT: v_cmp_class_f32_e32 vcc, v2, v10
; GFX6-NEXT: s_mov_b32 s7, 0xf000
; GFX6-NEXT: s_mov_b32 s4, s6
; GFX6-NEXT: s_mov_b32 s5, s6
; GFX6-NEXT: v_cndmask_b32_e64 v2, v8, 0, vcc
; GFX6-NEXT: v_floor_f32_e32 v6, v1
; GFX6-NEXT: buffer_store_dword v7, v[3:4], s[4:7], 0 addr64 offset:8
; GFX6-NEXT: buffer_store_dwordx2 v[5:6], v[3:4], s[4:7], 0 addr64
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: safe_math_fract_v3f32:
; GFX7: ; %bb.0:
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_mov_b32_e32 v10, 0x204
; GFX7-NEXT: v_fract_f32_e32 v8, v0
; GFX7-NEXT: v_cmp_class_f32_e32 vcc, v0, v10
; GFX7-NEXT: s_mov_b32 s6, 0
; GFX7-NEXT: v_floor_f32_e32 v5, v0
; GFX7-NEXT: v_fract_f32_e32 v9, v2
; GFX7-NEXT: v_cndmask_b32_e64 v0, v8, 0, vcc
; GFX7-NEXT: v_cmp_class_f32_e32 vcc, v2, v10
; GFX7-NEXT: s_mov_b32 s7, 0xf000
; GFX7-NEXT: s_mov_b32 s4, s6
; GFX7-NEXT: s_mov_b32 s5, s6
; GFX7-NEXT: v_floor_f32_e32 v7, v2
; GFX7-NEXT: v_floor_f32_e32 v6, v1
; GFX7-NEXT: v_fract_f32_e32 v1, v1
; GFX7-NEXT: v_cndmask_b32_e64 v2, v9, 0, vcc
; GFX7-NEXT: buffer_store_dwordx3 v[5:7], v[3:4], s[4:7], 0 addr64
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: safe_math_fract_v3f32:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_mov_b32_e32 v10, 0x204
; GFX8-NEXT: v_fract_f32_e32 v8, v0
; GFX8-NEXT: v_cmp_class_f32_e32 vcc, v0, v10
; GFX8-NEXT: v_floor_f32_e32 v5, v0
; GFX8-NEXT: v_fract_f32_e32 v9, v2
; GFX8-NEXT: v_cndmask_b32_e64 v0, v8, 0, vcc
; GFX8-NEXT: v_cmp_class_f32_e32 vcc, v2, v10
; GFX8-NEXT: v_floor_f32_e32 v7, v2
; GFX8-NEXT: v_floor_f32_e32 v6, v1
; GFX8-NEXT: v_fract_f32_e32 v1, v1
; GFX8-NEXT: v_cndmask_b32_e64 v2, v9, 0, vcc
; GFX8-NEXT: global_store_dwordx3 v[3:4], v[5:7], off
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: safe_math_fract_v3f32:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v8, v0
; GFX11-NEXT: v_cmp_class_f32_e64 s0, v0, 0x204
; GFX11-NEXT: v_floor_f32_e32 v5, v0
; GFX11-NEXT: v_fract_f32_e32 v9, v2
; GFX11-NEXT: v_floor_f32_e32 v7, v2
; GFX11-NEXT: v_floor_f32_e32 v6, v1
; GFX11-NEXT: v_cndmask_b32_e64 v0, v8, 0, s0
; GFX11-NEXT: v_cmp_class_f32_e64 s0, v2, 0x204
; GFX11-NEXT: v_fract_f32_e32 v1, v1
; GFX11-NEXT: global_store_b96 v[3:4], v[5:7], off
; GFX11-NEXT: v_cndmask_b32_e64 v2, v9, 0, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: safe_math_fract_v3f32:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v8, v0
; GFX12-NEXT: v_cmp_class_f32_e64 s0, v0, 0x204
; GFX12-NEXT: v_floor_f32_e32 v5, v0
; GFX12-NEXT: v_fract_f32_e32 v9, v2
; GFX12-NEXT: v_floor_f32_e32 v7, v2
; GFX12-NEXT: v_floor_f32_e32 v6, v1
; GFX12-NEXT: s_wait_alu depctr_va_sdst(0)
; GFX12-NEXT: v_cndmask_b32_e64 v0, v8, 0, s0
; GFX12-NEXT: v_cmp_class_f32_e64 s0, v2, 0x204
; GFX12-NEXT: v_fract_f32_e32 v1, v1
; GFX12-NEXT: global_store_b96 v[3:4], v[5:7], off
; GFX12-NEXT: s_wait_alu depctr_va_sdst(0)
; GFX12-NEXT: v_cndmask_b32_e64 v2, v9, 0, s0
; GFX12-NEXT: s_setpc_b64 s[30:31]
; GFX6-IR-LABEL: define <3 x float> @safe_math_fract_v3f32(
; GFX6-IR-SAME: <3 x float> [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call <3 x float> @llvm.floor.v3f32(<3 x float> [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub <3 x float> [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call <3 x float> @llvm.minnum.v3f32(<3 x float> [[SUB]], <3 x float> <float 0x3FEFFFFFE0000000, float poison, float 0x3FEFFFFFE0000000>)
; GFX6-IR-NEXT: [[UNO:%.*]] = fcmp uno <3 x float> [[X]], <float 0.000000e+00, float poison, float 0.000000e+00>
; GFX6-IR-NEXT: [[COND:%.*]] = select <3 x i1> [[UNO]], <3 x float> [[X]], <3 x float> [[MIN]]
; GFX6-IR-NEXT: [[FABS:%.*]] = tail call <3 x float> @llvm.fabs.v3f32(<3 x float> [[X]])
; GFX6-IR-NEXT: [[CMPINF:%.*]] = fcmp oeq <3 x float> [[FABS]], <float 0x7FF0000000000000, float poison, float 0x7FF0000000000000>
; GFX6-IR-NEXT: [[COND6:%.*]] = select <3 x i1> [[CMPINF]], <3 x float> <float 0.000000e+00, float poison, float 0.000000e+00>, <3 x float> [[COND]]
; GFX6-IR-NEXT: store <3 x float> [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; GFX6-IR-NEXT: ret <3 x float> [[COND6]]
; IR-FRACT-LABEL: define <3 x float> @safe_math_fract_v3f32(
; IR-FRACT-SAME: <3 x float> [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0]] {
; IR-FRACT-NEXT: [[FLOOR:%.*]] = tail call <3 x float> @llvm.floor.v3f32(<3 x float> [[X]])
; IR-FRACT-NEXT: [[TMP1:%.*]] = extractelement <3 x float> [[X]], i64 0
; IR-FRACT-NEXT: [[TMP2:%.*]] = extractelement <3 x float> [[X]], i64 1
; IR-FRACT-NEXT: [[TMP3:%.*]] = extractelement <3 x float> [[X]], i64 2
; IR-FRACT-NEXT: [[TMP4:%.*]] = call float @llvm.amdgcn.fract.f32(float [[TMP1]])
; IR-FRACT-NEXT: [[TMP5:%.*]] = call float @llvm.amdgcn.fract.f32(float [[TMP2]])
; IR-FRACT-NEXT: [[TMP6:%.*]] = call float @llvm.amdgcn.fract.f32(float [[TMP3]])
; IR-FRACT-NEXT: [[TMP7:%.*]] = insertelement <3 x float> poison, float [[TMP4]], i64 0
; IR-FRACT-NEXT: [[TMP8:%.*]] = insertelement <3 x float> [[TMP7]], float [[TMP5]], i64 1
; IR-FRACT-NEXT: [[COND:%.*]] = insertelement <3 x float> [[TMP8]], float [[TMP6]], i64 2
; IR-FRACT-NEXT: [[FABS:%.*]] = tail call <3 x float> @llvm.fabs.v3f32(<3 x float> [[X]])
; IR-FRACT-NEXT: [[CMPINF:%.*]] = fcmp oeq <3 x float> [[FABS]], <float 0x7FF0000000000000, float poison, float 0x7FF0000000000000>
; IR-FRACT-NEXT: [[COND6:%.*]] = select <3 x i1> [[CMPINF]], <3 x float> <float 0.000000e+00, float poison, float 0.000000e+00>, <3 x float> [[COND]]
; IR-FRACT-NEXT: store <3 x float> [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; IR-FRACT-NEXT: ret <3 x float> [[COND6]]
%floor = tail call <3 x float> @llvm.floor.v3f32(<3 x float> %x)
%sub = fsub <3 x float> %x, %floor
%min = tail call <3 x float> @llvm.minnum.v3f32(<3 x float> %sub, <3 x float> <float 0x3FEFFFFFE0000000, float poison, float 0x3FEFFFFFE0000000>)
%uno = fcmp uno <3 x float> %x, <float 0.0, float poison, float 0.0>
%cond = select <3 x i1> %uno, <3 x float> %x, <3 x float> %min
%fabs = tail call <3 x float> @llvm.fabs.v3f32(<3 x float> %x)
%cmpinf = fcmp oeq <3 x float> %fabs, <float 0x7FF0000000000000, float poison, float 0x7FF0000000000000>
%cond6 = select <3 x i1> %cmpinf, <3 x float> <float 0.0, float poison, float 0.0>, <3 x float> %cond
store <3 x float> %floor, ptr addrspace(1) %ip, align 4
ret <3 x float> %cond6
}
define <2 x float> @safe_math_fract_v2f32_const_splat_poison(<2 x float> %x, ptr addrspace(1) writeonly captures(none) %ip) {
; IR-LABEL: define <2 x float> @safe_math_fract_v2f32_const_splat_poison(
; IR-SAME: <2 x float> [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0:[0-9]+]] {
; IR-NEXT: [[FLOOR:%.*]] = tail call <2 x float> @llvm.floor.v2f32(<2 x float> [[X]])
; IR-NEXT: [[SUB:%.*]] = fsub <2 x float> [[X]], [[FLOOR]]
; IR-NEXT: [[MIN:%.*]] = tail call <2 x float> @llvm.minnum.v2f32(<2 x float> [[SUB]], <2 x float> poison)
; IR-NEXT: [[UNO:%.*]] = fcmp uno <2 x float> [[X]], zeroinitializer
; IR-NEXT: [[COND:%.*]] = select <2 x i1> [[UNO]], <2 x float> [[X]], <2 x float> [[MIN]]
; IR-NEXT: [[FABS:%.*]] = tail call <2 x float> @llvm.fabs.v2f32(<2 x float> [[X]])
; IR-NEXT: [[CMPINF:%.*]] = fcmp oeq <2 x float> [[FABS]], splat (float 0x7FF0000000000000)
; IR-NEXT: [[COND6:%.*]] = select <2 x i1> [[CMPINF]], <2 x float> zeroinitializer, <2 x float> [[COND]]
; IR-NEXT: store <2 x float> [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; IR-NEXT: ret <2 x float> [[COND6]]
;
; GFX6-LABEL: safe_math_fract_v2f32_const_splat_poison:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v5, v1
; GFX6-NEXT: v_floor_f32_e32 v4, v0
; GFX6-NEXT: v_sub_f32_e32 v7, v1, v5
; GFX6-NEXT: v_cmp_u_f32_e32 vcc, v1, v1
; GFX6-NEXT: v_sub_f32_e32 v6, v0, v4
; GFX6-NEXT: v_cndmask_b32_e32 v7, v7, v1, vcc
; GFX6-NEXT: v_cmp_u_f32_e32 vcc, v0, v0
; GFX6-NEXT: v_mov_b32_e32 v8, 0x204
; GFX6-NEXT: v_cndmask_b32_e32 v6, v6, v0, vcc
; GFX6-NEXT: v_cmp_class_f32_e32 vcc, v0, v8
; GFX6-NEXT: s_mov_b32 s6, 0
; GFX6-NEXT: v_cndmask_b32_e64 v0, v6, 0, vcc
; GFX6-NEXT: v_cmp_class_f32_e32 vcc, v1, v8
; GFX6-NEXT: s_mov_b32 s7, 0xf000
; GFX6-NEXT: s_mov_b32 s4, s6
; GFX6-NEXT: s_mov_b32 s5, s6
; GFX6-NEXT: v_cndmask_b32_e64 v1, v7, 0, vcc
; GFX6-NEXT: buffer_store_dwordx2 v[4:5], v[2:3], s[4:7], 0 addr64
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: safe_math_fract_v2f32_const_splat_poison:
; GFX7: ; %bb.0:
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_floor_f32_e32 v5, v1
; GFX7-NEXT: v_floor_f32_e32 v4, v0
; GFX7-NEXT: v_sub_f32_e32 v7, v1, v5
; GFX7-NEXT: v_cmp_u_f32_e32 vcc, v1, v1
; GFX7-NEXT: v_sub_f32_e32 v6, v0, v4
; GFX7-NEXT: v_cndmask_b32_e32 v7, v7, v1, vcc
; GFX7-NEXT: v_cmp_u_f32_e32 vcc, v0, v0
; GFX7-NEXT: v_mov_b32_e32 v8, 0x204
; GFX7-NEXT: v_cndmask_b32_e32 v6, v6, v0, vcc
; GFX7-NEXT: v_cmp_class_f32_e32 vcc, v0, v8
; GFX7-NEXT: s_mov_b32 s6, 0
; GFX7-NEXT: v_cndmask_b32_e64 v0, v6, 0, vcc
; GFX7-NEXT: v_cmp_class_f32_e32 vcc, v1, v8
; GFX7-NEXT: s_mov_b32 s7, 0xf000
; GFX7-NEXT: s_mov_b32 s4, s6
; GFX7-NEXT: s_mov_b32 s5, s6
; GFX7-NEXT: v_cndmask_b32_e64 v1, v7, 0, vcc
; GFX7-NEXT: buffer_store_dwordx2 v[4:5], v[2:3], s[4:7], 0 addr64
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: safe_math_fract_v2f32_const_splat_poison:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_floor_f32_e32 v5, v1
; GFX8-NEXT: v_floor_f32_e32 v4, v0
; GFX8-NEXT: v_sub_f32_e32 v7, v1, v5
; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v1, v1
; GFX8-NEXT: v_sub_f32_e32 v6, v0, v4
; GFX8-NEXT: v_cndmask_b32_e32 v7, v7, v1, vcc
; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v0, v0
; GFX8-NEXT: v_mov_b32_e32 v8, 0x204
; GFX8-NEXT: v_cndmask_b32_e32 v6, v6, v0, vcc
; GFX8-NEXT: v_cmp_class_f32_e32 vcc, v0, v8
; GFX8-NEXT: v_cndmask_b32_e64 v0, v6, 0, vcc
; GFX8-NEXT: v_cmp_class_f32_e32 vcc, v1, v8
; GFX8-NEXT: v_cndmask_b32_e64 v1, v7, 0, vcc
; GFX8-NEXT: global_store_dwordx2 v[2:3], v[4:5], off
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: safe_math_fract_v2f32_const_splat_poison:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_floor_f32_e32 v4, v0
; GFX11-NEXT: v_floor_f32_e32 v5, v1
; GFX11-NEXT: v_cmp_u_f32_e32 vcc_lo, v0, v0
; GFX11-NEXT: v_cmp_class_f32_e64 s0, v0, 0x204
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_4) | instid1(VALU_DEP_3)
; GFX11-NEXT: v_dual_sub_f32 v6, v0, v4 :: v_dual_sub_f32 v7, v1, v5
; GFX11-NEXT: global_store_b64 v[2:3], v[4:5], off
; GFX11-NEXT: v_cndmask_b32_e32 v6, v6, v0, vcc_lo
; GFX11-NEXT: v_cmp_u_f32_e32 vcc_lo, v1, v1
; GFX11-NEXT: v_cndmask_b32_e32 v7, v7, v1, vcc_lo
; GFX11-NEXT: v_cndmask_b32_e64 v0, v6, 0, s0
; GFX11-NEXT: v_cmp_class_f32_e64 s0, v1, 0x204
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT: v_cndmask_b32_e64 v1, v7, 0, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: safe_math_fract_v2f32_const_splat_poison:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_floor_f32_e32 v4, v0
; GFX12-NEXT: v_floor_f32_e32 v5, v1
; GFX12-NEXT: v_cmp_u_f32_e32 vcc_lo, v0, v0
; GFX12-NEXT: v_cmp_class_f32_e64 s0, v0, 0x204
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_3)
; GFX12-NEXT: v_dual_sub_f32 v6, v0, v4 :: v_dual_sub_f32 v7, v1, v5
; GFX12-NEXT: global_store_b64 v[2:3], v[4:5], off
; GFX12-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-NEXT: v_cndmask_b32_e32 v6, v6, v0, vcc_lo
; GFX12-NEXT: v_cmp_u_f32_e32 vcc_lo, v1, v1
; GFX12-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-NEXT: v_cndmask_b32_e32 v7, v7, v1, vcc_lo
; GFX12-NEXT: s_wait_alu depctr_va_sdst(0)
; GFX12-NEXT: v_cndmask_b32_e64 v0, v6, 0, s0
; GFX12-NEXT: v_cmp_class_f32_e64 s0, v1, 0x204
; GFX12-NEXT: s_wait_alu depctr_va_sdst(0)
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT: v_cndmask_b32_e64 v1, v7, 0, s0
; GFX12-NEXT: s_setpc_b64 s[30:31]
%floor = tail call <2 x float> @llvm.floor.v2f32(<2 x float> %x)
%sub = fsub <2 x float> %x, %floor
%min = tail call <2 x float> @llvm.minnum.v2f32(<2 x float> %sub, <2 x float> poison)
%uno = fcmp uno <2 x float> %x, zeroinitializer
%cond = select <2 x i1> %uno, <2 x float> %x, <2 x float> %min
%fabs = tail call <2 x float> @llvm.fabs.v2f32(<2 x float> %x)
%cmpinf = fcmp oeq <2 x float> %fabs, splat (float 0x7FF0000000000000)
%cond6 = select <2 x i1> %cmpinf, <2 x float> zeroinitializer, <2 x float> %cond
store <2 x float> %floor, ptr addrspace(1) %ip, align 4
ret <2 x float> %cond6
}
define float @safe_math_fract_f32_swap(float %x, ptr addrspace(1) writeonly captures(none) %ip) {
; GFX6-IR-LABEL: define float @safe_math_fract_f32_swap(
; GFX6-IR-SAME: float [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; GFX6-IR-NEXT: [[UNO:%.*]] = fcmp ord float [[X]], 0.000000e+00
; GFX6-IR-NEXT: [[COND:%.*]] = select i1 [[UNO]], float [[MIN]], float [[X]]
; GFX6-IR-NEXT: [[FABS:%.*]] = tail call float @llvm.fabs.f32(float [[X]])
; GFX6-IR-NEXT: [[CMPINF:%.*]] = fcmp oeq float [[FABS]], 0x7FF0000000000000
; GFX6-IR-NEXT: [[COND6:%.*]] = select i1 [[CMPINF]], float 0.000000e+00, float [[COND]]
; GFX6-IR-NEXT: store float [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; GFX6-IR-NEXT: ret float [[COND6]]
;
; IR-FRACT-LABEL: define float @safe_math_fract_f32_swap(
; IR-FRACT-SAME: float [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; IR-FRACT-NEXT: [[COND:%.*]] = call float @llvm.amdgcn.fract.f32(float [[X]])
; IR-FRACT-NEXT: [[FABS:%.*]] = tail call float @llvm.fabs.f32(float [[X]])
; IR-FRACT-NEXT: [[CMPINF:%.*]] = fcmp oeq float [[FABS]], 0x7FF0000000000000
; IR-FRACT-NEXT: [[COND6:%.*]] = select i1 [[CMPINF]], float 0.000000e+00, float [[COND]]
; IR-FRACT-NEXT: store float [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; IR-FRACT-NEXT: ret float [[COND6]]
;
; GFX6-LABEL: safe_math_fract_f32_swap:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v3, v0
; GFX6-NEXT: v_sub_f32_e32 v4, v0, v3
; GFX6-NEXT: v_min_f32_e32 v4, 0x3f7fffff, v4
; GFX6-NEXT: v_cmp_o_f32_e32 vcc, v0, v0
; GFX6-NEXT: s_mov_b32 s8, 0x7f800000
; GFX6-NEXT: s_mov_b32 s6, 0
; GFX6-NEXT: v_cndmask_b32_e32 v4, v0, v4, vcc
; GFX6-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s8
; GFX6-NEXT: s_mov_b32 s7, 0xf000
; GFX6-NEXT: s_mov_b32 s4, s6
; GFX6-NEXT: s_mov_b32 s5, s6
; GFX6-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc
; GFX6-NEXT: buffer_store_dword v3, v[1:2], s[4:7], 0 addr64
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: safe_math_fract_f32_swap:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: s_mov_b32 s8, 0x7f800000
; GFX7-NEXT: s_mov_b32 s6, 0
; GFX7-NEXT: v_fract_f32_e32 v4, v0
; GFX7-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s8
; GFX7-NEXT: s_mov_b32 s7, 0xf000
; GFX7-NEXT: s_mov_b32 s4, s6
; GFX7-NEXT: s_mov_b32 s5, s6
; GFX7-NEXT: v_floor_f32_e32 v3, v0
; GFX7-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc
; GFX7-NEXT: buffer_store_dword v3, v[1:2], s[4:7], 0 addr64
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: safe_math_fract_f32_swap:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: s_mov_b32 s4, 0x7f800000
; GFX8-NEXT: v_fract_f32_e32 v4, v0
; GFX8-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX8-NEXT: v_floor_f32_e32 v3, v0
; GFX8-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc
; GFX8-NEXT: global_store_dword v[1:2], v3, off
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: safe_math_fract_f32_swap:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v3, v0
; GFX11-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX11-NEXT: v_floor_f32_e32 v4, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_3)
; GFX11-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc_lo
; GFX11-NEXT: global_store_b32 v[1:2], v4, off
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: safe_math_fract_f32_swap:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v3, v0
; GFX12-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX12-NEXT: v_floor_f32_e32 v4, v0
; GFX12-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_3)
; GFX12-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc_lo
; GFX12-NEXT: global_store_b32 v[1:2], v4, off
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = tail call float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
%uno = fcmp ord float %x, 0.000000e+00
%cond = select i1 %uno, float %min, float %x
%fabs = tail call float @llvm.fabs.f32(float %x)
%cmpinf = fcmp oeq float %fabs, 0x7FF0000000000000
%cond6 = select i1 %cmpinf, float 0.000000e+00, float %cond
store float %floor, ptr addrspace(1) %ip, align 4
ret float %cond6
}
define float @safe_math_fract_f32_noinf_check(float %x, ptr addrspace(1) writeonly captures(none) %ip) {
; GFX6-IR-LABEL: define float @safe_math_fract_f32_noinf_check(
; GFX6-IR-SAME: float [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; GFX6-IR-NEXT: [[UNO:%.*]] = fcmp uno float [[X]], 0.000000e+00
; GFX6-IR-NEXT: [[COND:%.*]] = select i1 [[UNO]], float [[X]], float [[MIN]]
; GFX6-IR-NEXT: store float [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; GFX6-IR-NEXT: ret float [[COND]]
;
; IR-FRACT-LABEL: define float @safe_math_fract_f32_noinf_check(
; IR-FRACT-SAME: float [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; IR-FRACT-NEXT: [[COND:%.*]] = call float @llvm.amdgcn.fract.f32(float [[X]])
; IR-FRACT-NEXT: store float [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; IR-FRACT-NEXT: ret float [[COND]]
;
; GFX6-LABEL: safe_math_fract_f32_noinf_check:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v3, v0
; GFX6-NEXT: v_sub_f32_e32 v4, v0, v3
; GFX6-NEXT: s_mov_b32 s6, 0
; GFX6-NEXT: v_min_f32_e32 v4, 0x3f7fffff, v4
; GFX6-NEXT: v_cmp_u_f32_e32 vcc, v0, v0
; GFX6-NEXT: s_mov_b32 s7, 0xf000
; GFX6-NEXT: s_mov_b32 s4, s6
; GFX6-NEXT: s_mov_b32 s5, s6
; GFX6-NEXT: v_cndmask_b32_e32 v0, v4, v0, vcc
; GFX6-NEXT: buffer_store_dword v3, v[1:2], s[4:7], 0 addr64
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: safe_math_fract_f32_noinf_check:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: s_mov_b32 s6, 0
; GFX7-NEXT: s_mov_b32 s7, 0xf000
; GFX7-NEXT: s_mov_b32 s4, s6
; GFX7-NEXT: s_mov_b32 s5, s6
; GFX7-NEXT: v_floor_f32_e32 v3, v0
; GFX7-NEXT: v_fract_f32_e32 v0, v0
; GFX7-NEXT: buffer_store_dword v3, v[1:2], s[4:7], 0 addr64
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: safe_math_fract_f32_noinf_check:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_floor_f32_e32 v3, v0
; GFX8-NEXT: v_fract_f32_e32 v0, v0
; GFX8-NEXT: global_store_dword v[1:2], v3, off
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: safe_math_fract_f32_noinf_check:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_floor_f32_e32 v3, v0
; GFX11-NEXT: v_fract_f32_e32 v0, v0
; GFX11-NEXT: global_store_b32 v[1:2], v3, off
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: safe_math_fract_f32_noinf_check:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_floor_f32_e32 v3, v0
; GFX12-NEXT: v_fract_f32_e32 v0, v0
; GFX12-NEXT: global_store_b32 v[1:2], v3, off
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = tail call float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
%uno = fcmp uno float %x, 0.000000e+00
%cond = select i1 %uno, float %x, float %min
store float %floor, ptr addrspace(1) %ip, align 4
ret float %cond
}
; Cannot match fract without a nan check or no-nans.
define float @no_nan_check_math_fract_f32(float %x, ptr addrspace(1) writeonly captures(none) %ip) {
; IR-LABEL: define float @no_nan_check_math_fract_f32(
; IR-SAME: float [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0]] {
; IR-NEXT: [[ENTRY:.*:]]
; IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; IR-NEXT: [[MIN:%.*]] = tail call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; IR-NEXT: [[FABS:%.*]] = tail call float @llvm.fabs.f32(float [[X]])
; IR-NEXT: [[CMPINF:%.*]] = fcmp oeq float [[FABS]], 0x7FF0000000000000
; IR-NEXT: [[COND6:%.*]] = select i1 [[CMPINF]], float 0.000000e+00, float [[MIN]]
; IR-NEXT: store float [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; IR-NEXT: ret float [[COND6]]
;
; GFX6-LABEL: no_nan_check_math_fract_f32:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v3, v0
; GFX6-NEXT: v_sub_f32_e32 v4, v0, v3
; GFX6-NEXT: s_mov_b32 s8, 0x7f800000
; GFX6-NEXT: s_mov_b32 s6, 0
; GFX6-NEXT: v_min_f32_e32 v4, 0x3f7fffff, v4
; GFX6-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s8
; GFX6-NEXT: s_mov_b32 s7, 0xf000
; GFX6-NEXT: s_mov_b32 s4, s6
; GFX6-NEXT: s_mov_b32 s5, s6
; GFX6-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc
; GFX6-NEXT: buffer_store_dword v3, v[1:2], s[4:7], 0 addr64
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: no_nan_check_math_fract_f32:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_floor_f32_e32 v3, v0
; GFX7-NEXT: v_sub_f32_e32 v4, v0, v3
; GFX7-NEXT: s_mov_b32 s8, 0x7f800000
; GFX7-NEXT: s_mov_b32 s6, 0
; GFX7-NEXT: v_min_f32_e32 v4, 0x3f7fffff, v4
; GFX7-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s8
; GFX7-NEXT: s_mov_b32 s7, 0xf000
; GFX7-NEXT: s_mov_b32 s4, s6
; GFX7-NEXT: s_mov_b32 s5, s6
; GFX7-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc
; GFX7-NEXT: buffer_store_dword v3, v[1:2], s[4:7], 0 addr64
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: no_nan_check_math_fract_f32:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_floor_f32_e32 v3, v0
; GFX8-NEXT: v_sub_f32_e32 v4, v0, v3
; GFX8-NEXT: s_mov_b32 s4, 0x7f800000
; GFX8-NEXT: v_min_f32_e32 v4, 0x3f7fffff, v4
; GFX8-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX8-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc
; GFX8-NEXT: global_store_dword v[1:2], v3, off
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: no_nan_check_math_fract_f32:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_floor_f32_e32 v3, v0
; GFX11-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_sub_f32_e32 v4, v0, v3
; GFX11-NEXT: global_store_b32 v[1:2], v3, off
; GFX11-NEXT: v_min_f32_e32 v4, 0x3f7fffff, v4
; GFX11-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc_lo
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: no_nan_check_math_fract_f32:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_floor_f32_e32 v3, v0
; GFX12-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_3) | instid1(VALU_DEP_1)
; GFX12-NEXT: v_sub_f32_e32 v4, v0, v3
; GFX12-NEXT: global_store_b32 v[1:2], v3, off
; GFX12-NEXT: v_min_num_f32_e32 v4, 0x3f7fffff, v4
; GFX12-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc_lo
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = tail call float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
%fabs = tail call float @llvm.fabs.f32(float %x)
%cmpinf = fcmp oeq float %fabs, 0x7FF0000000000000
%cond6 = select i1 %cmpinf, float 0.000000e+00, float %min
store float %floor, ptr addrspace(1) %ip, align 4
ret float %cond6
}
define float @basic_fract_f32_nonans(float nofpclass(nan) %x) {
; GFX6-IR-LABEL: define float @basic_fract_f32_nonans(
; GFX6-IR-SAME: float nofpclass(nan) [[X:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; GFX6-IR-NEXT: ret float [[MIN]]
;
; IR-FRACT-LABEL: define float @basic_fract_f32_nonans(
; IR-FRACT-SAME: float nofpclass(nan) [[X:%.*]]) #[[ATTR0]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[MIN:%.*]] = call nnan float @llvm.amdgcn.fract.f32(float [[X]])
; IR-FRACT-NEXT: ret float [[MIN]]
;
; GFX6-LABEL: basic_fract_f32_nonans:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX6-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: basic_fract_f32_nonans:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_fract_f32_e32 v0, v0
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: basic_fract_f32_nonans:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_fract_f32_e32 v0, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: basic_fract_f32_nonans:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v0, v0
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: basic_fract_f32_nonans:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v0, v0
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = tail call float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
ret float %min
}
define float @basic_fract_f32_flags_minnum(float %x) {
; IR-LABEL: define float @basic_fract_f32_flags_minnum(
; IR-SAME: float [[X:%.*]]) #[[ATTR0]] {
; IR-NEXT: [[ENTRY:.*:]]
; IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; IR-NEXT: [[MIN:%.*]] = tail call nsz float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; IR-NEXT: ret float [[MIN]]
;
; GFX6-LABEL: basic_fract_f32_flags_minnum:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX6-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: basic_fract_f32_flags_minnum:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_floor_f32_e32 v1, v0
; GFX7-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX7-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: basic_fract_f32_flags_minnum:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_floor_f32_e32 v1, v0
; GFX8-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX8-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: basic_fract_f32_flags_minnum:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_floor_f32_e32 v1, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX11-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: basic_fract_f32_flags_minnum:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_floor_f32_e32 v1, v0
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX12-NEXT: v_min_num_f32_e32 v0, 0x3f7fffff, v0
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = tail call nsz float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
ret float %min
}
define float @basic_fract_f32_flags_fsub(float nofpclass(nan) %x) {
; GFX6-IR-LABEL: define float @basic_fract_f32_flags_fsub(
; GFX6-IR-SAME: float nofpclass(nan) [[X:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub nsz float [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; GFX6-IR-NEXT: ret float [[MIN]]
;
; IR-FRACT-LABEL: define float @basic_fract_f32_flags_fsub(
; IR-FRACT-SAME: float nofpclass(nan) [[X:%.*]]) #[[ATTR0]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[MIN:%.*]] = call nnan float @llvm.amdgcn.fract.f32(float [[X]])
; IR-FRACT-NEXT: ret float [[MIN]]
;
; GFX6-LABEL: basic_fract_f32_flags_fsub:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX6-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: basic_fract_f32_flags_fsub:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_fract_f32_e32 v0, v0
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: basic_fract_f32_flags_fsub:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_fract_f32_e32 v0, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: basic_fract_f32_flags_fsub:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v0, v0
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: basic_fract_f32_flags_fsub:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v0, v0
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %x)
%sub = fsub nsz float %x, %floor
%min = tail call float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
ret float %min
}
define <2 x float> @basic_fract_v2f32_nonans(<2 x float> nofpclass(nan) %x) {
; GFX6-IR-LABEL: define <2 x float> @basic_fract_v2f32_nonans(
; GFX6-IR-SAME: <2 x float> nofpclass(nan) [[X:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call <2 x float> @llvm.floor.v2f32(<2 x float> [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub <2 x float> [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call <2 x float> @llvm.minnum.v2f32(<2 x float> [[SUB]], <2 x float> splat (float 0x3FEFFFFFE0000000))
; GFX6-IR-NEXT: ret <2 x float> [[MIN]]
;
; IR-FRACT-LABEL: define <2 x float> @basic_fract_v2f32_nonans(
; IR-FRACT-SAME: <2 x float> nofpclass(nan) [[X:%.*]]) #[[ATTR0]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[TMP0:%.*]] = extractelement <2 x float> [[X]], i64 0
; IR-FRACT-NEXT: [[TMP1:%.*]] = extractelement <2 x float> [[X]], i64 1
; IR-FRACT-NEXT: [[TMP2:%.*]] = call nnan float @llvm.amdgcn.fract.f32(float [[TMP0]])
; IR-FRACT-NEXT: [[TMP3:%.*]] = call nnan float @llvm.amdgcn.fract.f32(float [[TMP1]])
; IR-FRACT-NEXT: [[TMP4:%.*]] = insertelement <2 x float> poison, float [[TMP2]], i64 0
; IR-FRACT-NEXT: [[MIN:%.*]] = insertelement <2 x float> [[TMP4]], float [[TMP3]], i64 1
; IR-FRACT-NEXT: ret <2 x float> [[MIN]]
;
; GFX6-LABEL: basic_fract_v2f32_nonans:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v2, v0
; GFX6-NEXT: v_floor_f32_e32 v3, v1
; GFX6-NEXT: v_sub_f32_e32 v1, v1, v3
; GFX6-NEXT: v_sub_f32_e32 v0, v0, v2
; GFX6-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX6-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v1
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: basic_fract_v2f32_nonans:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_fract_f32_e32 v0, v0
; GFX7-NEXT: v_fract_f32_e32 v1, v1
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: basic_fract_v2f32_nonans:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_fract_f32_e32 v0, v0
; GFX8-NEXT: v_fract_f32_e32 v1, v1
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: basic_fract_v2f32_nonans:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v0, v0
; GFX11-NEXT: v_fract_f32_e32 v1, v1
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: basic_fract_v2f32_nonans:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v0, v0
; GFX12-NEXT: v_fract_f32_e32 v1, v1
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call <2 x float> @llvm.floor.v2f32(<2 x float> %x)
%sub = fsub <2 x float> %x, %floor
%min = tail call <2 x float> @llvm.minnum.v2f32(<2 x float> %sub, <2 x float> <float 0x3FEFFFFFE0000000, float 0x3FEFFFFFE0000000>)
ret <2 x float> %min
}
define float @basic_fract_f32_multi_use_fsub_nonans(float nofpclass(nan) %x, ptr addrspace(1) %ptr) {
; GFX6-IR-LABEL: define float @basic_fract_f32_multi_use_fsub_nonans(
; GFX6-IR-SAME: float nofpclass(nan) [[X:%.*]], ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; GFX6-IR-NEXT: store float [[SUB]], ptr addrspace(1) [[PTR]], align 4
; GFX6-IR-NEXT: ret float [[MIN]]
;
; IR-FRACT-LABEL: define float @basic_fract_f32_multi_use_fsub_nonans(
; IR-FRACT-SAME: float nofpclass(nan) [[X:%.*]], ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; IR-FRACT-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; IR-FRACT-NEXT: [[MIN:%.*]] = call nnan float @llvm.amdgcn.fract.f32(float [[X]])
; IR-FRACT-NEXT: store float [[SUB]], ptr addrspace(1) [[PTR]], align 4
; IR-FRACT-NEXT: ret float [[MIN]]
;
; GFX6-LABEL: basic_fract_f32_multi_use_fsub_nonans:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v3, v0
; GFX6-NEXT: s_mov_b32 s6, 0
; GFX6-NEXT: v_sub_f32_e32 v3, v0, v3
; GFX6-NEXT: s_mov_b32 s7, 0xf000
; GFX6-NEXT: s_mov_b32 s4, s6
; GFX6-NEXT: s_mov_b32 s5, s6
; GFX6-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v3
; GFX6-NEXT: buffer_store_dword v3, v[1:2], s[4:7], 0 addr64
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: basic_fract_f32_multi_use_fsub_nonans:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: s_mov_b32 s6, 0
; GFX7-NEXT: v_floor_f32_e32 v3, v0
; GFX7-NEXT: s_mov_b32 s7, 0xf000
; GFX7-NEXT: s_mov_b32 s4, s6
; GFX7-NEXT: s_mov_b32 s5, s6
; GFX7-NEXT: v_sub_f32_e32 v3, v0, v3
; GFX7-NEXT: v_fract_f32_e32 v0, v0
; GFX7-NEXT: buffer_store_dword v3, v[1:2], s[4:7], 0 addr64
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: basic_fract_f32_multi_use_fsub_nonans:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_floor_f32_e32 v3, v0
; GFX8-NEXT: v_sub_f32_e32 v3, v0, v3
; GFX8-NEXT: v_fract_f32_e32 v0, v0
; GFX8-NEXT: global_store_dword v[1:2], v3, off
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: basic_fract_f32_multi_use_fsub_nonans:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_floor_f32_e32 v3, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT: v_sub_f32_e32 v3, v0, v3
; GFX11-NEXT: v_fract_f32_e32 v0, v0
; GFX11-NEXT: global_store_b32 v[1:2], v3, off
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: basic_fract_f32_multi_use_fsub_nonans:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_floor_f32_e32 v3, v0
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT: v_sub_f32_e32 v3, v0, v3
; GFX12-NEXT: v_fract_f32_e32 v0, v0
; GFX12-NEXT: global_store_b32 v[1:2], v3, off
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = tail call float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
store float %sub, ptr addrspace(1) %ptr
ret float %min
}
define float @nnan_minnum_fract_f32(float %x) {
; GFX6-IR-LABEL: define float @nnan_minnum_fract_f32(
; GFX6-IR-SAME: float [[X:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call nnan float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; GFX6-IR-NEXT: ret float [[MIN]]
;
; IR-FRACT-LABEL: define float @nnan_minnum_fract_f32(
; IR-FRACT-SAME: float [[X:%.*]]) #[[ATTR0]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[MIN:%.*]] = call nnan float @llvm.amdgcn.fract.f32(float [[X]])
; IR-FRACT-NEXT: ret float [[MIN]]
;
; GFX6-LABEL: nnan_minnum_fract_f32:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX6-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: nnan_minnum_fract_f32:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_fract_f32_e32 v0, v0
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: nnan_minnum_fract_f32:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_fract_f32_e32 v0, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: nnan_minnum_fract_f32:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v0, v0
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: nnan_minnum_fract_f32:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v0, v0
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = tail call nnan float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
ret float %min
}
; TODO: Could match if we checked isKnownNeverNaN on the minnum src
; instead of the pattern input source.
define float @nnan_fsub_fract_f32(float %x) {
; IR-LABEL: define float @nnan_fsub_fract_f32(
; IR-SAME: float [[X:%.*]]) #[[ATTR0]] {
; IR-NEXT: [[ENTRY:.*:]]
; IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; IR-NEXT: [[SUB:%.*]] = fsub nnan float [[X]], [[FLOOR]]
; IR-NEXT: [[MIN:%.*]] = tail call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; IR-NEXT: ret float [[MIN]]
;
; GFX6-LABEL: nnan_fsub_fract_f32:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX6-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: nnan_fsub_fract_f32:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_floor_f32_e32 v1, v0
; GFX7-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX7-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: nnan_fsub_fract_f32:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_floor_f32_e32 v1, v0
; GFX8-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX8-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: nnan_fsub_fract_f32:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_floor_f32_e32 v1, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX11-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: nnan_fsub_fract_f32:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_floor_f32_e32 v1, v0
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX12-NEXT: v_min_num_f32_e32 v0, 0x3f7fffff, v0
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %x)
%sub = fsub nnan float %x, %floor
%min = tail call float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
ret float %min
}
define float @nnan_floor_fract_f32(float %x) {
; IR-LABEL: define float @nnan_floor_fract_f32(
; IR-SAME: float [[X:%.*]]) #[[ATTR0]] {
; IR-NEXT: [[ENTRY:.*:]]
; IR-NEXT: [[FLOOR:%.*]] = tail call nnan float @llvm.floor.f32(float [[X]])
; IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; IR-NEXT: [[MIN:%.*]] = tail call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; IR-NEXT: ret float [[MIN]]
;
; GFX6-LABEL: nnan_floor_fract_f32:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX6-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: nnan_floor_fract_f32:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_floor_f32_e32 v1, v0
; GFX7-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX7-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: nnan_floor_fract_f32:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_floor_f32_e32 v1, v0
; GFX8-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX8-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: nnan_floor_fract_f32:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_floor_f32_e32 v1, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX11-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: nnan_floor_fract_f32:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_floor_f32_e32 v1, v0
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX12-NEXT: v_min_num_f32_e32 v0, 0x3f7fffff, v0
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call nnan float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = tail call float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
ret float %min
}
define float @nnan_src_fract_f32(float nofpclass(nan) %x) {
; GFX6-IR-LABEL: define float @nnan_src_fract_f32(
; GFX6-IR-SAME: float nofpclass(nan) [[X:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; GFX6-IR-NEXT: ret float [[MIN]]
;
; IR-FRACT-LABEL: define float @nnan_src_fract_f32(
; IR-FRACT-SAME: float nofpclass(nan) [[X:%.*]]) #[[ATTR0]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[MIN:%.*]] = call nnan float @llvm.amdgcn.fract.f32(float [[X]])
; IR-FRACT-NEXT: ret float [[MIN]]
;
; GFX6-LABEL: nnan_src_fract_f32:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX6-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: nnan_src_fract_f32:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_fract_f32_e32 v0, v0
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: nnan_src_fract_f32:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_fract_f32_e32 v0, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: nnan_src_fract_f32:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v0, v0
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: nnan_src_fract_f32:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v0, v0
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = tail call float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
ret float %min
}
; Negative test
define float @not_fract_f32_wrong_const(float nofpclass(nan) %x) {
; IR-LABEL: define float @not_fract_f32_wrong_const(
; IR-SAME: float nofpclass(nan) [[X:%.*]]) #[[ATTR0]] {
; IR-NEXT: [[ENTRY:.*:]]
; IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; IR-NEXT: [[MIN:%.*]] = tail call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFC0000000)
; IR-NEXT: ret float [[MIN]]
;
; GFX6-LABEL: not_fract_f32_wrong_const:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX6-NEXT: v_min_f32_e32 v0, 0x3f7ffffe, v0
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: not_fract_f32_wrong_const:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_floor_f32_e32 v1, v0
; GFX7-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX7-NEXT: v_min_f32_e32 v0, 0x3f7ffffe, v0
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: not_fract_f32_wrong_const:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_floor_f32_e32 v1, v0
; GFX8-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX8-NEXT: v_min_f32_e32 v0, 0x3f7ffffe, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: not_fract_f32_wrong_const:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_floor_f32_e32 v1, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX11-NEXT: v_min_f32_e32 v0, 0x3f7ffffe, v0
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: not_fract_f32_wrong_const:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_floor_f32_e32 v1, v0
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX12-NEXT: v_min_num_f32_e32 v0, 0x3f7ffffe, v0
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = tail call float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFC0000000)
ret float %min
}
; Negative test
define float @not_fract_f32_swapped_fsub(float nofpclass(nan) %x) {
; IR-LABEL: define float @not_fract_f32_swapped_fsub(
; IR-SAME: float nofpclass(nan) [[X:%.*]]) #[[ATTR0]] {
; IR-NEXT: [[ENTRY:.*:]]
; IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; IR-NEXT: [[SUB:%.*]] = fsub float [[FLOOR]], [[X]]
; IR-NEXT: [[MIN:%.*]] = tail call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; IR-NEXT: ret float [[MIN]]
;
; GFX6-LABEL: not_fract_f32_swapped_fsub:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v0, v1, v0
; GFX6-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: not_fract_f32_swapped_fsub:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_floor_f32_e32 v1, v0
; GFX7-NEXT: v_sub_f32_e32 v0, v1, v0
; GFX7-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: not_fract_f32_swapped_fsub:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_floor_f32_e32 v1, v0
; GFX8-NEXT: v_sub_f32_e32 v0, v1, v0
; GFX8-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: not_fract_f32_swapped_fsub:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_floor_f32_e32 v1, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_sub_f32_e32 v0, v1, v0
; GFX11-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: not_fract_f32_swapped_fsub:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_floor_f32_e32 v1, v0
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-NEXT: v_sub_f32_e32 v0, v1, v0
; GFX12-NEXT: v_min_num_f32_e32 v0, 0x3f7fffff, v0
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %x)
%sub = fsub float %floor, %x
%min = tail call float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
ret float %min
}
; Negative test
define float @not_fract_f32_not_floor(float nofpclass(nan) %x) {
; IR-LABEL: define float @not_fract_f32_not_floor(
; IR-SAME: float nofpclass(nan) [[X:%.*]]) #[[ATTR0]] {
; IR-NEXT: [[ENTRY:.*:]]
; IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.trunc.f32(float [[X]])
; IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; IR-NEXT: [[MIN:%.*]] = tail call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; IR-NEXT: ret float [[MIN]]
;
; GFX6-LABEL: not_fract_f32_not_floor:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_trunc_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX6-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: not_fract_f32_not_floor:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_trunc_f32_e32 v1, v0
; GFX7-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX7-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: not_fract_f32_not_floor:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_trunc_f32_e32 v1, v0
; GFX8-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX8-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: not_fract_f32_not_floor:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_trunc_f32_e32 v1, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX11-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: not_fract_f32_not_floor:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_trunc_f32_e32 v1, v0
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX12-NEXT: v_min_num_f32_e32 v0, 0x3f7fffff, v0
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.trunc.f32(float %x)
%sub = fsub float %x, %floor
%min = tail call float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
ret float %min
}
; Negative test
define float @not_fract_f32_different_floor(float %x, float %y) {
; IR-LABEL: define float @not_fract_f32_different_floor(
; IR-SAME: float [[X:%.*]], float [[Y:%.*]]) #[[ATTR0]] {
; IR-NEXT: [[ENTRY:.*:]]
; IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[Y]])
; IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; IR-NEXT: [[MIN:%.*]] = tail call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; IR-NEXT: ret float [[MIN]]
;
; GFX6-LABEL: not_fract_f32_different_floor:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v1, v1
; GFX6-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX6-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: not_fract_f32_different_floor:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_floor_f32_e32 v1, v1
; GFX7-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX7-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: not_fract_f32_different_floor:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_floor_f32_e32 v1, v1
; GFX8-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX8-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: not_fract_f32_different_floor:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_floor_f32_e32 v1, v1
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX11-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: not_fract_f32_different_floor:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_floor_f32_e32 v1, v1
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX12-NEXT: v_min_num_f32_e32 v0, 0x3f7fffff, v0
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %y)
%sub = fsub float %x, %floor
%min = tail call float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
ret float %min
}
; Negative test
define float @not_fract_f32_maxnum(float nofpclass(nan) %x) {
; IR-LABEL: define float @not_fract_f32_maxnum(
; IR-SAME: float nofpclass(nan) [[X:%.*]]) #[[ATTR0]] {
; IR-NEXT: [[ENTRY:.*:]]
; IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; IR-NEXT: [[MIN:%.*]] = tail call float @llvm.maxnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; IR-NEXT: ret float [[MIN]]
;
; GFX6-LABEL: not_fract_f32_maxnum:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX6-NEXT: v_max_f32_e32 v0, 0x3f7fffff, v0
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: not_fract_f32_maxnum:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_floor_f32_e32 v1, v0
; GFX7-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX7-NEXT: v_max_f32_e32 v0, 0x3f7fffff, v0
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: not_fract_f32_maxnum:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_floor_f32_e32 v1, v0
; GFX8-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX8-NEXT: v_max_f32_e32 v0, 0x3f7fffff, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: not_fract_f32_maxnum:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_floor_f32_e32 v1, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX11-NEXT: v_max_f32_e32 v0, 0x3f7fffff, v0
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: not_fract_f32_maxnum:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_floor_f32_e32 v1, v0
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX12-NEXT: v_max_num_f32_e32 v0, 0x3f7fffff, v0
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = tail call float @llvm.maxnum.f32(float %sub, float 0x3FEFFFFFE0000000)
ret float %min
}
define float @fcmp_uno_check_is_nan_f32(float %x) {
; IR-LABEL: define float @fcmp_uno_check_is_nan_f32(
; IR-SAME: float [[X:%.*]]) #[[ATTR0]] {
; IR-NEXT: [[ENTRY:.*:]]
; IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; IR-NEXT: [[MIN:%.*]] = tail call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; IR-NEXT: [[UNO:%.*]] = fcmp uno float [[X]], 0x7FF8000000000000
; IR-NEXT: [[COND:%.*]] = select i1 [[UNO]], float [[X]], float [[MIN]]
; IR-NEXT: ret float [[COND]]
;
; GCN-LABEL: fcmp_uno_check_is_nan_f32:
; GCN: ; %bb.0: ; %entry
; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: fcmp_uno_check_is_nan_f32:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = tail call float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
%uno = fcmp uno float %x, 0x7FF8000000000000
%cond = select i1 %uno, float %x, float %min
ret float %cond
}
; No inf check
define float @select_nan_fract_f32(float %x) {
; GFX6-IR-LABEL: define float @select_nan_fract_f32(
; GFX6-IR-SAME: float [[X:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; GFX6-IR-NEXT: [[UNO:%.*]] = fcmp uno float [[X]], 0.000000e+00
; GFX6-IR-NEXT: [[COND:%.*]] = select i1 [[UNO]], float [[X]], float [[MIN]]
; GFX6-IR-NEXT: ret float [[COND]]
;
; IR-FRACT-LABEL: define float @select_nan_fract_f32(
; IR-FRACT-SAME: float [[X:%.*]]) #[[ATTR0]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[COND:%.*]] = call float @llvm.amdgcn.fract.f32(float [[X]])
; IR-FRACT-NEXT: ret float [[COND]]
;
; GFX6-LABEL: select_nan_fract_f32:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v1, v0, v1
; GFX6-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v1
; GFX6-NEXT: v_cmp_u_f32_e32 vcc, v0, v0
; GFX6-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: select_nan_fract_f32:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_fract_f32_e32 v0, v0
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: select_nan_fract_f32:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_fract_f32_e32 v0, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: select_nan_fract_f32:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v0, v0
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: select_nan_fract_f32:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v0, v0
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = tail call float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
%uno = fcmp uno float %x, 0.000000e+00
%cond = select i1 %uno, float %x, float %min
ret float %cond
}
define float @commuted_select_nan_fract_f32(float %x) {
; GFX6-IR-LABEL: define float @commuted_select_nan_fract_f32(
; GFX6-IR-SAME: float [[X:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; GFX6-IR-NEXT: [[UNO:%.*]] = fcmp ord float [[X]], 0.000000e+00
; GFX6-IR-NEXT: [[COND:%.*]] = select i1 [[UNO]], float [[MIN]], float [[X]]
; GFX6-IR-NEXT: ret float [[COND]]
;
; IR-FRACT-LABEL: define float @commuted_select_nan_fract_f32(
; IR-FRACT-SAME: float [[X:%.*]]) #[[ATTR0]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[COND:%.*]] = call float @llvm.amdgcn.fract.f32(float [[X]])
; IR-FRACT-NEXT: ret float [[COND]]
;
; GFX6-LABEL: commuted_select_nan_fract_f32:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v1, v0, v1
; GFX6-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v1
; GFX6-NEXT: v_cmp_o_f32_e32 vcc, v0, v0
; GFX6-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: commuted_select_nan_fract_f32:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_fract_f32_e32 v0, v0
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: commuted_select_nan_fract_f32:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_fract_f32_e32 v0, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: commuted_select_nan_fract_f32:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v0, v0
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: commuted_select_nan_fract_f32:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v0, v0
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = tail call float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
%uno = fcmp ord float %x, 0.000000e+00
%cond = select i1 %uno, float %min, float %x
ret float %cond
}
define float @wrong_commuted_nan_select_f32(float %x) {
; IR-LABEL: define float @wrong_commuted_nan_select_f32(
; IR-SAME: float [[X:%.*]]) #[[ATTR0]] {
; IR-NEXT: [[ENTRY:.*:]]
; IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; IR-NEXT: [[MIN:%.*]] = tail call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; IR-NEXT: [[UNO:%.*]] = fcmp uno float [[X]], 0.000000e+00
; IR-NEXT: [[COND:%.*]] = select i1 [[UNO]], float [[MIN]], float [[X]]
; IR-NEXT: ret float [[COND]]
;
; GFX6-LABEL: wrong_commuted_nan_select_f32:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v1, v0, v1
; GFX6-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v1
; GFX6-NEXT: v_cmp_u_f32_e32 vcc, v0, v0
; GFX6-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: wrong_commuted_nan_select_f32:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_floor_f32_e32 v1, v0
; GFX7-NEXT: v_sub_f32_e32 v1, v0, v1
; GFX7-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v1
; GFX7-NEXT: v_cmp_u_f32_e32 vcc, v0, v0
; GFX7-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: wrong_commuted_nan_select_f32:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_floor_f32_e32 v1, v0
; GFX8-NEXT: v_sub_f32_e32 v1, v0, v1
; GFX8-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v1
; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v0, v0
; GFX8-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: wrong_commuted_nan_select_f32:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_floor_f32_e32 v1, v0
; GFX11-NEXT: v_cmp_u_f32_e32 vcc_lo, v0, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_sub_f32_e32 v1, v0, v1
; GFX11-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v1
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc_lo
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: wrong_commuted_nan_select_f32:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_floor_f32_e32 v1, v0
; GFX12-NEXT: v_cmp_u_f32_e32 vcc_lo, v0, v0
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-NEXT: v_sub_f32_e32 v1, v0, v1
; GFX12-NEXT: v_min_num_f32_e32 v1, 0x3f7fffff, v1
; GFX12-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc_lo
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = tail call float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
%uno = fcmp uno float %x, 0.000000e+00
%cond = select i1 %uno, float %min, float %x
ret float %cond
}
define half @basic_fract_f16_nonan(half nofpclass(nan) %x) {
; GFX6-IR-LABEL: define half @basic_fract_f16_nonan(
; GFX6-IR-SAME: half nofpclass(nan) [[X:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call half @llvm.floor.f16(half [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub half [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call half @llvm.minnum.f16(half [[SUB]], half 0xH3BFF)
; GFX6-IR-NEXT: ret half [[MIN]]
;
; GFX7-IR-LABEL: define half @basic_fract_f16_nonan(
; GFX7-IR-SAME: half nofpclass(nan) [[X:%.*]]) #[[ATTR0]] {
; GFX7-IR-NEXT: [[ENTRY:.*:]]
; GFX7-IR-NEXT: [[FLOOR:%.*]] = tail call half @llvm.floor.f16(half [[X]])
; GFX7-IR-NEXT: [[SUB:%.*]] = fsub half [[X]], [[FLOOR]]
; GFX7-IR-NEXT: [[MIN:%.*]] = tail call half @llvm.minnum.f16(half [[SUB]], half 0xH3BFF)
; GFX7-IR-NEXT: ret half [[MIN]]
;
; IR-LEGALF16-LABEL: define half @basic_fract_f16_nonan(
; IR-LEGALF16-SAME: half nofpclass(nan) [[X:%.*]]) #[[ATTR0]] {
; IR-LEGALF16-NEXT: [[ENTRY:.*:]]
; IR-LEGALF16-NEXT: [[MIN:%.*]] = call nnan half @llvm.amdgcn.fract.f16(half [[X]])
; IR-LEGALF16-NEXT: ret half [[MIN]]
;
; GFX6-LABEL: basic_fract_f16_nonan:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_cvt_f32_f16_e32 v0, v0
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX6-NEXT: v_cvt_f32_f16_e32 v1, v1
; GFX6-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX6-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX6-NEXT: v_cvt_f32_f16_e32 v0, v0
; GFX6-NEXT: v_min_f32_e32 v0, 0x3f7fe000, v0
; GFX6-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: basic_fract_f16_nonan:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_cvt_f32_f16_e32 v0, v0
; GFX7-NEXT: v_floor_f32_e32 v1, v0
; GFX7-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX7-NEXT: v_cvt_f32_f16_e32 v1, v1
; GFX7-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX7-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX7-NEXT: v_cvt_f32_f16_e32 v0, v0
; GFX7-NEXT: v_min_f32_e32 v0, 0x3f7fe000, v0
; GFX7-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: basic_fract_f16_nonan:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_fract_f16_e32 v0, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-TRUE16-LABEL: basic_fract_f16_nonan:
; GFX11-TRUE16: ; %bb.0: ; %entry
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT: v_fract_f16_e32 v0.l, v0.l
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: basic_fract_f16_nonan:
; GFX11-FAKE16: ; %bb.0: ; %entry
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-FAKE16-NEXT: v_fract_f16_e32 v0, v0
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-TRUE16-LABEL: basic_fract_f16_nonan:
; GFX12-TRUE16: ; %bb.0: ; %entry
; GFX12-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-TRUE16-NEXT: s_wait_expcnt 0x0
; GFX12-TRUE16-NEXT: s_wait_samplecnt 0x0
; GFX12-TRUE16-NEXT: s_wait_bvhcnt 0x0
; GFX12-TRUE16-NEXT: s_wait_kmcnt 0x0
; GFX12-TRUE16-NEXT: v_fract_f16_e32 v0.l, v0.l
; GFX12-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-FAKE16-LABEL: basic_fract_f16_nonan:
; GFX12-FAKE16: ; %bb.0: ; %entry
; GFX12-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-FAKE16-NEXT: s_wait_expcnt 0x0
; GFX12-FAKE16-NEXT: s_wait_samplecnt 0x0
; GFX12-FAKE16-NEXT: s_wait_bvhcnt 0x0
; GFX12-FAKE16-NEXT: s_wait_kmcnt 0x0
; GFX12-FAKE16-NEXT: v_fract_f16_e32 v0, v0
; GFX12-FAKE16-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call half @llvm.floor.f16(half %x)
%sub = fsub half %x, %floor
%min = tail call half @llvm.minnum.f16(half %sub, half 0xH3BFF)
ret half %min
}
define <2 x half> @basic_fract_v2f16_nonan(<2 x half> nofpclass(nan) %x) {
; GFX6-IR-LABEL: define <2 x half> @basic_fract_v2f16_nonan(
; GFX6-IR-SAME: <2 x half> nofpclass(nan) [[X:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call <2 x half> @llvm.floor.v2f16(<2 x half> [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub <2 x half> [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call <2 x half> @llvm.minnum.v2f16(<2 x half> [[SUB]], <2 x half> splat (half 0xH3BFF))
; GFX6-IR-NEXT: ret <2 x half> [[MIN]]
;
; GFX7-IR-LABEL: define <2 x half> @basic_fract_v2f16_nonan(
; GFX7-IR-SAME: <2 x half> nofpclass(nan) [[X:%.*]]) #[[ATTR0]] {
; GFX7-IR-NEXT: [[ENTRY:.*:]]
; GFX7-IR-NEXT: [[FLOOR:%.*]] = tail call <2 x half> @llvm.floor.v2f16(<2 x half> [[X]])
; GFX7-IR-NEXT: [[SUB:%.*]] = fsub <2 x half> [[X]], [[FLOOR]]
; GFX7-IR-NEXT: [[MIN:%.*]] = tail call <2 x half> @llvm.minnum.v2f16(<2 x half> [[SUB]], <2 x half> splat (half 0xH3BFF))
; GFX7-IR-NEXT: ret <2 x half> [[MIN]]
;
; IR-LEGALF16-LABEL: define <2 x half> @basic_fract_v2f16_nonan(
; IR-LEGALF16-SAME: <2 x half> nofpclass(nan) [[X:%.*]]) #[[ATTR0]] {
; IR-LEGALF16-NEXT: [[ENTRY:.*:]]
; IR-LEGALF16-NEXT: [[TMP0:%.*]] = extractelement <2 x half> [[X]], i64 0
; IR-LEGALF16-NEXT: [[TMP1:%.*]] = extractelement <2 x half> [[X]], i64 1
; IR-LEGALF16-NEXT: [[TMP2:%.*]] = call nnan half @llvm.amdgcn.fract.f16(half [[TMP0]])
; IR-LEGALF16-NEXT: [[TMP3:%.*]] = call nnan half @llvm.amdgcn.fract.f16(half [[TMP1]])
; IR-LEGALF16-NEXT: [[TMP4:%.*]] = insertelement <2 x half> poison, half [[TMP2]], i64 0
; IR-LEGALF16-NEXT: [[MIN:%.*]] = insertelement <2 x half> [[TMP4]], half [[TMP3]], i64 1
; IR-LEGALF16-NEXT: ret <2 x half> [[MIN]]
;
; GFX6-LABEL: basic_fract_v2f16_nonan:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_lshrrev_b32_e32 v1, 16, v0
; GFX6-NEXT: v_cvt_f32_f16_e32 v1, v1
; GFX6-NEXT: v_cvt_f32_f16_e32 v0, v0
; GFX6-NEXT: v_floor_f32_e32 v2, v1
; GFX6-NEXT: v_cvt_f16_f32_e32 v2, v2
; GFX6-NEXT: v_floor_f32_e32 v3, v0
; GFX6-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX6-NEXT: v_cvt_f32_f16_e32 v2, v2
; GFX6-NEXT: v_cvt_f32_f16_e32 v3, v3
; GFX6-NEXT: v_sub_f32_e32 v1, v1, v2
; GFX6-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX6-NEXT: v_sub_f32_e32 v0, v0, v3
; GFX6-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX6-NEXT: v_cvt_f32_f16_e32 v1, v1
; GFX6-NEXT: v_cvt_f32_f16_e32 v0, v0
; GFX6-NEXT: v_min_f32_e32 v1, 0x3f7fe000, v1
; GFX6-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX6-NEXT: v_min_f32_e32 v0, 0x3f7fe000, v0
; GFX6-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX6-NEXT: v_lshlrev_b32_e32 v1, 16, v1
; GFX6-NEXT: v_or_b32_e32 v0, v0, v1
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: basic_fract_v2f16_nonan:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_lshrrev_b32_e32 v1, 16, v0
; GFX7-NEXT: v_cvt_f32_f16_e32 v1, v1
; GFX7-NEXT: v_cvt_f32_f16_e32 v0, v0
; GFX7-NEXT: v_floor_f32_e32 v2, v1
; GFX7-NEXT: v_cvt_f16_f32_e32 v2, v2
; GFX7-NEXT: v_floor_f32_e32 v3, v0
; GFX7-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX7-NEXT: v_cvt_f32_f16_e32 v2, v2
; GFX7-NEXT: v_cvt_f32_f16_e32 v3, v3
; GFX7-NEXT: v_sub_f32_e32 v1, v1, v2
; GFX7-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX7-NEXT: v_sub_f32_e32 v0, v0, v3
; GFX7-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX7-NEXT: v_cvt_f32_f16_e32 v1, v1
; GFX7-NEXT: v_cvt_f32_f16_e32 v0, v0
; GFX7-NEXT: v_min_f32_e32 v1, 0x3f7fe000, v1
; GFX7-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX7-NEXT: v_min_f32_e32 v0, 0x3f7fe000, v0
; GFX7-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX7-NEXT: v_lshlrev_b32_e32 v1, 16, v1
; GFX7-NEXT: v_or_b32_e32 v0, v0, v1
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: basic_fract_v2f16_nonan:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_fract_f16_e32 v1, v0
; GFX8-NEXT: v_fract_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX8-NEXT: v_pack_b32_f16 v0, v1, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-TRUE16-LABEL: basic_fract_v2f16_nonan:
; GFX11-TRUE16: ; %bb.0: ; %entry
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT: v_fract_f16_e32 v0.l, v0.l
; GFX11-TRUE16-NEXT: v_fract_f16_e32 v0.h, v0.h
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: basic_fract_v2f16_nonan:
; GFX11-FAKE16: ; %bb.0: ; %entry
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v1, 16, v0
; GFX11-FAKE16-NEXT: v_fract_f16_e32 v0, v0
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-FAKE16-NEXT: v_fract_f16_e32 v1, v1
; GFX11-FAKE16-NEXT: v_pack_b32_f16 v0, v0, v1
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-TRUE16-LABEL: basic_fract_v2f16_nonan:
; GFX12-TRUE16: ; %bb.0: ; %entry
; GFX12-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-TRUE16-NEXT: s_wait_expcnt 0x0
; GFX12-TRUE16-NEXT: s_wait_samplecnt 0x0
; GFX12-TRUE16-NEXT: s_wait_bvhcnt 0x0
; GFX12-TRUE16-NEXT: s_wait_kmcnt 0x0
; GFX12-TRUE16-NEXT: v_fract_f16_e32 v0.l, v0.l
; GFX12-TRUE16-NEXT: v_fract_f16_e32 v0.h, v0.h
; GFX12-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-FAKE16-LABEL: basic_fract_v2f16_nonan:
; GFX12-FAKE16: ; %bb.0: ; %entry
; GFX12-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-FAKE16-NEXT: s_wait_expcnt 0x0
; GFX12-FAKE16-NEXT: s_wait_samplecnt 0x0
; GFX12-FAKE16-NEXT: s_wait_bvhcnt 0x0
; GFX12-FAKE16-NEXT: s_wait_kmcnt 0x0
; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v1, 16, v0
; GFX12-FAKE16-NEXT: v_fract_f16_e32 v0, v0
; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-FAKE16-NEXT: v_fract_f16_e32 v1, v1
; GFX12-FAKE16-NEXT: v_pack_b32_f16 v0, v0, v1
; GFX12-FAKE16-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call <2 x half> @llvm.floor.v2f16(<2 x half> %x)
%sub = fsub <2 x half> %x, %floor
%min = tail call <2 x half> @llvm.minnum.v2f16(<2 x half> %sub, <2 x half> <half 0xH3BFF, half 0xH3BFF>)
ret <2 x half> %min
}
define double @basic_fract_f64_nanans(double nofpclass(nan) %x) {
; GFX6-IR-LABEL: define double @basic_fract_f64_nanans(
; GFX6-IR-SAME: double nofpclass(nan) [[X:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call double @llvm.floor.f64(double [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub double [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call double @llvm.minnum.f64(double [[SUB]], double 0x3FEFFFFFFFFFFFFF)
; GFX6-IR-NEXT: ret double [[MIN]]
;
; IR-FRACT-LABEL: define double @basic_fract_f64_nanans(
; IR-FRACT-SAME: double nofpclass(nan) [[X:%.*]]) #[[ATTR0]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[MIN:%.*]] = call nnan double @llvm.amdgcn.fract.f64(double [[X]])
; IR-FRACT-NEXT: ret double [[MIN]]
;
; GFX6-LABEL: basic_fract_f64_nanans:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_fract_f64_e32 v[2:3], v[0:1]
; GFX6-NEXT: v_mov_b32_e32 v4, -1
; GFX6-NEXT: v_mov_b32_e32 v5, 0x3fefffff
; GFX6-NEXT: v_min_f64 v[2:3], v[2:3], v[4:5]
; GFX6-NEXT: v_cmp_class_f64_e64 vcc, v[0:1], 3
; GFX6-NEXT: s_mov_b32 s4, -1
; GFX6-NEXT: v_cndmask_b32_e32 v2, v2, v0, vcc
; GFX6-NEXT: v_cndmask_b32_e32 v3, v3, v1, vcc
; GFX6-NEXT: v_add_f64 v[2:3], v[0:1], -v[2:3]
; GFX6-NEXT: s_mov_b32 s5, 0x3fefffff
; GFX6-NEXT: v_add_f64 v[0:1], v[0:1], -v[2:3]
; GFX6-NEXT: v_min_f64 v[0:1], v[0:1], s[4:5]
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: basic_fract_f64_nanans:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_fract_f64_e32 v[0:1], v[0:1]
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: basic_fract_f64_nanans:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_fract_f64_e32 v[0:1], v[0:1]
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: basic_fract_f64_nanans:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f64_e32 v[0:1], v[0:1]
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: basic_fract_f64_nanans:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f64_e32 v[0:1], v[0:1]
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call double @llvm.floor.f64(double %x)
%sub = fsub double %x, %floor
%min = tail call double @llvm.minnum.f64(double %sub, double 0x3FEFFFFFFFFFFFFF)
ret double %min
}
define half @safe_math_fract_f16_noinf_check(half %x, ptr addrspace(1) writeonly captures(none) %ip) {
; GFX6-IR-LABEL: define half @safe_math_fract_f16_noinf_check(
; GFX6-IR-SAME: half [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call half @llvm.floor.f16(half [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub half [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call half @llvm.minnum.f16(half [[SUB]], half 0xH3BFF)
; GFX6-IR-NEXT: [[UNO:%.*]] = fcmp uno half [[X]], 0xH0000
; GFX6-IR-NEXT: [[COND:%.*]] = select i1 [[UNO]], half [[X]], half [[MIN]]
; GFX6-IR-NEXT: store half [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; GFX6-IR-NEXT: ret half [[COND]]
;
; GFX7-IR-LABEL: define half @safe_math_fract_f16_noinf_check(
; GFX7-IR-SAME: half [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0]] {
; GFX7-IR-NEXT: [[ENTRY:.*:]]
; GFX7-IR-NEXT: [[FLOOR:%.*]] = tail call half @llvm.floor.f16(half [[X]])
; GFX7-IR-NEXT: [[SUB:%.*]] = fsub half [[X]], [[FLOOR]]
; GFX7-IR-NEXT: [[MIN:%.*]] = tail call half @llvm.minnum.f16(half [[SUB]], half 0xH3BFF)
; GFX7-IR-NEXT: [[UNO:%.*]] = fcmp uno half [[X]], 0xH0000
; GFX7-IR-NEXT: [[COND:%.*]] = select i1 [[UNO]], half [[X]], half [[MIN]]
; GFX7-IR-NEXT: store half [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; GFX7-IR-NEXT: ret half [[COND]]
;
; IR-LEGALF16-LABEL: define half @safe_math_fract_f16_noinf_check(
; IR-LEGALF16-SAME: half [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0]] {
; IR-LEGALF16-NEXT: [[ENTRY:.*:]]
; IR-LEGALF16-NEXT: [[FLOOR:%.*]] = tail call half @llvm.floor.f16(half [[X]])
; IR-LEGALF16-NEXT: [[COND:%.*]] = call half @llvm.amdgcn.fract.f16(half [[X]])
; IR-LEGALF16-NEXT: store half [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; IR-LEGALF16-NEXT: ret half [[COND]]
;
; GFX6-LABEL: safe_math_fract_f16_noinf_check:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_cvt_f32_f16_e32 v3, v0
; GFX6-NEXT: s_mov_b32 s6, 0
; GFX6-NEXT: s_mov_b32 s7, 0xf000
; GFX6-NEXT: s_mov_b32 s4, s6
; GFX6-NEXT: v_floor_f32_e32 v4, v3
; GFX6-NEXT: v_cvt_f16_f32_e32 v4, v4
; GFX6-NEXT: v_cmp_u_f32_e32 vcc, v3, v3
; GFX6-NEXT: s_mov_b32 s5, s6
; GFX6-NEXT: v_cvt_f32_f16_e32 v5, v4
; GFX6-NEXT: buffer_store_short v4, v[1:2], s[4:7], 0 addr64
; GFX6-NEXT: v_sub_f32_e32 v5, v3, v5
; GFX6-NEXT: v_cvt_f16_f32_e32 v5, v5
; GFX6-NEXT: v_cvt_f32_f16_e32 v5, v5
; GFX6-NEXT: v_min_f32_e32 v5, 0x3f7fe000, v5
; GFX6-NEXT: v_cvt_f16_f32_e32 v5, v5
; GFX6-NEXT: v_cndmask_b32_e32 v0, v5, v0, vcc
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: safe_math_fract_f16_noinf_check:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_cvt_f32_f16_e32 v3, v0
; GFX7-NEXT: s_mov_b32 s6, 0
; GFX7-NEXT: s_mov_b32 s7, 0xf000
; GFX7-NEXT: s_mov_b32 s4, s6
; GFX7-NEXT: v_floor_f32_e32 v4, v3
; GFX7-NEXT: v_cvt_f16_f32_e32 v4, v4
; GFX7-NEXT: v_cmp_u_f32_e32 vcc, v3, v3
; GFX7-NEXT: s_mov_b32 s5, s6
; GFX7-NEXT: v_cvt_f32_f16_e32 v5, v4
; GFX7-NEXT: buffer_store_short v4, v[1:2], s[4:7], 0 addr64
; GFX7-NEXT: v_sub_f32_e32 v5, v3, v5
; GFX7-NEXT: v_cvt_f16_f32_e32 v5, v5
; GFX7-NEXT: v_cvt_f32_f16_e32 v5, v5
; GFX7-NEXT: v_min_f32_e32 v5, 0x3f7fe000, v5
; GFX7-NEXT: v_cvt_f16_f32_e32 v5, v5
; GFX7-NEXT: v_cndmask_b32_e32 v0, v5, v0, vcc
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: safe_math_fract_f16_noinf_check:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_floor_f16_e32 v3, v0
; GFX8-NEXT: v_fract_f16_e32 v0, v0
; GFX8-NEXT: global_store_short v[1:2], v3, off
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-TRUE16-LABEL: safe_math_fract_f16_noinf_check:
; GFX11-TRUE16: ; %bb.0: ; %entry
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT: v_floor_f16_e32 v0.h, v0.l
; GFX11-TRUE16-NEXT: v_fract_f16_e32 v0.l, v0.l
; GFX11-TRUE16-NEXT: global_store_d16_hi_b16 v[1:2], v0, off
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: safe_math_fract_f16_noinf_check:
; GFX11-FAKE16: ; %bb.0: ; %entry
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-FAKE16-NEXT: v_floor_f16_e32 v3, v0
; GFX11-FAKE16-NEXT: v_fract_f16_e32 v0, v0
; GFX11-FAKE16-NEXT: global_store_b16 v[1:2], v3, off
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-TRUE16-LABEL: safe_math_fract_f16_noinf_check:
; GFX12-TRUE16: ; %bb.0: ; %entry
; GFX12-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-TRUE16-NEXT: s_wait_expcnt 0x0
; GFX12-TRUE16-NEXT: s_wait_samplecnt 0x0
; GFX12-TRUE16-NEXT: s_wait_bvhcnt 0x0
; GFX12-TRUE16-NEXT: s_wait_kmcnt 0x0
; GFX12-TRUE16-NEXT: v_floor_f16_e32 v0.h, v0.l
; GFX12-TRUE16-NEXT: v_fract_f16_e32 v0.l, v0.l
; GFX12-TRUE16-NEXT: global_store_d16_hi_b16 v[1:2], v0, off
; GFX12-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-FAKE16-LABEL: safe_math_fract_f16_noinf_check:
; GFX12-FAKE16: ; %bb.0: ; %entry
; GFX12-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-FAKE16-NEXT: s_wait_expcnt 0x0
; GFX12-FAKE16-NEXT: s_wait_samplecnt 0x0
; GFX12-FAKE16-NEXT: s_wait_bvhcnt 0x0
; GFX12-FAKE16-NEXT: s_wait_kmcnt 0x0
; GFX12-FAKE16-NEXT: v_floor_f16_e32 v3, v0
; GFX12-FAKE16-NEXT: v_fract_f16_e32 v0, v0
; GFX12-FAKE16-NEXT: global_store_b16 v[1:2], v3, off
; GFX12-FAKE16-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call half @llvm.floor.f16(half %x)
%sub = fsub half %x, %floor
%min = tail call half @llvm.minnum.f16(half %sub, half 0xH3BFF)
%uno = fcmp uno half %x, 0.000000e+00
%cond = select i1 %uno, half %x, half %min
store half %floor, ptr addrspace(1) %ip, align 4
ret half %cond
}
define double @safe_math_fract_f64_noinf_check(double %x, ptr addrspace(1) writeonly captures(none) %ip) {
; GFX6-IR-LABEL: define double @safe_math_fract_f64_noinf_check(
; GFX6-IR-SAME: double [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call double @llvm.floor.f64(double [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub double [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call double @llvm.minnum.f64(double [[SUB]], double 0x3FEFFFFFFFFFFFFF)
; GFX6-IR-NEXT: [[UNO:%.*]] = fcmp uno double [[X]], 0.000000e+00
; GFX6-IR-NEXT: [[COND:%.*]] = select i1 [[UNO]], double [[X]], double [[MIN]]
; GFX6-IR-NEXT: store double [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; GFX6-IR-NEXT: ret double [[COND]]
;
; IR-FRACT-LABEL: define double @safe_math_fract_f64_noinf_check(
; IR-FRACT-SAME: double [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[FLOOR:%.*]] = tail call double @llvm.floor.f64(double [[X]])
; IR-FRACT-NEXT: [[COND:%.*]] = call double @llvm.amdgcn.fract.f64(double [[X]])
; IR-FRACT-NEXT: store double [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; IR-FRACT-NEXT: ret double [[COND]]
;
; GFX6-LABEL: safe_math_fract_f64_noinf_check:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_fract_f64_e32 v[4:5], v[0:1]
; GFX6-NEXT: v_mov_b32_e32 v6, -1
; GFX6-NEXT: v_mov_b32_e32 v7, 0x3fefffff
; GFX6-NEXT: v_min_f64 v[4:5], v[4:5], v[6:7]
; GFX6-NEXT: v_cmp_class_f64_e64 vcc, v[0:1], 3
; GFX6-NEXT: s_mov_b32 s8, -1
; GFX6-NEXT: v_cndmask_b32_e32 v4, v4, v0, vcc
; GFX6-NEXT: v_cndmask_b32_e32 v5, v5, v1, vcc
; GFX6-NEXT: v_add_f64 v[4:5], v[0:1], -v[4:5]
; GFX6-NEXT: s_mov_b32 s9, 0x3fefffff
; GFX6-NEXT: v_add_f64 v[6:7], v[0:1], -v[4:5]
; GFX6-NEXT: v_cmp_u_f64_e32 vcc, v[0:1], v[0:1]
; GFX6-NEXT: v_min_f64 v[6:7], v[6:7], s[8:9]
; GFX6-NEXT: s_mov_b32 s6, 0
; GFX6-NEXT: s_mov_b32 s7, 0xf000
; GFX6-NEXT: s_mov_b32 s4, s6
; GFX6-NEXT: s_mov_b32 s5, s6
; GFX6-NEXT: v_cndmask_b32_e32 v0, v6, v0, vcc
; GFX6-NEXT: v_cndmask_b32_e32 v1, v7, v1, vcc
; GFX6-NEXT: buffer_store_dwordx2 v[4:5], v[2:3], s[4:7], 0 addr64
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: safe_math_fract_f64_noinf_check:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_floor_f64_e32 v[4:5], v[0:1]
; GFX7-NEXT: v_fract_f64_e32 v[0:1], v[0:1]
; GFX7-NEXT: s_mov_b32 s6, 0
; GFX7-NEXT: s_mov_b32 s7, 0xf000
; GFX7-NEXT: s_mov_b32 s4, s6
; GFX7-NEXT: s_mov_b32 s5, s6
; GFX7-NEXT: buffer_store_dwordx2 v[4:5], v[2:3], s[4:7], 0 addr64
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: safe_math_fract_f64_noinf_check:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_floor_f64_e32 v[4:5], v[0:1]
; GFX8-NEXT: v_fract_f64_e32 v[0:1], v[0:1]
; GFX8-NEXT: global_store_dwordx2 v[2:3], v[4:5], off
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: safe_math_fract_f64_noinf_check:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_floor_f64_e32 v[4:5], v[0:1]
; GFX11-NEXT: v_fract_f64_e32 v[0:1], v[0:1]
; GFX11-NEXT: global_store_b64 v[2:3], v[4:5], off
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: safe_math_fract_f64_noinf_check:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_floor_f64_e32 v[4:5], v[0:1]
; GFX12-NEXT: v_fract_f64_e32 v[0:1], v[0:1]
; GFX12-NEXT: global_store_b64 v[2:3], v[4:5], off
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call double @llvm.floor.f64(double %x)
%sub = fsub double %x, %floor
%min = tail call double @llvm.minnum.f64(double %sub, double 0x3FEFFFFFFFFFFFFF)
%uno = fcmp uno double %x, 0.000000e+00
%cond = select i1 %uno, double %x, double %min
store double %floor, ptr addrspace(1) %ip, align 4
ret double %cond
}
define float @select_nan_fract_f32_flags_select(float %x) {
; GFX6-IR-LABEL: define float @select_nan_fract_f32_flags_select(
; GFX6-IR-SAME: float [[X:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; GFX6-IR-NEXT: [[UNO:%.*]] = fcmp uno float [[X]], 0.000000e+00
; GFX6-IR-NEXT: [[COND:%.*]] = select nsz i1 [[UNO]], float [[X]], float [[MIN]]
; GFX6-IR-NEXT: ret float [[COND]]
;
; IR-FRACT-LABEL: define float @select_nan_fract_f32_flags_select(
; IR-FRACT-SAME: float [[X:%.*]]) #[[ATTR0]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[COND:%.*]] = call nsz float @llvm.amdgcn.fract.f32(float [[X]])
; IR-FRACT-NEXT: ret float [[COND]]
;
; GFX6-LABEL: select_nan_fract_f32_flags_select:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v1, v0, v1
; GFX6-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v1
; GFX6-NEXT: v_cmp_u_f32_e32 vcc, v0, v0
; GFX6-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: select_nan_fract_f32_flags_select:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_fract_f32_e32 v0, v0
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: select_nan_fract_f32_flags_select:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_fract_f32_e32 v0, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: select_nan_fract_f32_flags_select:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v0, v0
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: select_nan_fract_f32_flags_select:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v0, v0
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = tail call float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
%uno = fcmp uno float %x, 0.000000e+00
%cond = select nsz i1 %uno, float %x, float %min
ret float %cond
}
define float @select_nan_fract_f32_flags_minnum(float %x) {
; GFX6-IR-LABEL: define float @select_nan_fract_f32_flags_minnum(
; GFX6-IR-SAME: float [[X:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call nsz float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; GFX6-IR-NEXT: [[UNO:%.*]] = fcmp uno float [[X]], 0.000000e+00
; GFX6-IR-NEXT: [[COND:%.*]] = select i1 [[UNO]], float [[X]], float [[MIN]]
; GFX6-IR-NEXT: ret float [[COND]]
;
; IR-FRACT-LABEL: define float @select_nan_fract_f32_flags_minnum(
; IR-FRACT-SAME: float [[X:%.*]]) #[[ATTR0]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[COND:%.*]] = call float @llvm.amdgcn.fract.f32(float [[X]])
; IR-FRACT-NEXT: ret float [[COND]]
;
; GFX6-LABEL: select_nan_fract_f32_flags_minnum:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v1, v0, v1
; GFX6-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v1
; GFX6-NEXT: v_cmp_u_f32_e32 vcc, v0, v0
; GFX6-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: select_nan_fract_f32_flags_minnum:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_fract_f32_e32 v0, v0
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: select_nan_fract_f32_flags_minnum:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_fract_f32_e32 v0, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: select_nan_fract_f32_flags_minnum:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v0, v0
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: select_nan_fract_f32_flags_minnum:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v0, v0
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = tail call nsz float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
%uno = fcmp uno float %x, 0.000000e+00
%cond = select i1 %uno, float %x, float %min
ret float %cond
}
define <2 x float> @safe_math_fract_v2f32(<2 x float> %x, ptr addrspace(1) writeonly captures(none) %ip) {
; GFX6-IR-LABEL: define <2 x float> @safe_math_fract_v2f32(
; GFX6-IR-SAME: <2 x float> [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call <2 x float> @llvm.floor.v2f32(<2 x float> [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub <2 x float> [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call <2 x float> @llvm.minnum.v2f32(<2 x float> [[SUB]], <2 x float> splat (float 0x3FEFFFFFE0000000))
; GFX6-IR-NEXT: [[UNO:%.*]] = fcmp uno <2 x float> [[X]], zeroinitializer
; GFX6-IR-NEXT: [[COND:%.*]] = select <2 x i1> [[UNO]], <2 x float> [[X]], <2 x float> [[MIN]]
; GFX6-IR-NEXT: [[FABS:%.*]] = tail call <2 x float> @llvm.fabs.v2f32(<2 x float> [[X]])
; GFX6-IR-NEXT: [[CMPINF:%.*]] = fcmp oeq <2 x float> [[FABS]], splat (float 0x7FF0000000000000)
; GFX6-IR-NEXT: [[COND6:%.*]] = select <2 x i1> [[CMPINF]], <2 x float> zeroinitializer, <2 x float> [[COND]]
; GFX6-IR-NEXT: store <2 x float> [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; GFX6-IR-NEXT: ret <2 x float> [[COND6]]
;
; IR-FRACT-LABEL: define <2 x float> @safe_math_fract_v2f32(
; IR-FRACT-SAME: <2 x float> [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[FLOOR:%.*]] = tail call <2 x float> @llvm.floor.v2f32(<2 x float> [[X]])
; IR-FRACT-NEXT: [[TMP0:%.*]] = extractelement <2 x float> [[X]], i64 0
; IR-FRACT-NEXT: [[TMP1:%.*]] = extractelement <2 x float> [[X]], i64 1
; IR-FRACT-NEXT: [[TMP2:%.*]] = call float @llvm.amdgcn.fract.f32(float [[TMP0]])
; IR-FRACT-NEXT: [[TMP3:%.*]] = call float @llvm.amdgcn.fract.f32(float [[TMP1]])
; IR-FRACT-NEXT: [[TMP4:%.*]] = insertelement <2 x float> poison, float [[TMP2]], i64 0
; IR-FRACT-NEXT: [[COND:%.*]] = insertelement <2 x float> [[TMP4]], float [[TMP3]], i64 1
; IR-FRACT-NEXT: [[FABS:%.*]] = tail call <2 x float> @llvm.fabs.v2f32(<2 x float> [[X]])
; IR-FRACT-NEXT: [[CMPINF:%.*]] = fcmp oeq <2 x float> [[FABS]], splat (float 0x7FF0000000000000)
; IR-FRACT-NEXT: [[COND6:%.*]] = select <2 x i1> [[CMPINF]], <2 x float> zeroinitializer, <2 x float> [[COND]]
; IR-FRACT-NEXT: store <2 x float> [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; IR-FRACT-NEXT: ret <2 x float> [[COND6]]
;
; GFX6-LABEL: safe_math_fract_v2f32:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v5, v1
; GFX6-NEXT: v_floor_f32_e32 v4, v0
; GFX6-NEXT: v_sub_f32_e32 v6, v1, v5
; GFX6-NEXT: v_sub_f32_e32 v7, v0, v4
; GFX6-NEXT: v_min_f32_e32 v6, 0x3f7fffff, v6
; GFX6-NEXT: v_cmp_u_f32_e32 vcc, v1, v1
; GFX6-NEXT: v_min_f32_e32 v7, 0x3f7fffff, v7
; GFX6-NEXT: v_cndmask_b32_e32 v6, v6, v1, vcc
; GFX6-NEXT: v_cmp_u_f32_e32 vcc, v0, v0
; GFX6-NEXT: v_mov_b32_e32 v8, 0x204
; GFX6-NEXT: v_cndmask_b32_e32 v7, v7, v0, vcc
; GFX6-NEXT: v_cmp_class_f32_e32 vcc, v0, v8
; GFX6-NEXT: s_mov_b32 s6, 0
; GFX6-NEXT: v_cndmask_b32_e64 v0, v7, 0, vcc
; GFX6-NEXT: v_cmp_class_f32_e32 vcc, v1, v8
; GFX6-NEXT: s_mov_b32 s7, 0xf000
; GFX6-NEXT: s_mov_b32 s4, s6
; GFX6-NEXT: s_mov_b32 s5, s6
; GFX6-NEXT: v_cndmask_b32_e64 v1, v6, 0, vcc
; GFX6-NEXT: buffer_store_dwordx2 v[4:5], v[2:3], s[4:7], 0 addr64
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: safe_math_fract_v2f32:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_mov_b32_e32 v8, 0x204
; GFX7-NEXT: v_fract_f32_e32 v6, v0
; GFX7-NEXT: v_cmp_class_f32_e32 vcc, v0, v8
; GFX7-NEXT: s_mov_b32 s6, 0
; GFX7-NEXT: v_floor_f32_e32 v4, v0
; GFX7-NEXT: v_fract_f32_e32 v7, v1
; GFX7-NEXT: v_cndmask_b32_e64 v0, v6, 0, vcc
; GFX7-NEXT: v_cmp_class_f32_e32 vcc, v1, v8
; GFX7-NEXT: s_mov_b32 s7, 0xf000
; GFX7-NEXT: s_mov_b32 s4, s6
; GFX7-NEXT: s_mov_b32 s5, s6
; GFX7-NEXT: v_floor_f32_e32 v5, v1
; GFX7-NEXT: v_cndmask_b32_e64 v1, v7, 0, vcc
; GFX7-NEXT: buffer_store_dwordx2 v[4:5], v[2:3], s[4:7], 0 addr64
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: safe_math_fract_v2f32:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_mov_b32_e32 v8, 0x204
; GFX8-NEXT: v_fract_f32_e32 v6, v0
; GFX8-NEXT: v_cmp_class_f32_e32 vcc, v0, v8
; GFX8-NEXT: v_floor_f32_e32 v4, v0
; GFX8-NEXT: v_fract_f32_e32 v7, v1
; GFX8-NEXT: v_cndmask_b32_e64 v0, v6, 0, vcc
; GFX8-NEXT: v_cmp_class_f32_e32 vcc, v1, v8
; GFX8-NEXT: v_floor_f32_e32 v5, v1
; GFX8-NEXT: v_cndmask_b32_e64 v1, v7, 0, vcc
; GFX8-NEXT: global_store_dwordx2 v[2:3], v[4:5], off
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: safe_math_fract_v2f32:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v6, v0
; GFX11-NEXT: v_cmp_class_f32_e64 s0, v0, 0x204
; GFX11-NEXT: v_fract_f32_e32 v7, v1
; GFX11-NEXT: v_floor_f32_e32 v4, v0
; GFX11-NEXT: v_floor_f32_e32 v5, v1
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_4)
; GFX11-NEXT: v_cndmask_b32_e64 v0, v6, 0, s0
; GFX11-NEXT: v_cmp_class_f32_e64 s0, v1, 0x204
; GFX11-NEXT: global_store_b64 v[2:3], v[4:5], off
; GFX11-NEXT: v_cndmask_b32_e64 v1, v7, 0, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: safe_math_fract_v2f32:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v6, v0
; GFX12-NEXT: v_cmp_class_f32_e64 s0, v0, 0x204
; GFX12-NEXT: v_fract_f32_e32 v7, v1
; GFX12-NEXT: v_floor_f32_e32 v4, v0
; GFX12-NEXT: v_floor_f32_e32 v5, v1
; GFX12-NEXT: s_wait_alu depctr_va_sdst(0)
; GFX12-NEXT: v_cndmask_b32_e64 v0, v6, 0, s0
; GFX12-NEXT: v_cmp_class_f32_e64 s0, v1, 0x204
; GFX12-NEXT: global_store_b64 v[2:3], v[4:5], off
; GFX12-NEXT: s_wait_alu depctr_va_sdst(0)
; GFX12-NEXT: v_cndmask_b32_e64 v1, v7, 0, s0
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call <2 x float> @llvm.floor.v2f32(<2 x float> %x)
%sub = fsub <2 x float> %x, %floor
%min = tail call <2 x float> @llvm.minnum.v2f32(<2 x float> %sub, <2 x float> <float 0x3FEFFFFFE0000000, float 0x3FEFFFFFE0000000>)
%uno = fcmp uno <2 x float> %x, zeroinitializer
%cond = select <2 x i1> %uno, <2 x float> %x, <2 x float> %min
%fabs = tail call <2 x float> @llvm.fabs.v2f32(<2 x float> %x)
%cmpinf = fcmp oeq <2 x float> %fabs, <float 0x7FF0000000000000, float 0x7FF0000000000000>
%cond6 = select <2 x i1> %cmpinf, <2 x float> zeroinitializer, <2 x float> %cond
store <2 x float> %floor, ptr addrspace(1) %ip, align 4
ret <2 x float> %cond6
}
define double @safe_math_fract_f64(double %x, ptr addrspace(1) writeonly captures(none) %ip) {
; GFX6-IR-LABEL: define double @safe_math_fract_f64(
; GFX6-IR-SAME: double [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call double @llvm.floor.f64(double [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub double [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call double @llvm.minnum.f64(double [[SUB]], double 0x3FEFFFFFFFFFFFFF)
; GFX6-IR-NEXT: [[UNO:%.*]] = fcmp uno double [[X]], 0.000000e+00
; GFX6-IR-NEXT: [[COND:%.*]] = select i1 [[UNO]], double [[X]], double [[MIN]]
; GFX6-IR-NEXT: [[FABS:%.*]] = tail call double @llvm.fabs.f64(double [[X]])
; GFX6-IR-NEXT: [[CMPINF:%.*]] = fcmp oeq double [[FABS]], 0x7FF0000000000000
; GFX6-IR-NEXT: [[COND6:%.*]] = select i1 [[CMPINF]], double 0.000000e+00, double [[COND]]
; GFX6-IR-NEXT: store double [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; GFX6-IR-NEXT: ret double [[COND6]]
;
; IR-FRACT-LABEL: define double @safe_math_fract_f64(
; IR-FRACT-SAME: double [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[FLOOR:%.*]] = tail call double @llvm.floor.f64(double [[X]])
; IR-FRACT-NEXT: [[COND:%.*]] = call double @llvm.amdgcn.fract.f64(double [[X]])
; IR-FRACT-NEXT: [[FABS:%.*]] = tail call double @llvm.fabs.f64(double [[X]])
; IR-FRACT-NEXT: [[CMPINF:%.*]] = fcmp oeq double [[FABS]], 0x7FF0000000000000
; IR-FRACT-NEXT: [[COND6:%.*]] = select i1 [[CMPINF]], double 0.000000e+00, double [[COND]]
; IR-FRACT-NEXT: store double [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; IR-FRACT-NEXT: ret double [[COND6]]
;
; GFX6-LABEL: safe_math_fract_f64:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_fract_f64_e32 v[4:5], v[0:1]
; GFX6-NEXT: v_mov_b32_e32 v6, -1
; GFX6-NEXT: v_mov_b32_e32 v7, 0x3fefffff
; GFX6-NEXT: v_min_f64 v[4:5], v[4:5], v[6:7]
; GFX6-NEXT: v_cmp_class_f64_e64 vcc, v[0:1], 3
; GFX6-NEXT: s_mov_b32 s8, -1
; GFX6-NEXT: v_cndmask_b32_e32 v4, v4, v0, vcc
; GFX6-NEXT: v_cndmask_b32_e32 v5, v5, v1, vcc
; GFX6-NEXT: v_add_f64 v[4:5], v[0:1], -v[4:5]
; GFX6-NEXT: s_mov_b32 s9, 0x3fefffff
; GFX6-NEXT: v_add_f64 v[6:7], v[0:1], -v[4:5]
; GFX6-NEXT: v_cmp_u_f64_e32 vcc, v[0:1], v[0:1]
; GFX6-NEXT: v_min_f64 v[6:7], v[6:7], s[8:9]
; GFX6-NEXT: s_mov_b32 s8, 0
; GFX6-NEXT: s_mov_b32 s9, 0x7ff00000
; GFX6-NEXT: v_cndmask_b32_e32 v7, v7, v1, vcc
; GFX6-NEXT: v_cndmask_b32_e32 v6, v6, v0, vcc
; GFX6-NEXT: v_cmp_neq_f64_e64 vcc, |v[0:1]|, s[8:9]
; GFX6-NEXT: s_mov_b32 s6, 0
; GFX6-NEXT: s_mov_b32 s7, 0xf000
; GFX6-NEXT: s_mov_b32 s4, s6
; GFX6-NEXT: s_mov_b32 s5, s6
; GFX6-NEXT: v_cndmask_b32_e32 v0, 0, v6, vcc
; GFX6-NEXT: v_cndmask_b32_e32 v1, 0, v7, vcc
; GFX6-NEXT: buffer_store_dwordx2 v[4:5], v[2:3], s[4:7], 0 addr64
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: safe_math_fract_f64:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: s_mov_b32 s4, 0
; GFX7-NEXT: s_mov_b32 s5, 0x7ff00000
; GFX7-NEXT: v_fract_f64_e32 v[4:5], v[0:1]
; GFX7-NEXT: v_cmp_neq_f64_e64 vcc, |v[0:1]|, s[4:5]
; GFX7-NEXT: v_floor_f64_e32 v[6:7], v[0:1]
; GFX7-NEXT: s_mov_b32 s6, 0
; GFX7-NEXT: s_mov_b32 s7, 0xf000
; GFX7-NEXT: s_mov_b32 s4, s6
; GFX7-NEXT: s_mov_b32 s5, s6
; GFX7-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc
; GFX7-NEXT: v_cndmask_b32_e32 v1, 0, v5, vcc
; GFX7-NEXT: buffer_store_dwordx2 v[6:7], v[2:3], s[4:7], 0 addr64
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: safe_math_fract_f64:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: s_mov_b32 s4, 0
; GFX8-NEXT: s_mov_b32 s5, 0x7ff00000
; GFX8-NEXT: v_fract_f64_e32 v[4:5], v[0:1]
; GFX8-NEXT: v_cmp_neq_f64_e64 vcc, |v[0:1]|, s[4:5]
; GFX8-NEXT: v_floor_f64_e32 v[6:7], v[0:1]
; GFX8-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc
; GFX8-NEXT: v_cndmask_b32_e32 v1, 0, v5, vcc
; GFX8-NEXT: global_store_dwordx2 v[2:3], v[6:7], off
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: safe_math_fract_f64:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f64_e32 v[4:5], v[0:1]
; GFX11-NEXT: v_cmp_neq_f64_e64 vcc_lo, 0x7ff00000, |v[0:1]|
; GFX11-NEXT: v_floor_f64_e32 v[6:7], v[0:1]
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_3)
; GFX11-NEXT: v_dual_cndmask_b32 v0, 0, v4 :: v_dual_cndmask_b32 v1, 0, v5
; GFX11-NEXT: global_store_b64 v[2:3], v[6:7], off
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: safe_math_fract_f64:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f64_e32 v[4:5], v[0:1]
; GFX12-NEXT: v_cmp_neq_f64_e64 vcc_lo, 0x7ff00000, |v[0:1]|
; GFX12-NEXT: v_floor_f64_e32 v[6:7], v[0:1]
; GFX12-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_3)
; GFX12-NEXT: v_dual_cndmask_b32 v0, 0, v4 :: v_dual_cndmask_b32 v1, 0, v5
; GFX12-NEXT: global_store_b64 v[2:3], v[6:7], off
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call double @llvm.floor.f64(double %x)
%sub = fsub double %x, %floor
%min = tail call double @llvm.minnum.f64(double %sub, double 0x3FEFFFFFFFFFFFFF)
%uno = fcmp uno double %x, 0.000000e+00
%cond = select i1 %uno, double %x, double %min
%fabs = tail call double @llvm.fabs.f64(double %x)
%cmpinf = fcmp oeq double %fabs, 0x7FF0000000000000
%cond6 = select i1 %cmpinf, double 0.000000e+00, double %cond
store double %floor, ptr addrspace(1) %ip, align 4
ret double %cond6
}
define half @safe_math_fract_f16(half %x, ptr addrspace(1) writeonly captures(none) %ip) {
; GFX6-IR-LABEL: define half @safe_math_fract_f16(
; GFX6-IR-SAME: half [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call half @llvm.floor.f16(half [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub half [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call half @llvm.minnum.f16(half [[SUB]], half 0xH3BFF)
; GFX6-IR-NEXT: [[UNO:%.*]] = fcmp uno half [[X]], 0xH0000
; GFX6-IR-NEXT: [[COND:%.*]] = select i1 [[UNO]], half [[X]], half [[MIN]]
; GFX6-IR-NEXT: [[FABS:%.*]] = tail call half @llvm.fabs.f16(half [[X]])
; GFX6-IR-NEXT: [[CMPINF:%.*]] = fcmp oeq half [[FABS]], 0xH7C00
; GFX6-IR-NEXT: [[COND6:%.*]] = select i1 [[CMPINF]], half 0xH0000, half [[COND]]
; GFX6-IR-NEXT: store half [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; GFX6-IR-NEXT: ret half [[COND6]]
;
; GFX7-IR-LABEL: define half @safe_math_fract_f16(
; GFX7-IR-SAME: half [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0]] {
; GFX7-IR-NEXT: [[ENTRY:.*:]]
; GFX7-IR-NEXT: [[FLOOR:%.*]] = tail call half @llvm.floor.f16(half [[X]])
; GFX7-IR-NEXT: [[SUB:%.*]] = fsub half [[X]], [[FLOOR]]
; GFX7-IR-NEXT: [[MIN:%.*]] = tail call half @llvm.minnum.f16(half [[SUB]], half 0xH3BFF)
; GFX7-IR-NEXT: [[UNO:%.*]] = fcmp uno half [[X]], 0xH0000
; GFX7-IR-NEXT: [[COND:%.*]] = select i1 [[UNO]], half [[X]], half [[MIN]]
; GFX7-IR-NEXT: [[FABS:%.*]] = tail call half @llvm.fabs.f16(half [[X]])
; GFX7-IR-NEXT: [[CMPINF:%.*]] = fcmp oeq half [[FABS]], 0xH7C00
; GFX7-IR-NEXT: [[COND6:%.*]] = select i1 [[CMPINF]], half 0xH0000, half [[COND]]
; GFX7-IR-NEXT: store half [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; GFX7-IR-NEXT: ret half [[COND6]]
;
; IR-LEGALF16-LABEL: define half @safe_math_fract_f16(
; IR-LEGALF16-SAME: half [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0]] {
; IR-LEGALF16-NEXT: [[ENTRY:.*:]]
; IR-LEGALF16-NEXT: [[FLOOR:%.*]] = tail call half @llvm.floor.f16(half [[X]])
; IR-LEGALF16-NEXT: [[COND:%.*]] = call half @llvm.amdgcn.fract.f16(half [[X]])
; IR-LEGALF16-NEXT: [[FABS:%.*]] = tail call half @llvm.fabs.f16(half [[X]])
; IR-LEGALF16-NEXT: [[CMPINF:%.*]] = fcmp oeq half [[FABS]], 0xH7C00
; IR-LEGALF16-NEXT: [[COND6:%.*]] = select i1 [[CMPINF]], half 0xH0000, half [[COND]]
; IR-LEGALF16-NEXT: store half [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; IR-LEGALF16-NEXT: ret half [[COND6]]
;
; GFX6-LABEL: safe_math_fract_f16:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_cvt_f32_f16_e32 v3, v0
; GFX6-NEXT: v_cvt_f32_f16_e64 v6, |v0|
; GFX6-NEXT: s_mov_b32 s8, 0x7f800000
; GFX6-NEXT: s_mov_b32 s6, 0
; GFX6-NEXT: v_floor_f32_e32 v4, v3
; GFX6-NEXT: v_cvt_f16_f32_e32 v4, v4
; GFX6-NEXT: v_cmp_u_f32_e32 vcc, v3, v3
; GFX6-NEXT: s_mov_b32 s7, 0xf000
; GFX6-NEXT: s_mov_b32 s4, s6
; GFX6-NEXT: v_cvt_f32_f16_e32 v5, v4
; GFX6-NEXT: s_mov_b32 s5, s6
; GFX6-NEXT: buffer_store_short v4, v[1:2], s[4:7], 0 addr64
; GFX6-NEXT: v_sub_f32_e32 v5, v3, v5
; GFX6-NEXT: v_cvt_f16_f32_e32 v5, v5
; GFX6-NEXT: v_cvt_f32_f16_e32 v5, v5
; GFX6-NEXT: v_min_f32_e32 v5, 0x3f7fe000, v5
; GFX6-NEXT: v_cvt_f16_f32_e32 v5, v5
; GFX6-NEXT: v_cndmask_b32_e32 v0, v5, v0, vcc
; GFX6-NEXT: v_cmp_neq_f32_e32 vcc, s8, v6
; GFX6-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: safe_math_fract_f16:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_cvt_f32_f16_e32 v3, v0
; GFX7-NEXT: v_cvt_f32_f16_e64 v6, |v0|
; GFX7-NEXT: s_mov_b32 s8, 0x7f800000
; GFX7-NEXT: s_mov_b32 s6, 0
; GFX7-NEXT: v_floor_f32_e32 v4, v3
; GFX7-NEXT: v_cvt_f16_f32_e32 v4, v4
; GFX7-NEXT: v_cmp_u_f32_e32 vcc, v3, v3
; GFX7-NEXT: s_mov_b32 s7, 0xf000
; GFX7-NEXT: s_mov_b32 s4, s6
; GFX7-NEXT: v_cvt_f32_f16_e32 v5, v4
; GFX7-NEXT: s_mov_b32 s5, s6
; GFX7-NEXT: buffer_store_short v4, v[1:2], s[4:7], 0 addr64
; GFX7-NEXT: v_sub_f32_e32 v5, v3, v5
; GFX7-NEXT: v_cvt_f16_f32_e32 v5, v5
; GFX7-NEXT: v_cvt_f32_f16_e32 v5, v5
; GFX7-NEXT: v_min_f32_e32 v5, 0x3f7fe000, v5
; GFX7-NEXT: v_cvt_f16_f32_e32 v5, v5
; GFX7-NEXT: v_cndmask_b32_e32 v0, v5, v0, vcc
; GFX7-NEXT: v_cmp_neq_f32_e32 vcc, s8, v6
; GFX7-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: safe_math_fract_f16:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: s_movk_i32 s4, 0x7c00
; GFX8-NEXT: v_fract_f16_e32 v4, v0
; GFX8-NEXT: v_cmp_neq_f16_e64 vcc, |v0|, s4
; GFX8-NEXT: v_floor_f16_e32 v3, v0
; GFX8-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc
; GFX8-NEXT: global_store_short v[1:2], v3, off
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-TRUE16-LABEL: safe_math_fract_f16:
; GFX11-TRUE16: ; %bb.0: ; %entry
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT: v_fract_f16_e32 v0.h, v0.l
; GFX11-TRUE16-NEXT: v_cmp_neq_f16_e64 s0, 0x7c00, |v0.l|
; GFX11-TRUE16-NEXT: v_floor_f16_e32 v3.l, v0.l
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, 0, v0.h, s0
; GFX11-TRUE16-NEXT: global_store_b16 v[1:2], v3, off
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: safe_math_fract_f16:
; GFX11-FAKE16: ; %bb.0: ; %entry
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-FAKE16-NEXT: v_fract_f16_e32 v3, v0
; GFX11-FAKE16-NEXT: v_cmp_neq_f16_e64 vcc_lo, 0x7c00, |v0|
; GFX11-FAKE16-NEXT: v_floor_f16_e32 v4, v0
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3)
; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc_lo
; GFX11-FAKE16-NEXT: global_store_b16 v[1:2], v4, off
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-TRUE16-LABEL: safe_math_fract_f16:
; GFX12-TRUE16: ; %bb.0: ; %entry
; GFX12-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-TRUE16-NEXT: s_wait_expcnt 0x0
; GFX12-TRUE16-NEXT: s_wait_samplecnt 0x0
; GFX12-TRUE16-NEXT: s_wait_bvhcnt 0x0
; GFX12-TRUE16-NEXT: s_wait_kmcnt 0x0
; GFX12-TRUE16-NEXT: v_fract_f16_e32 v0.h, v0.l
; GFX12-TRUE16-NEXT: v_cmp_neq_f16_e64 s0, 0x7c00, |v0.l|
; GFX12-TRUE16-NEXT: v_floor_f16_e32 v3.l, v0.l
; GFX12-TRUE16-NEXT: s_wait_alu depctr_va_sdst(0)
; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX12-TRUE16-NEXT: v_cndmask_b16 v0.l, 0, v0.h, s0
; GFX12-TRUE16-NEXT: global_store_b16 v[1:2], v3, off
; GFX12-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-FAKE16-LABEL: safe_math_fract_f16:
; GFX12-FAKE16: ; %bb.0: ; %entry
; GFX12-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-FAKE16-NEXT: s_wait_expcnt 0x0
; GFX12-FAKE16-NEXT: s_wait_samplecnt 0x0
; GFX12-FAKE16-NEXT: s_wait_bvhcnt 0x0
; GFX12-FAKE16-NEXT: s_wait_kmcnt 0x0
; GFX12-FAKE16-NEXT: v_fract_f16_e32 v3, v0
; GFX12-FAKE16-NEXT: v_cmp_neq_f16_e64 vcc_lo, 0x7c00, |v0|
; GFX12-FAKE16-NEXT: v_floor_f16_e32 v4, v0
; GFX12-FAKE16-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3)
; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc_lo
; GFX12-FAKE16-NEXT: global_store_b16 v[1:2], v4, off
; GFX12-FAKE16-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call half @llvm.floor.f16(half %x)
%sub = fsub half %x, %floor
%min = tail call half @llvm.minnum.f16(half %sub, half 0xH3BFF)
%uno = fcmp uno half %x, 0.000000e+00
%cond = select i1 %uno, half %x, half %min
%fabs = tail call half @llvm.fabs.f16(half %x)
%cmpinf = fcmp oeq half %fabs, 0xH7C00
%cond6 = select i1 %cmpinf, half 0.000000e+00, half %cond
store half %floor, ptr addrspace(1) %ip, align 4
ret half %cond6
}
define <2 x half> @safe_math_fract_v2f16(<2 x half> %x, ptr addrspace(1) writeonly captures(none) %ip) {
; GFX6-IR-LABEL: define <2 x half> @safe_math_fract_v2f16(
; GFX6-IR-SAME: <2 x half> [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call <2 x half> @llvm.floor.v2f16(<2 x half> [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub <2 x half> [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call <2 x half> @llvm.minnum.v2f16(<2 x half> [[SUB]], <2 x half> splat (half 0xH3BFF))
; GFX6-IR-NEXT: [[UNO:%.*]] = fcmp uno <2 x half> [[X]], zeroinitializer
; GFX6-IR-NEXT: [[COND:%.*]] = select <2 x i1> [[UNO]], <2 x half> [[X]], <2 x half> [[MIN]]
; GFX6-IR-NEXT: [[FABS:%.*]] = tail call <2 x half> @llvm.fabs.v2f16(<2 x half> [[X]])
; GFX6-IR-NEXT: [[CMPINF:%.*]] = fcmp oeq <2 x half> [[FABS]], splat (half 0xH7C00)
; GFX6-IR-NEXT: [[COND6:%.*]] = select <2 x i1> [[CMPINF]], <2 x half> zeroinitializer, <2 x half> [[COND]]
; GFX6-IR-NEXT: store <2 x half> [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; GFX6-IR-NEXT: ret <2 x half> [[COND6]]
;
; GFX7-IR-LABEL: define <2 x half> @safe_math_fract_v2f16(
; GFX7-IR-SAME: <2 x half> [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0]] {
; GFX7-IR-NEXT: [[ENTRY:.*:]]
; GFX7-IR-NEXT: [[FLOOR:%.*]] = tail call <2 x half> @llvm.floor.v2f16(<2 x half> [[X]])
; GFX7-IR-NEXT: [[SUB:%.*]] = fsub <2 x half> [[X]], [[FLOOR]]
; GFX7-IR-NEXT: [[MIN:%.*]] = tail call <2 x half> @llvm.minnum.v2f16(<2 x half> [[SUB]], <2 x half> splat (half 0xH3BFF))
; GFX7-IR-NEXT: [[UNO:%.*]] = fcmp uno <2 x half> [[X]], zeroinitializer
; GFX7-IR-NEXT: [[COND:%.*]] = select <2 x i1> [[UNO]], <2 x half> [[X]], <2 x half> [[MIN]]
; GFX7-IR-NEXT: [[FABS:%.*]] = tail call <2 x half> @llvm.fabs.v2f16(<2 x half> [[X]])
; GFX7-IR-NEXT: [[CMPINF:%.*]] = fcmp oeq <2 x half> [[FABS]], splat (half 0xH7C00)
; GFX7-IR-NEXT: [[COND6:%.*]] = select <2 x i1> [[CMPINF]], <2 x half> zeroinitializer, <2 x half> [[COND]]
; GFX7-IR-NEXT: store <2 x half> [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; GFX7-IR-NEXT: ret <2 x half> [[COND6]]
;
; IR-LEGALF16-LABEL: define <2 x half> @safe_math_fract_v2f16(
; IR-LEGALF16-SAME: <2 x half> [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0]] {
; IR-LEGALF16-NEXT: [[ENTRY:.*:]]
; IR-LEGALF16-NEXT: [[FLOOR:%.*]] = tail call <2 x half> @llvm.floor.v2f16(<2 x half> [[X]])
; IR-LEGALF16-NEXT: [[TMP0:%.*]] = extractelement <2 x half> [[X]], i64 0
; IR-LEGALF16-NEXT: [[TMP1:%.*]] = extractelement <2 x half> [[X]], i64 1
; IR-LEGALF16-NEXT: [[TMP2:%.*]] = call half @llvm.amdgcn.fract.f16(half [[TMP0]])
; IR-LEGALF16-NEXT: [[TMP3:%.*]] = call half @llvm.amdgcn.fract.f16(half [[TMP1]])
; IR-LEGALF16-NEXT: [[TMP4:%.*]] = insertelement <2 x half> poison, half [[TMP2]], i64 0
; IR-LEGALF16-NEXT: [[COND:%.*]] = insertelement <2 x half> [[TMP4]], half [[TMP3]], i64 1
; IR-LEGALF16-NEXT: [[FABS:%.*]] = tail call <2 x half> @llvm.fabs.v2f16(<2 x half> [[X]])
; IR-LEGALF16-NEXT: [[CMPINF:%.*]] = fcmp oeq <2 x half> [[FABS]], splat (half 0xH7C00)
; IR-LEGALF16-NEXT: [[COND6:%.*]] = select <2 x i1> [[CMPINF]], <2 x half> zeroinitializer, <2 x half> [[COND]]
; IR-LEGALF16-NEXT: store <2 x half> [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; IR-LEGALF16-NEXT: ret <2 x half> [[COND6]]
;
; GFX6-LABEL: safe_math_fract_v2f16:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_lshrrev_b32_e32 v3, 16, v0
; GFX6-NEXT: v_cvt_f32_f16_e32 v4, v3
; GFX6-NEXT: v_cvt_f32_f16_e32 v5, v0
; GFX6-NEXT: s_movk_i32 s8, 0x7c00
; GFX6-NEXT: s_mov_b32 s6, 0
; GFX6-NEXT: v_floor_f32_e32 v6, v4
; GFX6-NEXT: v_cvt_f16_f32_e32 v6, v6
; GFX6-NEXT: v_floor_f32_e32 v7, v5
; GFX6-NEXT: v_cvt_f16_f32_e32 v7, v7
; GFX6-NEXT: v_cmp_u_f32_e32 vcc, v4, v4
; GFX6-NEXT: v_cvt_f32_f16_e32 v8, v6
; GFX6-NEXT: v_lshlrev_b32_e32 v6, 16, v6
; GFX6-NEXT: v_cvt_f32_f16_e32 v9, v7
; GFX6-NEXT: v_or_b32_e32 v6, v7, v6
; GFX6-NEXT: v_sub_f32_e32 v8, v4, v8
; GFX6-NEXT: v_cvt_f16_f32_e32 v8, v8
; GFX6-NEXT: v_sub_f32_e32 v9, v5, v9
; GFX6-NEXT: v_cvt_f16_f32_e32 v9, v9
; GFX6-NEXT: s_mov_b32 s7, 0xf000
; GFX6-NEXT: v_cvt_f32_f16_e32 v8, v8
; GFX6-NEXT: s_mov_b32 s4, s6
; GFX6-NEXT: v_cvt_f32_f16_e32 v9, v9
; GFX6-NEXT: s_mov_b32 s5, s6
; GFX6-NEXT: v_min_f32_e32 v7, 0x3f7fe000, v8
; GFX6-NEXT: v_cvt_f16_f32_e32 v7, v7
; GFX6-NEXT: v_min_f32_e32 v8, 0x3f7fe000, v9
; GFX6-NEXT: v_cvt_f16_f32_e32 v8, v8
; GFX6-NEXT: buffer_store_dword v6, v[1:2], s[4:7], 0 addr64
; GFX6-NEXT: v_cndmask_b32_e32 v3, v7, v3, vcc
; GFX6-NEXT: v_cmp_u_f32_e32 vcc, v5, v5
; GFX6-NEXT: v_and_b32_e32 v5, 0x7fff, v0
; GFX6-NEXT: v_cndmask_b32_e32 v4, v8, v0, vcc
; GFX6-NEXT: v_bfe_u32 v0, v0, 16, 15
; GFX6-NEXT: v_cmp_ne_u32_e32 vcc, s8, v5
; GFX6-NEXT: v_cndmask_b32_e32 v4, 0, v4, vcc
; GFX6-NEXT: v_cmp_ne_u32_e32 vcc, s8, v0
; GFX6-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc
; GFX6-NEXT: v_and_b32_e32 v4, 0xffff, v4
; GFX6-NEXT: v_lshlrev_b32_e32 v0, 16, v0
; GFX6-NEXT: v_or_b32_e32 v0, v4, v0
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: safe_math_fract_v2f16:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_lshrrev_b32_e32 v3, 16, v0
; GFX7-NEXT: v_cvt_f32_f16_e32 v4, v3
; GFX7-NEXT: v_cvt_f32_f16_e32 v5, v0
; GFX7-NEXT: s_movk_i32 s8, 0x7c00
; GFX7-NEXT: s_mov_b32 s6, 0
; GFX7-NEXT: v_floor_f32_e32 v6, v4
; GFX7-NEXT: v_cvt_f16_f32_e32 v6, v6
; GFX7-NEXT: v_floor_f32_e32 v7, v5
; GFX7-NEXT: v_cvt_f16_f32_e32 v7, v7
; GFX7-NEXT: v_cmp_u_f32_e32 vcc, v4, v4
; GFX7-NEXT: v_cvt_f32_f16_e32 v8, v6
; GFX7-NEXT: v_lshlrev_b32_e32 v6, 16, v6
; GFX7-NEXT: v_cvt_f32_f16_e32 v9, v7
; GFX7-NEXT: v_or_b32_e32 v6, v7, v6
; GFX7-NEXT: v_sub_f32_e32 v8, v4, v8
; GFX7-NEXT: v_cvt_f16_f32_e32 v8, v8
; GFX7-NEXT: v_sub_f32_e32 v9, v5, v9
; GFX7-NEXT: v_cvt_f16_f32_e32 v9, v9
; GFX7-NEXT: s_mov_b32 s7, 0xf000
; GFX7-NEXT: v_cvt_f32_f16_e32 v8, v8
; GFX7-NEXT: s_mov_b32 s4, s6
; GFX7-NEXT: v_cvt_f32_f16_e32 v9, v9
; GFX7-NEXT: s_mov_b32 s5, s6
; GFX7-NEXT: v_min_f32_e32 v7, 0x3f7fe000, v8
; GFX7-NEXT: v_cvt_f16_f32_e32 v7, v7
; GFX7-NEXT: v_min_f32_e32 v8, 0x3f7fe000, v9
; GFX7-NEXT: v_cvt_f16_f32_e32 v8, v8
; GFX7-NEXT: buffer_store_dword v6, v[1:2], s[4:7], 0 addr64
; GFX7-NEXT: v_cndmask_b32_e32 v3, v7, v3, vcc
; GFX7-NEXT: v_cmp_u_f32_e32 vcc, v5, v5
; GFX7-NEXT: v_and_b32_e32 v5, 0x7fff, v0
; GFX7-NEXT: v_cndmask_b32_e32 v4, v8, v0, vcc
; GFX7-NEXT: v_bfe_u32 v0, v0, 16, 15
; GFX7-NEXT: v_cmp_ne_u32_e32 vcc, s8, v5
; GFX7-NEXT: v_cndmask_b32_e32 v4, 0, v4, vcc
; GFX7-NEXT: v_cmp_ne_u32_e32 vcc, s8, v0
; GFX7-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc
; GFX7-NEXT: v_and_b32_e32 v4, 0xffff, v4
; GFX7-NEXT: v_lshlrev_b32_e32 v0, 16, v0
; GFX7-NEXT: v_or_b32_e32 v0, v4, v0
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: safe_math_fract_v2f16:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: s_movk_i32 s6, 0x204
; GFX8-NEXT: v_floor_f16_sdwa v3, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX8-NEXT: v_floor_f16_e32 v4, v0
; GFX8-NEXT: v_fract_f16_sdwa v5, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX8-NEXT: v_cmp_class_f16_sdwa s[4:5], v0, s6 src0_sel:WORD_1 src1_sel:DWORD
; GFX8-NEXT: v_pack_b32_f16 v3, v4, v3
; GFX8-NEXT: v_fract_f16_e32 v4, v0
; GFX8-NEXT: v_cndmask_b32_e64 v5, v5, 0, s[4:5]
; GFX8-NEXT: v_cmp_class_f16_e64 s[4:5], v0, s6
; GFX8-NEXT: v_cndmask_b32_e64 v0, v4, 0, s[4:5]
; GFX8-NEXT: v_pack_b32_f16 v0, v0, v5
; GFX8-NEXT: global_store_dword v[1:2], v3, off
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-TRUE16-LABEL: safe_math_fract_v2f16:
; GFX11-TRUE16: ; %bb.0: ; %entry
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT: v_fract_f16_e32 v3.l, v0.l
; GFX11-TRUE16-NEXT: v_fract_f16_e32 v3.h, v0.h
; GFX11-TRUE16-NEXT: v_cmp_class_f16_e64 s0, v0.h, 0x204
; GFX11-TRUE16-NEXT: v_cmp_class_f16_e64 s1, v0.l, 0x204
; GFX11-TRUE16-NEXT: v_floor_f16_e32 v4.h, v0.h
; GFX11-TRUE16-NEXT: v_floor_f16_e32 v4.l, v0.l
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.h, 0, s0
; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, 0, s1
; GFX11-TRUE16-NEXT: global_store_b32 v[1:2], v4, off
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: safe_math_fract_v2f16:
; GFX11-FAKE16: ; %bb.0: ; %entry
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v0
; GFX11-FAKE16-NEXT: v_fract_f16_e32 v6, v0
; GFX11-FAKE16-NEXT: v_floor_f16_e32 v5, v0
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_2) | instid1(VALU_DEP_2)
; GFX11-FAKE16-NEXT: v_fract_f16_e32 v4, v3
; GFX11-FAKE16-NEXT: v_cmp_class_f16_e64 s0, v3, 0x204
; GFX11-FAKE16-NEXT: v_floor_f16_e32 v7, v3
; GFX11-FAKE16-NEXT: v_cndmask_b32_e64 v3, v4, 0, s0
; GFX11-FAKE16-NEXT: v_cmp_class_f16_e64 s0, v0, 0x204
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-FAKE16-NEXT: v_pack_b32_f16 v4, v5, v7
; GFX11-FAKE16-NEXT: v_cndmask_b32_e64 v0, v6, 0, s0
; GFX11-FAKE16-NEXT: global_store_b32 v[1:2], v4, off
; GFX11-FAKE16-NEXT: v_pack_b32_f16 v0, v0, v3
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-TRUE16-LABEL: safe_math_fract_v2f16:
; GFX12-TRUE16: ; %bb.0: ; %entry
; GFX12-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-TRUE16-NEXT: s_wait_expcnt 0x0
; GFX12-TRUE16-NEXT: s_wait_samplecnt 0x0
; GFX12-TRUE16-NEXT: s_wait_bvhcnt 0x0
; GFX12-TRUE16-NEXT: s_wait_kmcnt 0x0
; GFX12-TRUE16-NEXT: v_fract_f16_e32 v3.l, v0.l
; GFX12-TRUE16-NEXT: v_fract_f16_e32 v3.h, v0.h
; GFX12-TRUE16-NEXT: v_cmp_class_f16_e64 s0, v0.h, 0x204
; GFX12-TRUE16-NEXT: v_cmp_class_f16_e64 s1, v0.l, 0x204
; GFX12-TRUE16-NEXT: v_floor_f16_e32 v4.h, v0.h
; GFX12-TRUE16-NEXT: v_floor_f16_e32 v4.l, v0.l
; GFX12-TRUE16-NEXT: s_wait_alu depctr_va_sdst(0)
; GFX12-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.h, 0, s0
; GFX12-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, 0, s1
; GFX12-TRUE16-NEXT: global_store_b32 v[1:2], v4, off
; GFX12-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-FAKE16-LABEL: safe_math_fract_v2f16:
; GFX12-FAKE16: ; %bb.0: ; %entry
; GFX12-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-FAKE16-NEXT: s_wait_expcnt 0x0
; GFX12-FAKE16-NEXT: s_wait_samplecnt 0x0
; GFX12-FAKE16-NEXT: s_wait_bvhcnt 0x0
; GFX12-FAKE16-NEXT: s_wait_kmcnt 0x0
; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v0
; GFX12-FAKE16-NEXT: v_fract_f16_e32 v6, v0
; GFX12-FAKE16-NEXT: v_floor_f16_e32 v5, v0
; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_3) | instid1(VALU_DEP_2)
; GFX12-FAKE16-NEXT: v_fract_f16_e32 v4, v3
; GFX12-FAKE16-NEXT: v_cmp_class_f16_e64 s0, v3, 0x204
; GFX12-FAKE16-NEXT: v_floor_f16_e32 v7, v3
; GFX12-FAKE16-NEXT: s_wait_alu depctr_va_sdst(0)
; GFX12-FAKE16-NEXT: v_cndmask_b32_e64 v3, v4, 0, s0
; GFX12-FAKE16-NEXT: v_cmp_class_f16_e64 s0, v0, 0x204
; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX12-FAKE16-NEXT: v_pack_b32_f16 v4, v5, v7
; GFX12-FAKE16-NEXT: s_wait_alu depctr_va_sdst(0)
; GFX12-FAKE16-NEXT: v_cndmask_b32_e64 v0, v6, 0, s0
; GFX12-FAKE16-NEXT: global_store_b32 v[1:2], v4, off
; GFX12-FAKE16-NEXT: v_pack_b32_f16 v0, v0, v3
; GFX12-FAKE16-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call <2 x half> @llvm.floor.v2f16(<2 x half> %x)
%sub = fsub <2 x half> %x, %floor
%min = tail call <2 x half> @llvm.minnum.v2f16(<2 x half> %sub, <2 x half> <half 0xH3BFF, half 0xH3BFF>)
%uno = fcmp uno <2 x half> %x, zeroinitializer
%cond = select <2 x i1> %uno, <2 x half> %x, <2 x half> %min
%fabs = tail call <2 x half> @llvm.fabs.v2f16(<2 x half> %x)
%cmpinf = fcmp oeq <2 x half> %fabs, <half 0xH7C00, half 0xH7C00>
%cond6 = select <2 x i1> %cmpinf, <2 x half> zeroinitializer, <2 x half> %cond
store <2 x half> %floor, ptr addrspace(1) %ip, align 4
ret <2 x half> %cond6
}
define <2 x double> @safe_math_fract_v2f64(<2 x double> %x, ptr addrspace(1) writeonly captures(none) %ip) {
; GFX6-IR-LABEL: define <2 x double> @safe_math_fract_v2f64(
; GFX6-IR-SAME: <2 x double> [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call <2 x double> @llvm.floor.v2f64(<2 x double> [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub <2 x double> [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call <2 x double> @llvm.minnum.v2f64(<2 x double> [[SUB]], <2 x double> splat (double 0x3FEFFFFFFFFFFFFF))
; GFX6-IR-NEXT: [[UNO:%.*]] = fcmp uno <2 x double> [[X]], zeroinitializer
; GFX6-IR-NEXT: [[COND:%.*]] = select <2 x i1> [[UNO]], <2 x double> [[X]], <2 x double> [[MIN]]
; GFX6-IR-NEXT: [[FABS:%.*]] = tail call <2 x double> @llvm.fabs.v2f64(<2 x double> [[X]])
; GFX6-IR-NEXT: [[CMPINF:%.*]] = fcmp oeq <2 x double> [[FABS]], splat (double 0x7FF0000000000000)
; GFX6-IR-NEXT: [[COND6:%.*]] = select <2 x i1> [[CMPINF]], <2 x double> zeroinitializer, <2 x double> [[COND]]
; GFX6-IR-NEXT: store <2 x double> [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; GFX6-IR-NEXT: ret <2 x double> [[COND6]]
;
; IR-FRACT-LABEL: define <2 x double> @safe_math_fract_v2f64(
; IR-FRACT-SAME: <2 x double> [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[FLOOR:%.*]] = tail call <2 x double> @llvm.floor.v2f64(<2 x double> [[X]])
; IR-FRACT-NEXT: [[TMP0:%.*]] = extractelement <2 x double> [[X]], i64 0
; IR-FRACT-NEXT: [[TMP1:%.*]] = extractelement <2 x double> [[X]], i64 1
; IR-FRACT-NEXT: [[TMP2:%.*]] = call double @llvm.amdgcn.fract.f64(double [[TMP0]])
; IR-FRACT-NEXT: [[TMP3:%.*]] = call double @llvm.amdgcn.fract.f64(double [[TMP1]])
; IR-FRACT-NEXT: [[TMP4:%.*]] = insertelement <2 x double> poison, double [[TMP2]], i64 0
; IR-FRACT-NEXT: [[COND:%.*]] = insertelement <2 x double> [[TMP4]], double [[TMP3]], i64 1
; IR-FRACT-NEXT: [[FABS:%.*]] = tail call <2 x double> @llvm.fabs.v2f64(<2 x double> [[X]])
; IR-FRACT-NEXT: [[CMPINF:%.*]] = fcmp oeq <2 x double> [[FABS]], splat (double 0x7FF0000000000000)
; IR-FRACT-NEXT: [[COND6:%.*]] = select <2 x i1> [[CMPINF]], <2 x double> zeroinitializer, <2 x double> [[COND]]
; IR-FRACT-NEXT: store <2 x double> [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; IR-FRACT-NEXT: ret <2 x double> [[COND6]]
;
; GFX6-LABEL: safe_math_fract_v2f64:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_fract_f64_e32 v[6:7], v[2:3]
; GFX6-NEXT: v_mov_b32_e32 v10, -1
; GFX6-NEXT: v_mov_b32_e32 v11, 0x3fefffff
; GFX6-NEXT: v_min_f64 v[6:7], v[6:7], v[10:11]
; GFX6-NEXT: v_cmp_class_f64_e64 vcc, v[2:3], 3
; GFX6-NEXT: v_fract_f64_e32 v[12:13], v[0:1]
; GFX6-NEXT: v_cndmask_b32_e32 v6, v6, v2, vcc
; GFX6-NEXT: v_cndmask_b32_e32 v7, v7, v3, vcc
; GFX6-NEXT: v_add_f64 v[8:9], v[2:3], -v[6:7]
; GFX6-NEXT: v_min_f64 v[6:7], v[12:13], v[10:11]
; GFX6-NEXT: v_cmp_class_f64_e64 vcc, v[0:1], 3
; GFX6-NEXT: v_add_f64 v[10:11], v[2:3], -v[8:9]
; GFX6-NEXT: v_cndmask_b32_e32 v6, v6, v0, vcc
; GFX6-NEXT: v_cndmask_b32_e32 v7, v7, v1, vcc
; GFX6-NEXT: v_add_f64 v[6:7], v[0:1], -v[6:7]
; GFX6-NEXT: s_mov_b32 s8, -1
; GFX6-NEXT: s_mov_b32 s9, 0x3fefffff
; GFX6-NEXT: v_add_f64 v[12:13], v[0:1], -v[6:7]
; GFX6-NEXT: v_min_f64 v[10:11], v[10:11], s[8:9]
; GFX6-NEXT: v_cmp_u_f64_e32 vcc, v[2:3], v[2:3]
; GFX6-NEXT: v_min_f64 v[12:13], v[12:13], s[8:9]
; GFX6-NEXT: v_cndmask_b32_e32 v11, v11, v3, vcc
; GFX6-NEXT: v_cndmask_b32_e32 v10, v10, v2, vcc
; GFX6-NEXT: v_cmp_u_f64_e32 vcc, v[0:1], v[0:1]
; GFX6-NEXT: v_mov_b32_e32 v14, 0x204
; GFX6-NEXT: v_cndmask_b32_e32 v13, v13, v1, vcc
; GFX6-NEXT: v_cndmask_b32_e32 v12, v12, v0, vcc
; GFX6-NEXT: v_cmp_class_f64_e32 vcc, v[0:1], v14
; GFX6-NEXT: s_mov_b32 s6, 0
; GFX6-NEXT: v_cndmask_b32_e64 v0, v12, 0, vcc
; GFX6-NEXT: v_cndmask_b32_e64 v1, v13, 0, vcc
; GFX6-NEXT: v_cmp_class_f64_e32 vcc, v[2:3], v14
; GFX6-NEXT: s_mov_b32 s7, 0xf000
; GFX6-NEXT: s_mov_b32 s4, s6
; GFX6-NEXT: s_mov_b32 s5, s6
; GFX6-NEXT: v_cndmask_b32_e64 v2, v10, 0, vcc
; GFX6-NEXT: v_cndmask_b32_e64 v3, v11, 0, vcc
; GFX6-NEXT: buffer_store_dwordx4 v[6:9], v[4:5], s[4:7], 0 addr64
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: safe_math_fract_v2f64:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_mov_b32_e32 v6, 0x204
; GFX7-NEXT: v_fract_f64_e32 v[10:11], v[0:1]
; GFX7-NEXT: v_cmp_class_f64_e32 vcc, v[0:1], v6
; GFX7-NEXT: v_fract_f64_e32 v[12:13], v[2:3]
; GFX7-NEXT: v_cmp_class_f64_e64 s[4:5], v[2:3], v6
; GFX7-NEXT: v_floor_f64_e32 v[8:9], v[2:3]
; GFX7-NEXT: v_floor_f64_e32 v[6:7], v[0:1]
; GFX7-NEXT: s_mov_b32 s10, 0
; GFX7-NEXT: s_mov_b32 s11, 0xf000
; GFX7-NEXT: s_mov_b32 s8, s10
; GFX7-NEXT: s_mov_b32 s9, s10
; GFX7-NEXT: v_cndmask_b32_e64 v0, v10, 0, vcc
; GFX7-NEXT: v_cndmask_b32_e64 v1, v11, 0, vcc
; GFX7-NEXT: v_cndmask_b32_e64 v2, v12, 0, s[4:5]
; GFX7-NEXT: v_cndmask_b32_e64 v3, v13, 0, s[4:5]
; GFX7-NEXT: buffer_store_dwordx4 v[6:9], v[4:5], s[8:11], 0 addr64
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: safe_math_fract_v2f64:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_mov_b32_e32 v6, 0x204
; GFX8-NEXT: v_fract_f64_e32 v[10:11], v[0:1]
; GFX8-NEXT: v_cmp_class_f64_e32 vcc, v[0:1], v6
; GFX8-NEXT: v_fract_f64_e32 v[12:13], v[2:3]
; GFX8-NEXT: v_cmp_class_f64_e64 s[4:5], v[2:3], v6
; GFX8-NEXT: v_floor_f64_e32 v[8:9], v[2:3]
; GFX8-NEXT: v_floor_f64_e32 v[6:7], v[0:1]
; GFX8-NEXT: v_cndmask_b32_e64 v0, v10, 0, vcc
; GFX8-NEXT: v_cndmask_b32_e64 v1, v11, 0, vcc
; GFX8-NEXT: v_cndmask_b32_e64 v2, v12, 0, s[4:5]
; GFX8-NEXT: v_cndmask_b32_e64 v3, v13, 0, s[4:5]
; GFX8-NEXT: global_store_dwordx4 v[4:5], v[6:9], off
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: safe_math_fract_v2f64:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f64_e32 v[10:11], v[0:1]
; GFX11-NEXT: v_cmp_class_f64_e64 s0, v[0:1], 0x204
; GFX11-NEXT: v_fract_f64_e32 v[12:13], v[2:3]
; GFX11-NEXT: v_cmp_class_f64_e64 s1, v[2:3], 0x204
; GFX11-NEXT: v_floor_f64_e32 v[8:9], v[2:3]
; GFX11-NEXT: v_floor_f64_e32 v[6:7], v[0:1]
; GFX11-NEXT: v_cndmask_b32_e64 v0, v10, 0, s0
; GFX11-NEXT: v_cndmask_b32_e64 v1, v11, 0, s0
; GFX11-NEXT: v_cndmask_b32_e64 v2, v12, 0, s1
; GFX11-NEXT: v_cndmask_b32_e64 v3, v13, 0, s1
; GFX11-NEXT: global_store_b128 v[4:5], v[6:9], off
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: safe_math_fract_v2f64:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f64_e32 v[10:11], v[0:1]
; GFX12-NEXT: v_cmp_class_f64_e64 s0, v[0:1], 0x204
; GFX12-NEXT: v_fract_f64_e32 v[12:13], v[2:3]
; GFX12-NEXT: v_cmp_class_f64_e64 s1, v[2:3], 0x204
; GFX12-NEXT: v_floor_f64_e32 v[8:9], v[2:3]
; GFX12-NEXT: v_floor_f64_e32 v[6:7], v[0:1]
; GFX12-NEXT: s_wait_alu depctr_va_sdst(0)
; GFX12-NEXT: v_cndmask_b32_e64 v0, v10, 0, s0
; GFX12-NEXT: v_cndmask_b32_e64 v1, v11, 0, s0
; GFX12-NEXT: v_cndmask_b32_e64 v2, v12, 0, s1
; GFX12-NEXT: v_cndmask_b32_e64 v3, v13, 0, s1
; GFX12-NEXT: global_store_b128 v[4:5], v[6:9], off
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call <2 x double> @llvm.floor.v2f64(<2 x double> %x)
%sub = fsub <2 x double> %x, %floor
%min = tail call <2 x double> @llvm.minnum.v2f64(<2 x double> %sub, <2 x double> <double 0x3FEFFFFFFFFFFFFF, double 0x3FEFFFFFFFFFFFFF>)
%uno = fcmp uno <2 x double> %x, zeroinitializer
%cond = select <2 x i1> %uno, <2 x double> %x, <2 x double> %min
%fabs = tail call <2 x double> @llvm.fabs.v2f64(<2 x double> %x)
%cmpinf = fcmp oeq <2 x double> %fabs, <double 0x7FF0000000000000, double 0x7FF0000000000000>
%cond6 = select <2 x i1> %cmpinf, <2 x double> zeroinitializer, <2 x double> %cond
store <2 x double> %floor, ptr addrspace(1) %ip, align 4
ret <2 x double> %cond6
}
define float @safe_math_fract_f32_minimum(float %x, ptr addrspace(1) writeonly captures(none) %ip) {
; GFX6-IR-LABEL: define float @safe_math_fract_f32_minimum(
; GFX6-IR-SAME: float [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call float @llvm.minimum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; GFX6-IR-NEXT: [[UNO:%.*]] = fcmp uno float [[X]], 0.000000e+00
; GFX6-IR-NEXT: [[COND:%.*]] = select i1 [[UNO]], float [[X]], float [[MIN]]
; GFX6-IR-NEXT: [[FABS:%.*]] = tail call float @llvm.fabs.f32(float [[X]])
; GFX6-IR-NEXT: [[CMPINF:%.*]] = fcmp oeq float [[FABS]], 0x7FF0000000000000
; GFX6-IR-NEXT: [[COND6:%.*]] = select i1 [[CMPINF]], float 0.000000e+00, float [[COND]]
; GFX6-IR-NEXT: store float [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; GFX6-IR-NEXT: ret float [[COND6]]
;
; IR-FRACT-LABEL: define float @safe_math_fract_f32_minimum(
; IR-FRACT-SAME: float [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; IR-FRACT-NEXT: [[COND:%.*]] = call float @llvm.amdgcn.fract.f32(float [[X]])
; IR-FRACT-NEXT: [[FABS:%.*]] = tail call float @llvm.fabs.f32(float [[X]])
; IR-FRACT-NEXT: [[CMPINF:%.*]] = fcmp oeq float [[FABS]], 0x7FF0000000000000
; IR-FRACT-NEXT: [[COND6:%.*]] = select i1 [[CMPINF]], float 0.000000e+00, float [[COND]]
; IR-FRACT-NEXT: store float [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; IR-FRACT-NEXT: ret float [[COND6]]
;
; GFX6-LABEL: safe_math_fract_f32_minimum:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v3, v0
; GFX6-NEXT: v_sub_f32_e32 v4, v0, v3
; GFX6-NEXT: v_min_f32_e32 v5, 0x3f7fffff, v4
; GFX6-NEXT: v_mov_b32_e32 v6, 0x7fc00000
; GFX6-NEXT: v_cmp_o_f32_e32 vcc, v4, v4
; GFX6-NEXT: v_cndmask_b32_e32 v4, v6, v5, vcc
; GFX6-NEXT: v_cmp_u_f32_e32 vcc, v0, v0
; GFX6-NEXT: s_mov_b32 s8, 0x7f800000
; GFX6-NEXT: s_mov_b32 s6, 0
; GFX6-NEXT: v_cndmask_b32_e32 v4, v4, v0, vcc
; GFX6-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s8
; GFX6-NEXT: s_mov_b32 s7, 0xf000
; GFX6-NEXT: s_mov_b32 s4, s6
; GFX6-NEXT: s_mov_b32 s5, s6
; GFX6-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc
; GFX6-NEXT: buffer_store_dword v3, v[1:2], s[4:7], 0 addr64
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: safe_math_fract_f32_minimum:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: s_mov_b32 s8, 0x7f800000
; GFX7-NEXT: s_mov_b32 s6, 0
; GFX7-NEXT: v_fract_f32_e32 v4, v0
; GFX7-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s8
; GFX7-NEXT: s_mov_b32 s7, 0xf000
; GFX7-NEXT: s_mov_b32 s4, s6
; GFX7-NEXT: s_mov_b32 s5, s6
; GFX7-NEXT: v_floor_f32_e32 v3, v0
; GFX7-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc
; GFX7-NEXT: buffer_store_dword v3, v[1:2], s[4:7], 0 addr64
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: safe_math_fract_f32_minimum:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: s_mov_b32 s4, 0x7f800000
; GFX8-NEXT: v_fract_f32_e32 v4, v0
; GFX8-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX8-NEXT: v_floor_f32_e32 v3, v0
; GFX8-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc
; GFX8-NEXT: global_store_dword v[1:2], v3, off
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: safe_math_fract_f32_minimum:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v3, v0
; GFX11-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX11-NEXT: v_floor_f32_e32 v4, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_3)
; GFX11-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc_lo
; GFX11-NEXT: global_store_b32 v[1:2], v4, off
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: safe_math_fract_f32_minimum:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v3, v0
; GFX12-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX12-NEXT: v_floor_f32_e32 v4, v0
; GFX12-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_3)
; GFX12-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc_lo
; GFX12-NEXT: global_store_b32 v[1:2], v4, off
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = tail call float @llvm.minimum.f32(float %sub, float 0x3FEFFFFFE0000000)
%uno = fcmp uno float %x, 0.000000e+00
%cond = select i1 %uno, float %x, float %min
%fabs = tail call float @llvm.fabs.f32(float %x)
%cmpinf = fcmp oeq float %fabs, 0x7FF0000000000000
%cond6 = select i1 %cmpinf, float 0.000000e+00, float %cond
store float %floor, ptr addrspace(1) %ip, align 4
ret float %cond6
}
define float @safe_math_fract_f32_minimum_swap(float %x, ptr addrspace(1) writeonly captures(none) %ip) {
; GFX6-IR-LABEL: define float @safe_math_fract_f32_minimum_swap(
; GFX6-IR-SAME: float [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call float @llvm.minimum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; GFX6-IR-NEXT: [[UNO:%.*]] = fcmp ord float [[X]], 0.000000e+00
; GFX6-IR-NEXT: [[COND:%.*]] = select i1 [[UNO]], float [[MIN]], float [[X]]
; GFX6-IR-NEXT: [[FABS:%.*]] = tail call float @llvm.fabs.f32(float [[X]])
; GFX6-IR-NEXT: [[CMPINF:%.*]] = fcmp oeq float [[FABS]], 0x7FF0000000000000
; GFX6-IR-NEXT: [[COND6:%.*]] = select i1 [[CMPINF]], float 0.000000e+00, float [[COND]]
; GFX6-IR-NEXT: store float [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; GFX6-IR-NEXT: ret float [[COND6]]
;
; IR-FRACT-LABEL: define float @safe_math_fract_f32_minimum_swap(
; IR-FRACT-SAME: float [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; IR-FRACT-NEXT: [[COND:%.*]] = call float @llvm.amdgcn.fract.f32(float [[X]])
; IR-FRACT-NEXT: [[FABS:%.*]] = tail call float @llvm.fabs.f32(float [[X]])
; IR-FRACT-NEXT: [[CMPINF:%.*]] = fcmp oeq float [[FABS]], 0x7FF0000000000000
; IR-FRACT-NEXT: [[COND6:%.*]] = select i1 [[CMPINF]], float 0.000000e+00, float [[COND]]
; IR-FRACT-NEXT: store float [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; IR-FRACT-NEXT: ret float [[COND6]]
;
; GFX6-LABEL: safe_math_fract_f32_minimum_swap:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v3, v0
; GFX6-NEXT: v_sub_f32_e32 v4, v0, v3
; GFX6-NEXT: v_min_f32_e32 v5, 0x3f7fffff, v4
; GFX6-NEXT: v_mov_b32_e32 v6, 0x7fc00000
; GFX6-NEXT: v_cmp_o_f32_e32 vcc, v4, v4
; GFX6-NEXT: v_cndmask_b32_e32 v4, v6, v5, vcc
; GFX6-NEXT: v_cmp_o_f32_e32 vcc, v0, v0
; GFX6-NEXT: s_mov_b32 s8, 0x7f800000
; GFX6-NEXT: s_mov_b32 s6, 0
; GFX6-NEXT: v_cndmask_b32_e32 v4, v0, v4, vcc
; GFX6-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s8
; GFX6-NEXT: s_mov_b32 s7, 0xf000
; GFX6-NEXT: s_mov_b32 s4, s6
; GFX6-NEXT: s_mov_b32 s5, s6
; GFX6-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc
; GFX6-NEXT: buffer_store_dword v3, v[1:2], s[4:7], 0 addr64
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: safe_math_fract_f32_minimum_swap:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: s_mov_b32 s8, 0x7f800000
; GFX7-NEXT: s_mov_b32 s6, 0
; GFX7-NEXT: v_fract_f32_e32 v4, v0
; GFX7-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s8
; GFX7-NEXT: s_mov_b32 s7, 0xf000
; GFX7-NEXT: s_mov_b32 s4, s6
; GFX7-NEXT: s_mov_b32 s5, s6
; GFX7-NEXT: v_floor_f32_e32 v3, v0
; GFX7-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc
; GFX7-NEXT: buffer_store_dword v3, v[1:2], s[4:7], 0 addr64
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: safe_math_fract_f32_minimum_swap:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: s_mov_b32 s4, 0x7f800000
; GFX8-NEXT: v_fract_f32_e32 v4, v0
; GFX8-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX8-NEXT: v_floor_f32_e32 v3, v0
; GFX8-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc
; GFX8-NEXT: global_store_dword v[1:2], v3, off
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: safe_math_fract_f32_minimum_swap:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v3, v0
; GFX11-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX11-NEXT: v_floor_f32_e32 v4, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_3)
; GFX11-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc_lo
; GFX11-NEXT: global_store_b32 v[1:2], v4, off
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: safe_math_fract_f32_minimum_swap:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v3, v0
; GFX12-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX12-NEXT: v_floor_f32_e32 v4, v0
; GFX12-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_3)
; GFX12-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc_lo
; GFX12-NEXT: global_store_b32 v[1:2], v4, off
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = tail call float @llvm.minimum.f32(float %sub, float 0x3FEFFFFFE0000000)
%uno = fcmp ord float %x, 0.000000e+00
%cond = select i1 %uno, float %min, float %x
%fabs = tail call float @llvm.fabs.f32(float %x)
%cmpinf = fcmp oeq float %fabs, 0x7FF0000000000000
%cond6 = select i1 %cmpinf, float 0.000000e+00, float %cond
store float %floor, ptr addrspace(1) %ip, align 4
ret float %cond6
}
define float @safe_math_fract_f32_minimumnum(float %x, ptr addrspace(1) writeonly captures(none) %ip) {
; GFX6-IR-LABEL: define float @safe_math_fract_f32_minimumnum(
; GFX6-IR-SAME: float [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call float @llvm.minimumnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; GFX6-IR-NEXT: [[UNO:%.*]] = fcmp uno float [[X]], 0.000000e+00
; GFX6-IR-NEXT: [[COND:%.*]] = select i1 [[UNO]], float [[X]], float [[MIN]]
; GFX6-IR-NEXT: [[FABS:%.*]] = tail call float @llvm.fabs.f32(float [[X]])
; GFX6-IR-NEXT: [[CMPINF:%.*]] = fcmp oeq float [[FABS]], 0x7FF0000000000000
; GFX6-IR-NEXT: [[COND6:%.*]] = select i1 [[CMPINF]], float 0.000000e+00, float [[COND]]
; GFX6-IR-NEXT: store float [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; GFX6-IR-NEXT: ret float [[COND6]]
;
; IR-FRACT-LABEL: define float @safe_math_fract_f32_minimumnum(
; IR-FRACT-SAME: float [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; IR-FRACT-NEXT: [[COND:%.*]] = call float @llvm.amdgcn.fract.f32(float [[X]])
; IR-FRACT-NEXT: [[FABS:%.*]] = tail call float @llvm.fabs.f32(float [[X]])
; IR-FRACT-NEXT: [[CMPINF:%.*]] = fcmp oeq float [[FABS]], 0x7FF0000000000000
; IR-FRACT-NEXT: [[COND6:%.*]] = select i1 [[CMPINF]], float 0.000000e+00, float [[COND]]
; IR-FRACT-NEXT: store float [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; IR-FRACT-NEXT: ret float [[COND6]]
;
; GFX6-LABEL: safe_math_fract_f32_minimumnum:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v3, v0
; GFX6-NEXT: v_sub_f32_e32 v4, v0, v3
; GFX6-NEXT: v_min_f32_e32 v4, 0x3f7fffff, v4
; GFX6-NEXT: v_cmp_u_f32_e32 vcc, v0, v0
; GFX6-NEXT: s_mov_b32 s8, 0x7f800000
; GFX6-NEXT: s_mov_b32 s6, 0
; GFX6-NEXT: v_cndmask_b32_e32 v4, v4, v0, vcc
; GFX6-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s8
; GFX6-NEXT: s_mov_b32 s7, 0xf000
; GFX6-NEXT: s_mov_b32 s4, s6
; GFX6-NEXT: s_mov_b32 s5, s6
; GFX6-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc
; GFX6-NEXT: buffer_store_dword v3, v[1:2], s[4:7], 0 addr64
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: safe_math_fract_f32_minimumnum:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: s_mov_b32 s8, 0x7f800000
; GFX7-NEXT: s_mov_b32 s6, 0
; GFX7-NEXT: v_fract_f32_e32 v4, v0
; GFX7-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s8
; GFX7-NEXT: s_mov_b32 s7, 0xf000
; GFX7-NEXT: s_mov_b32 s4, s6
; GFX7-NEXT: s_mov_b32 s5, s6
; GFX7-NEXT: v_floor_f32_e32 v3, v0
; GFX7-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc
; GFX7-NEXT: buffer_store_dword v3, v[1:2], s[4:7], 0 addr64
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: safe_math_fract_f32_minimumnum:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: s_mov_b32 s4, 0x7f800000
; GFX8-NEXT: v_fract_f32_e32 v4, v0
; GFX8-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX8-NEXT: v_floor_f32_e32 v3, v0
; GFX8-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc
; GFX8-NEXT: global_store_dword v[1:2], v3, off
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: safe_math_fract_f32_minimumnum:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v3, v0
; GFX11-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX11-NEXT: v_floor_f32_e32 v4, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_3)
; GFX11-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc_lo
; GFX11-NEXT: global_store_b32 v[1:2], v4, off
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: safe_math_fract_f32_minimumnum:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v3, v0
; GFX12-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX12-NEXT: v_floor_f32_e32 v4, v0
; GFX12-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_3)
; GFX12-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc_lo
; GFX12-NEXT: global_store_b32 v[1:2], v4, off
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = tail call float @llvm.minimumnum.f32(float %sub, float 0x3FEFFFFFE0000000)
%uno = fcmp uno float %x, 0.000000e+00
%cond = select i1 %uno, float %x, float %min
%fabs = tail call float @llvm.fabs.f32(float %x)
%cmpinf = fcmp oeq float %fabs, 0x7FF0000000000000
%cond6 = select i1 %cmpinf, float 0.000000e+00, float %cond
store float %floor, ptr addrspace(1) %ip, align 4
ret float %cond6
}
define float @safe_math_fract_f32_minimumnum_swap(float %x, ptr addrspace(1) writeonly captures(none) %ip) {
; GFX6-IR-LABEL: define float @safe_math_fract_f32_minimumnum_swap(
; GFX6-IR-SAME: float [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call float @llvm.minimumnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; GFX6-IR-NEXT: [[UNO:%.*]] = fcmp ord float [[X]], 0.000000e+00
; GFX6-IR-NEXT: [[COND:%.*]] = select i1 [[UNO]], float [[MIN]], float [[X]]
; GFX6-IR-NEXT: [[FABS:%.*]] = tail call float @llvm.fabs.f32(float [[X]])
; GFX6-IR-NEXT: [[CMPINF:%.*]] = fcmp oeq float [[FABS]], 0x7FF0000000000000
; GFX6-IR-NEXT: [[COND6:%.*]] = select i1 [[CMPINF]], float 0.000000e+00, float [[COND]]
; GFX6-IR-NEXT: store float [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; GFX6-IR-NEXT: ret float [[COND6]]
;
; IR-FRACT-LABEL: define float @safe_math_fract_f32_minimumnum_swap(
; IR-FRACT-SAME: float [[X:%.*]], ptr addrspace(1) writeonly captures(none) [[IP:%.*]]) #[[ATTR0]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; IR-FRACT-NEXT: [[COND:%.*]] = call float @llvm.amdgcn.fract.f32(float [[X]])
; IR-FRACT-NEXT: [[FABS:%.*]] = tail call float @llvm.fabs.f32(float [[X]])
; IR-FRACT-NEXT: [[CMPINF:%.*]] = fcmp oeq float [[FABS]], 0x7FF0000000000000
; IR-FRACT-NEXT: [[COND6:%.*]] = select i1 [[CMPINF]], float 0.000000e+00, float [[COND]]
; IR-FRACT-NEXT: store float [[FLOOR]], ptr addrspace(1) [[IP]], align 4
; IR-FRACT-NEXT: ret float [[COND6]]
;
; GFX6-LABEL: safe_math_fract_f32_minimumnum_swap:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v3, v0
; GFX6-NEXT: v_sub_f32_e32 v4, v0, v3
; GFX6-NEXT: v_min_f32_e32 v4, 0x3f7fffff, v4
; GFX6-NEXT: v_cmp_o_f32_e32 vcc, v0, v0
; GFX6-NEXT: s_mov_b32 s8, 0x7f800000
; GFX6-NEXT: s_mov_b32 s6, 0
; GFX6-NEXT: v_cndmask_b32_e32 v4, v0, v4, vcc
; GFX6-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s8
; GFX6-NEXT: s_mov_b32 s7, 0xf000
; GFX6-NEXT: s_mov_b32 s4, s6
; GFX6-NEXT: s_mov_b32 s5, s6
; GFX6-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc
; GFX6-NEXT: buffer_store_dword v3, v[1:2], s[4:7], 0 addr64
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: safe_math_fract_f32_minimumnum_swap:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: s_mov_b32 s8, 0x7f800000
; GFX7-NEXT: s_mov_b32 s6, 0
; GFX7-NEXT: v_fract_f32_e32 v4, v0
; GFX7-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s8
; GFX7-NEXT: s_mov_b32 s7, 0xf000
; GFX7-NEXT: s_mov_b32 s4, s6
; GFX7-NEXT: s_mov_b32 s5, s6
; GFX7-NEXT: v_floor_f32_e32 v3, v0
; GFX7-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc
; GFX7-NEXT: buffer_store_dword v3, v[1:2], s[4:7], 0 addr64
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: safe_math_fract_f32_minimumnum_swap:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: s_mov_b32 s4, 0x7f800000
; GFX8-NEXT: v_fract_f32_e32 v4, v0
; GFX8-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX8-NEXT: v_floor_f32_e32 v3, v0
; GFX8-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc
; GFX8-NEXT: global_store_dword v[1:2], v3, off
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: safe_math_fract_f32_minimumnum_swap:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v3, v0
; GFX11-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX11-NEXT: v_floor_f32_e32 v4, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_3)
; GFX11-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc_lo
; GFX11-NEXT: global_store_b32 v[1:2], v4, off
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: safe_math_fract_f32_minimumnum_swap:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v3, v0
; GFX12-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX12-NEXT: v_floor_f32_e32 v4, v0
; GFX12-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_3)
; GFX12-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc_lo
; GFX12-NEXT: global_store_b32 v[1:2], v4, off
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = tail call float @llvm.minimumnum.f32(float %sub, float 0x3FEFFFFFE0000000)
%uno = fcmp ord float %x, 0.000000e+00
%cond = select i1 %uno, float %min, float %x
%fabs = tail call float @llvm.fabs.f32(float %x)
%cmpinf = fcmp oeq float %fabs, 0x7FF0000000000000
%cond6 = select i1 %cmpinf, float 0.000000e+00, float %cond
store float %floor, ptr addrspace(1) %ip, align 4
ret float %cond6
}
define float @basic_fract_f32_nonans_minimumnum(float nofpclass(nan) %x) {
; GFX6-IR-LABEL: define float @basic_fract_f32_nonans_minimumnum(
; GFX6-IR-SAME: float nofpclass(nan) [[X:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call float @llvm.minimumnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; GFX6-IR-NEXT: ret float [[MIN]]
;
; IR-FRACT-LABEL: define float @basic_fract_f32_nonans_minimumnum(
; IR-FRACT-SAME: float nofpclass(nan) [[X:%.*]]) #[[ATTR0]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[MIN:%.*]] = call nnan float @llvm.amdgcn.fract.f32(float [[X]])
; IR-FRACT-NEXT: ret float [[MIN]]
;
; GFX6-LABEL: basic_fract_f32_nonans_minimumnum:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX6-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: basic_fract_f32_nonans_minimumnum:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_fract_f32_e32 v0, v0
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: basic_fract_f32_nonans_minimumnum:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_fract_f32_e32 v0, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: basic_fract_f32_nonans_minimumnum:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v0, v0
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: basic_fract_f32_nonans_minimumnum:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v0, v0
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = tail call float @llvm.minimumnum.f32(float %sub, float 0x3FEFFFFFE0000000)
ret float %min
}
define float @basic_fract_f32_nonans_minimum(float nofpclass(nan) %x) {
; GFX6-IR-LABEL: define float @basic_fract_f32_nonans_minimum(
; GFX6-IR-SAME: float nofpclass(nan) [[X:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call float @llvm.minimum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; GFX6-IR-NEXT: ret float [[MIN]]
;
; IR-FRACT-LABEL: define float @basic_fract_f32_nonans_minimum(
; IR-FRACT-SAME: float nofpclass(nan) [[X:%.*]]) #[[ATTR0]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[MIN:%.*]] = call nnan float @llvm.amdgcn.fract.f32(float [[X]])
; IR-FRACT-NEXT: ret float [[MIN]]
;
; GFX6-LABEL: basic_fract_f32_nonans_minimum:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX6-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v0
; GFX6-NEXT: v_mov_b32_e32 v2, 0x7fc00000
; GFX6-NEXT: v_cmp_o_f32_e32 vcc, v0, v0
; GFX6-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: basic_fract_f32_nonans_minimum:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_fract_f32_e32 v0, v0
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: basic_fract_f32_nonans_minimum:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_fract_f32_e32 v0, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: basic_fract_f32_nonans_minimum:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v0, v0
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: basic_fract_f32_nonans_minimum:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v0, v0
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = tail call float @llvm.minimum.f32(float %sub, float 0x3FEFFFFFE0000000)
ret float %min
}
define float @nnan_minimum_fract_f32(float %x) {
; GFX6-IR-LABEL: define float @nnan_minimum_fract_f32(
; GFX6-IR-SAME: float [[X:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call nnan float @llvm.minimum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; GFX6-IR-NEXT: ret float [[MIN]]
;
; IR-FRACT-LABEL: define float @nnan_minimum_fract_f32(
; IR-FRACT-SAME: float [[X:%.*]]) #[[ATTR0]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[MIN:%.*]] = call nnan float @llvm.amdgcn.fract.f32(float [[X]])
; IR-FRACT-NEXT: ret float [[MIN]]
;
; GFX6-LABEL: nnan_minimum_fract_f32:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX6-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: nnan_minimum_fract_f32:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_fract_f32_e32 v0, v0
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: nnan_minimum_fract_f32:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_fract_f32_e32 v0, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: nnan_minimum_fract_f32:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v0, v0
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: nnan_minimum_fract_f32:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v0, v0
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = tail call nnan float @llvm.minimum.f32(float %sub, float 0x3FEFFFFFE0000000)
ret float %min
}
define float @nnan_minimumnum_fract_f32(float %x) {
; GFX6-IR-LABEL: define float @nnan_minimumnum_fract_f32(
; GFX6-IR-SAME: float [[X:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call nnan float @llvm.minimumnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; GFX6-IR-NEXT: ret float [[MIN]]
;
; IR-FRACT-LABEL: define float @nnan_minimumnum_fract_f32(
; IR-FRACT-SAME: float [[X:%.*]]) #[[ATTR0]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[MIN:%.*]] = call nnan float @llvm.amdgcn.fract.f32(float [[X]])
; IR-FRACT-NEXT: ret float [[MIN]]
;
; GFX6-LABEL: nnan_minimumnum_fract_f32:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX6-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: nnan_minimumnum_fract_f32:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_fract_f32_e32 v0, v0
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: nnan_minimumnum_fract_f32:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_fract_f32_e32 v0, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: nnan_minimumnum_fract_f32:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v0, v0
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: nnan_minimumnum_fract_f32:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v0, v0
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = tail call nnan float @llvm.minimumnum.f32(float %sub, float 0x3FEFFFFFE0000000)
ret float %min
}
define float @basic_fract_f32_flags_minimumnum(float %x) {
; IR-LABEL: define float @basic_fract_f32_flags_minimumnum(
; IR-SAME: float [[X:%.*]]) #[[ATTR0]] {
; IR-NEXT: [[ENTRY:.*:]]
; IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; IR-NEXT: [[MIN:%.*]] = tail call nsz float @llvm.minimumnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; IR-NEXT: ret float [[MIN]]
;
; GFX6-LABEL: basic_fract_f32_flags_minimumnum:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX6-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: basic_fract_f32_flags_minimumnum:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_floor_f32_e32 v1, v0
; GFX7-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX7-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: basic_fract_f32_flags_minimumnum:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_floor_f32_e32 v1, v0
; GFX8-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX8-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: basic_fract_f32_flags_minimumnum:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_floor_f32_e32 v1, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX11-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: basic_fract_f32_flags_minimumnum:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_floor_f32_e32 v1, v0
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX12-NEXT: v_min_num_f32_e32 v0, 0x3f7fffff, v0
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = tail call nsz float @llvm.minimumnum.f32(float %sub, float 0x3FEFFFFFE0000000)
ret float %min
}
define float @basic_fract_f32_flags_minimum(float %x) {
; GFX6-IR-LABEL: define float @basic_fract_f32_flags_minimum(
; GFX6-IR-SAME: float [[X:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call nsz float @llvm.minimum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; GFX6-IR-NEXT: ret float [[MIN]]
;
; IR-FRACT-LABEL: define float @basic_fract_f32_flags_minimum(
; IR-FRACT-SAME: float [[X:%.*]]) #[[ATTR0]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[MIN:%.*]] = call nnan nsz float @llvm.amdgcn.fract.f32(float [[X]])
; IR-FRACT-NEXT: ret float [[MIN]]
;
; GFX6-LABEL: basic_fract_f32_flags_minimum:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX6-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v0
; GFX6-NEXT: v_mov_b32_e32 v2, 0x7fc00000
; GFX6-NEXT: v_cmp_o_f32_e32 vcc, v0, v0
; GFX6-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: basic_fract_f32_flags_minimum:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_fract_f32_e32 v0, v0
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: basic_fract_f32_flags_minimum:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_fract_f32_e32 v0, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: basic_fract_f32_flags_minimum:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v0, v0
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: basic_fract_f32_flags_minimum:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v0, v0
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = tail call nsz float @llvm.minimum.f32(float %sub, float 0x3FEFFFFFE0000000)
ret float %min
}
define double @fract_match_f64_assume_not_nan(double %x) #0 {
; GFX6-LABEL: fract_match_f64_assume_not_nan:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_fract_f64_e32 v[2:3], v[0:1]
; GFX6-NEXT: v_mov_b32_e32 v4, -1
; GFX6-NEXT: v_mov_b32_e32 v5, 0x3fefffff
; GFX6-NEXT: v_min_f64 v[2:3], v[2:3], v[4:5]
; GFX6-NEXT: v_cmp_class_f64_e64 vcc, v[0:1], 3
; GFX6-NEXT: s_mov_b32 s4, -1
; GFX6-NEXT: v_cndmask_b32_e32 v2, v2, v0, vcc
; GFX6-NEXT: v_cndmask_b32_e32 v3, v3, v1, vcc
; GFX6-NEXT: v_add_f64 v[2:3], v[0:1], -v[2:3]
; GFX6-NEXT: s_mov_b32 s5, 0x3fefffff
; GFX6-NEXT: v_add_f64 v[2:3], v[0:1], -v[2:3]
; GFX6-NEXT: v_min_f64 v[2:3], v[2:3], s[4:5]
; GFX6-NEXT: s_mov_b32 s4, 0
; GFX6-NEXT: s_mov_b32 s5, 0x7ff00000
; GFX6-NEXT: v_cmp_neq_f64_e64 vcc, |v[0:1]|, s[4:5]
; GFX6-NEXT: v_cndmask_b32_e32 v0, 0, v2, vcc
; GFX6-NEXT: v_cndmask_b32_e32 v1, 0, v3, vcc
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: fract_match_f64_assume_not_nan:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: s_mov_b32 s4, 0
; GFX7-NEXT: s_mov_b32 s5, 0x7ff00000
; GFX7-NEXT: v_fract_f64_e32 v[2:3], v[0:1]
; GFX7-NEXT: v_cmp_neq_f64_e64 vcc, |v[0:1]|, s[4:5]
; GFX7-NEXT: v_cndmask_b32_e32 v0, 0, v2, vcc
; GFX7-NEXT: v_cndmask_b32_e32 v1, 0, v3, vcc
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: fract_match_f64_assume_not_nan:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: s_mov_b32 s4, 0
; GFX8-NEXT: s_mov_b32 s5, 0x7ff00000
; GFX8-NEXT: v_fract_f64_e32 v[2:3], v[0:1]
; GFX8-NEXT: v_cmp_neq_f64_e64 vcc, |v[0:1]|, s[4:5]
; GFX8-NEXT: v_cndmask_b32_e32 v0, 0, v2, vcc
; GFX8-NEXT: v_cndmask_b32_e32 v1, 0, v3, vcc
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: fract_match_f64_assume_not_nan:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f64_e32 v[2:3], v[0:1]
; GFX11-NEXT: v_cmp_neq_f64_e64 vcc_lo, 0x7ff00000, |v[0:1]|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX11-NEXT: v_dual_cndmask_b32 v0, 0, v2 :: v_dual_cndmask_b32 v1, 0, v3
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: fract_match_f64_assume_not_nan:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f64_e32 v[2:3], v[0:1]
; GFX12-NEXT: v_cmp_neq_f64_e64 vcc_lo, 0x7ff00000, |v[0:1]|
; GFX12-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX12-NEXT: v_dual_cndmask_b32 v0, 0, v2 :: v_dual_cndmask_b32 v1, 0, v3
; GFX12-NEXT: s_setpc_b64 s[30:31]
; GFX6-IR-LABEL: define double @fract_match_f64_assume_not_nan(
; GFX6-IR-SAME: double [[X:%.*]]) #[[ATTR1:[0-9]+]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[IS_ORD:%.*]] = fcmp ord double [[X]], 0.000000e+00
; GFX6-IR-NEXT: tail call void @llvm.assume(i1 [[IS_ORD]])
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call double @llvm.floor.f64(double [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub double [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call double @llvm.minnum.f64(double [[SUB]], double 0x3FEFFFFFFFFFFFFF)
; GFX6-IR-NEXT: [[X_ABS:%.*]] = tail call double @llvm.fabs.f64(double [[X]])
; GFX6-IR-NEXT: [[IS_INF:%.*]] = fcmp oeq double [[X_ABS]], 0x7FF0000000000000
; GFX6-IR-NEXT: [[RESULT:%.*]] = select i1 [[IS_INF]], double 0.000000e+00, double [[MIN]]
; GFX6-IR-NEXT: ret double [[RESULT]]
; IR-FRACT-LABEL: define double @fract_match_f64_assume_not_nan(
; IR-FRACT-SAME: double [[X:%.*]]) #[[ATTR1:[0-9]+]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[IS_ORD:%.*]] = fcmp ord double [[X]], 0.000000e+00
; IR-FRACT-NEXT: tail call void @llvm.assume(i1 [[IS_ORD]])
; IR-FRACT-NEXT: [[MIN:%.*]] = call nnan double @llvm.amdgcn.fract.f64(double [[X]])
; IR-FRACT-NEXT: [[X_ABS:%.*]] = tail call double @llvm.fabs.f64(double [[X]])
; IR-FRACT-NEXT: [[IS_INF:%.*]] = fcmp oeq double [[X_ABS]], 0x7FF0000000000000
; IR-FRACT-NEXT: [[RESULT:%.*]] = select i1 [[IS_INF]], double 0.000000e+00, double [[MIN]]
; IR-FRACT-NEXT: ret double [[RESULT]]
entry:
%is.ord = fcmp ord double %x, 0.000000e+00
tail call void @llvm.assume(i1 %is.ord)
%floor = tail call double @llvm.floor.f64(double %x)
%sub = fsub double %x, %floor
%min = tail call double @llvm.minnum.f64(double %sub, double 0x3FEFFFFFFFFFFFFF)
%x.abs = tail call double @llvm.fabs.f64(double %x)
%is.inf = fcmp oeq double %x.abs, 0x7FF0000000000000
%result = select i1 %is.inf, double 0.0, double %min
ret double %result
}
; isinf before isnan
define float @safe_math_fract_f32_swapped_edge_case(float %x) #0 {
; GFX6-IR-LABEL: define float @safe_math_fract_f32_swapped_edge_case(
; GFX6-IR-SAME: float [[X:%.*]]) #[[ATTR1]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = call float @llvm.floor.f32(float [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; GFX6-IR-NEXT: [[X_FABS:%.*]] = call float @llvm.fabs.f32(float [[X]])
; GFX6-IR-NEXT: [[NOT_INF:%.*]] = fcmp une float [[X_FABS]], 0x7FF0000000000000
; GFX6-IR-NEXT: [[COND:%.*]] = select i1 [[NOT_INF]], float [[MIN]], float 0.000000e+00
; GFX6-IR-NEXT: [[NOT_NAN:%.*]] = fcmp ord float [[X]], 0.000000e+00
; GFX6-IR-NEXT: [[COND8:%.*]] = select i1 [[NOT_NAN]], float [[COND]], float [[X]]
; GFX6-IR-NEXT: ret float [[COND8]]
;
; IR-FRACT-LABEL: define float @safe_math_fract_f32_swapped_edge_case(
; IR-FRACT-SAME: float [[X:%.*]]) #[[ATTR1]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[COND:%.*]] = call float @llvm.amdgcn.fract.f32(float [[X]])
; IR-FRACT-NEXT: [[X_FABS:%.*]] = call float @llvm.fabs.f32(float [[X]])
; IR-FRACT-NEXT: [[NOT_INF:%.*]] = fcmp une float [[X_FABS]], 0x7FF0000000000000
; IR-FRACT-NEXT: [[COND8:%.*]] = select i1 [[NOT_INF]], float [[COND]], float 0.000000e+00
; IR-FRACT-NEXT: ret float [[COND8]]
;
; GFX6-LABEL: safe_math_fract_f32_swapped_edge_case:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v1, v0, v1
; GFX6-NEXT: s_mov_b32 s4, 0x7f800000
; GFX6-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v1
; GFX6-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX6-NEXT: v_cndmask_b32_e32 v1, 0, v1, vcc
; GFX6-NEXT: v_cmp_o_f32_e32 vcc, v0, v0
; GFX6-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: safe_math_fract_f32_swapped_edge_case:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: s_mov_b32 s4, 0x7f800000
; GFX7-NEXT: v_fract_f32_e32 v1, v0
; GFX7-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX7-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: safe_math_fract_f32_swapped_edge_case:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: s_mov_b32 s4, 0x7f800000
; GFX8-NEXT: v_fract_f32_e32 v1, v0
; GFX8-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX8-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: safe_math_fract_f32_swapped_edge_case:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v1, v0
; GFX11-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX11-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc_lo
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: safe_math_fract_f32_swapped_edge_case:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v1, v0
; GFX12-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX12-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX12-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc_lo
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = call float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
%x.fabs = call float @llvm.fabs.f32(float %x)
%not.inf = fcmp une float %x.fabs, 0x7FF0000000000000
%cond = select i1 %not.inf, float %min, float 0.0
%not.nan = fcmp ord float %x, 0.0
%cond8 = select i1 %not.nan, float %cond, float %x
ret float %cond8
}
define float @safe_math_fract_f32_swapped_edge_case_multi_use_inner_select(float %x, ptr addrspace(1) %ptr) #0 {
; GFX6-LABEL: safe_math_fract_f32_swapped_edge_case_multi_use_inner_select:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v3, v0
; GFX6-NEXT: v_sub_f32_e32 v3, v0, v3
; GFX6-NEXT: s_mov_b32 s8, 0x7f800000
; GFX6-NEXT: v_min_f32_e32 v3, 0x3f7fffff, v3
; GFX6-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s8
; GFX6-NEXT: s_mov_b32 s6, 0
; GFX6-NEXT: v_cndmask_b32_e32 v3, 0, v3, vcc
; GFX6-NEXT: v_cmp_o_f32_e32 vcc, v0, v0
; GFX6-NEXT: s_mov_b32 s7, 0xf000
; GFX6-NEXT: s_mov_b32 s4, s6
; GFX6-NEXT: s_mov_b32 s5, s6
; GFX6-NEXT: v_cndmask_b32_e32 v0, v0, v3, vcc
; GFX6-NEXT: buffer_store_dword v3, v[1:2], s[4:7], 0 addr64
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: safe_math_fract_f32_swapped_edge_case_multi_use_inner_select:
; GFX7: ; %bb.0:
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: s_mov_b32 s8, 0x7f800000
; GFX7-NEXT: s_mov_b32 s6, 0
; GFX7-NEXT: v_fract_f32_e32 v3, v0
; GFX7-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s8
; GFX7-NEXT: s_mov_b32 s7, 0xf000
; GFX7-NEXT: s_mov_b32 s4, s6
; GFX7-NEXT: s_mov_b32 s5, s6
; GFX7-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc
; GFX7-NEXT: buffer_store_dword v0, v[1:2], s[4:7], 0 addr64
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: safe_math_fract_f32_swapped_edge_case_multi_use_inner_select:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: s_mov_b32 s4, 0x7f800000
; GFX8-NEXT: v_fract_f32_e32 v3, v0
; GFX8-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX8-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc
; GFX8-NEXT: global_store_dword v[1:2], v0, off
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: safe_math_fract_f32_swapped_edge_case_multi_use_inner_select:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v3, v0
; GFX11-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX11-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc_lo
; GFX11-NEXT: global_store_b32 v[1:2], v0, off
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: safe_math_fract_f32_swapped_edge_case_multi_use_inner_select:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v3, v0
; GFX12-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX12-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX12-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc_lo
; GFX12-NEXT: global_store_b32 v[1:2], v0, off
; GFX12-NEXT: s_setpc_b64 s[30:31]
; GFX6-IR-LABEL: define float @safe_math_fract_f32_swapped_edge_case_multi_use_inner_select(
; GFX6-IR-SAME: float [[X:%.*]], ptr addrspace(1) [[PTR:%.*]]) #[[ATTR1]] {
; GFX6-IR-NEXT: [[FLOOR:%.*]] = call float @llvm.floor.f32(float [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; GFX6-IR-NEXT: [[X_FABS:%.*]] = call float @llvm.fabs.f32(float [[X]])
; GFX6-IR-NEXT: [[NOT_INF:%.*]] = fcmp une float [[X_FABS]], 0x7FF0000000000000
; GFX6-IR-NEXT: [[COND:%.*]] = select i1 [[NOT_INF]], float [[MIN]], float 0.000000e+00
; GFX6-IR-NEXT: store float [[COND]], ptr addrspace(1) [[PTR]], align 4
; GFX6-IR-NEXT: [[NOT_NAN:%.*]] = fcmp ord float [[X]], 0.000000e+00
; GFX6-IR-NEXT: [[COND8:%.*]] = select i1 [[NOT_NAN]], float [[COND]], float [[X]]
; GFX6-IR-NEXT: ret float [[COND8]]
; IR-FRACT-LABEL: define float @safe_math_fract_f32_swapped_edge_case_multi_use_inner_select(
; IR-FRACT-SAME: float [[X:%.*]], ptr addrspace(1) [[PTR:%.*]]) #[[ATTR1]] {
; IR-FRACT-NEXT: [[COND:%.*]] = call float @llvm.amdgcn.fract.f32(float [[X]])
; IR-FRACT-NEXT: [[X_FABS:%.*]] = call float @llvm.fabs.f32(float [[X]])
; IR-FRACT-NEXT: [[NOT_INF:%.*]] = fcmp une float [[X_FABS]], 0x7FF0000000000000
; IR-FRACT-NEXT: [[COND8:%.*]] = select i1 [[NOT_INF]], float [[COND]], float 0.000000e+00
; IR-FRACT-NEXT: store float [[COND8]], ptr addrspace(1) [[PTR]], align 4
; IR-FRACT-NEXT: ret float [[COND8]]
%floor = call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = call float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
%x.fabs = call float @llvm.fabs.f32(float %x)
%not.inf = fcmp une float %x.fabs, 0x7FF0000000000000
%cond = select i1 %not.inf, float %min, float 0.0
store float %cond, ptr addrspace(1) %ptr
%not.nan = fcmp ord float %x, 0.0
%cond8 = select i1 %not.nan, float %cond, float %x
ret float %cond8
}
define float @safe_math_fract_f32_swapped_edge_case_multi_use_inner_select_fcmp(float %x, ptr addrspace(1) %ptr) #0 {
; GFX6-LABEL: safe_math_fract_f32_swapped_edge_case_multi_use_inner_select_fcmp:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v3, v0
; GFX6-NEXT: s_mov_b32 s8, 0x7f800000
; GFX6-NEXT: s_mov_b32 s6, 0
; GFX6-NEXT: v_sub_f32_e32 v3, v0, v3
; GFX6-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s8
; GFX6-NEXT: s_mov_b32 s7, 0xf000
; GFX6-NEXT: s_mov_b32 s4, s6
; GFX6-NEXT: s_mov_b32 s5, s6
; GFX6-NEXT: v_min_f32_e32 v3, 0x3f7fffff, v3
; GFX6-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
; GFX6-NEXT: buffer_store_byte v4, v[1:2], s[4:7], 0 addr64
; GFX6-NEXT: v_cndmask_b32_e32 v1, 0, v3, vcc
; GFX6-NEXT: v_cmp_o_f32_e32 vcc, v0, v0
; GFX6-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: safe_math_fract_f32_swapped_edge_case_multi_use_inner_select_fcmp:
; GFX7: ; %bb.0:
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: s_mov_b32 s8, 0x7f800000
; GFX7-NEXT: s_mov_b32 s6, 0
; GFX7-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s8
; GFX7-NEXT: s_mov_b32 s7, 0xf000
; GFX7-NEXT: s_mov_b32 s4, s6
; GFX7-NEXT: s_mov_b32 s5, s6
; GFX7-NEXT: v_fract_f32_e32 v3, v0
; GFX7-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; GFX7-NEXT: buffer_store_byte v0, v[1:2], s[4:7], 0 addr64
; GFX7-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: safe_math_fract_f32_swapped_edge_case_multi_use_inner_select_fcmp:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: s_mov_b32 s4, 0x7f800000
; GFX8-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX8-NEXT: v_fract_f32_e32 v3, v0
; GFX8-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; GFX8-NEXT: global_store_byte v[1:2], v0, off
; GFX8-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: safe_math_fract_f32_swapped_edge_case_multi_use_inner_select_fcmp:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v3, v0
; GFX11-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX11-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc_lo
; GFX11-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc_lo
; GFX11-NEXT: global_store_b8 v[1:2], v4, off
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: safe_math_fract_f32_swapped_edge_case_multi_use_inner_select_fcmp:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v3, v0
; GFX12-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX12-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX12-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc_lo
; GFX12-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc_lo
; GFX12-NEXT: global_store_b8 v[1:2], v4, off
; GFX12-NEXT: s_setpc_b64 s[30:31]
; GFX6-IR-LABEL: define float @safe_math_fract_f32_swapped_edge_case_multi_use_inner_select_fcmp(
; GFX6-IR-SAME: float [[X:%.*]], ptr addrspace(1) [[PTR:%.*]]) #[[ATTR1]] {
; GFX6-IR-NEXT: [[FLOOR:%.*]] = call float @llvm.floor.f32(float [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; GFX6-IR-NEXT: [[X_FABS:%.*]] = call float @llvm.fabs.f32(float [[X]])
; GFX6-IR-NEXT: [[NOT_INF:%.*]] = fcmp une float [[X_FABS]], 0x7FF0000000000000
; GFX6-IR-NEXT: store i1 [[NOT_INF]], ptr addrspace(1) [[PTR]], align 1
; GFX6-IR-NEXT: [[COND:%.*]] = select i1 [[NOT_INF]], float [[MIN]], float 0.000000e+00
; GFX6-IR-NEXT: [[NOT_NAN:%.*]] = fcmp ord float [[X]], 0.000000e+00
; GFX6-IR-NEXT: [[COND8:%.*]] = select i1 [[NOT_NAN]], float [[COND]], float [[X]]
; GFX6-IR-NEXT: ret float [[COND8]]
; IR-FRACT-LABEL: define float @safe_math_fract_f32_swapped_edge_case_multi_use_inner_select_fcmp(
; IR-FRACT-SAME: float [[X:%.*]], ptr addrspace(1) [[PTR:%.*]]) #[[ATTR1]] {
; IR-FRACT-NEXT: [[COND:%.*]] = call float @llvm.amdgcn.fract.f32(float [[X]])
; IR-FRACT-NEXT: [[X_FABS:%.*]] = call float @llvm.fabs.f32(float [[X]])
; IR-FRACT-NEXT: [[NOT_INF:%.*]] = fcmp une float [[X_FABS]], 0x7FF0000000000000
; IR-FRACT-NEXT: store i1 [[NOT_INF]], ptr addrspace(1) [[PTR]], align 1
; IR-FRACT-NEXT: [[COND8:%.*]] = select i1 [[NOT_INF]], float [[COND]], float 0.000000e+00
; IR-FRACT-NEXT: ret float [[COND8]]
%floor = call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = call float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
%x.fabs = call float @llvm.fabs.f32(float %x)
%not.inf = fcmp une float %x.fabs, 0x7FF0000000000000
store i1 %not.inf, ptr addrspace(1) %ptr
%cond = select i1 %not.inf, float %min, float 0.0
%not.nan = fcmp ord float %x, 0.0
%cond8 = select i1 %not.nan, float %cond, float %x
ret float %cond8
}
define float @safe_math_fract_f32_swapped_edge_case_multi_use_fabs(float %x, ptr addrspace(1) %ptr) #0 {
; GFX6-LABEL: safe_math_fract_f32_swapped_edge_case_multi_use_fabs:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: s_mov_b32 s6, 0
; GFX6-NEXT: s_mov_b32 s7, 0xf000
; GFX6-NEXT: s_mov_b32 s4, s6
; GFX6-NEXT: s_mov_b32 s5, s6
; GFX6-NEXT: v_floor_f32_e32 v3, v0
; GFX6-NEXT: v_and_b32_e32 v4, 0x7fffffff, v0
; GFX6-NEXT: v_sub_f32_e32 v3, v0, v3
; GFX6-NEXT: buffer_store_dword v4, v[1:2], s[4:7], 0 addr64
; GFX6-NEXT: s_mov_b32 s4, 0x7f800000
; GFX6-NEXT: v_min_f32_e32 v3, 0x3f7fffff, v3
; GFX6-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX6-NEXT: v_cndmask_b32_e32 v1, 0, v3, vcc
; GFX6-NEXT: v_cmp_o_f32_e32 vcc, v0, v0
; GFX6-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: safe_math_fract_f32_swapped_edge_case_multi_use_fabs:
; GFX7: ; %bb.0:
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: s_mov_b32 s6, 0
; GFX7-NEXT: s_mov_b32 s7, 0xf000
; GFX7-NEXT: s_mov_b32 s4, s6
; GFX7-NEXT: s_mov_b32 s5, s6
; GFX7-NEXT: v_and_b32_e32 v4, 0x7fffffff, v0
; GFX7-NEXT: buffer_store_dword v4, v[1:2], s[4:7], 0 addr64
; GFX7-NEXT: s_mov_b32 s4, 0x7f800000
; GFX7-NEXT: v_fract_f32_e32 v3, v0
; GFX7-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX7-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: safe_math_fract_f32_swapped_edge_case_multi_use_fabs:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: s_mov_b32 s4, 0x7f800000
; GFX8-NEXT: v_fract_f32_e32 v3, v0
; GFX8-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX8-NEXT: v_and_b32_e32 v4, 0x7fffffff, v0
; GFX8-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc
; GFX8-NEXT: global_store_dword v[1:2], v4, off
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: safe_math_fract_f32_swapped_edge_case_multi_use_fabs:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v3, v0
; GFX11-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX11-NEXT: v_and_b32_e32 v4, 0x7fffffff, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_3)
; GFX11-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc_lo
; GFX11-NEXT: global_store_b32 v[1:2], v4, off
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: safe_math_fract_f32_swapped_edge_case_multi_use_fabs:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v3, v0
; GFX12-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX12-NEXT: v_and_b32_e32 v4, 0x7fffffff, v0
; GFX12-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_3)
; GFX12-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc_lo
; GFX12-NEXT: global_store_b32 v[1:2], v4, off
; GFX12-NEXT: s_setpc_b64 s[30:31]
; GFX6-IR-LABEL: define float @safe_math_fract_f32_swapped_edge_case_multi_use_fabs(
; GFX6-IR-SAME: float [[X:%.*]], ptr addrspace(1) [[PTR:%.*]]) #[[ATTR1]] {
; GFX6-IR-NEXT: [[FLOOR:%.*]] = call float @llvm.floor.f32(float [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; GFX6-IR-NEXT: [[X_FABS:%.*]] = call float @llvm.fabs.f32(float [[X]])
; GFX6-IR-NEXT: store float [[X_FABS]], ptr addrspace(1) [[PTR]], align 4
; GFX6-IR-NEXT: [[NOT_INF:%.*]] = fcmp une float [[X_FABS]], 0x7FF0000000000000
; GFX6-IR-NEXT: [[COND:%.*]] = select i1 [[NOT_INF]], float [[MIN]], float 0.000000e+00
; GFX6-IR-NEXT: [[NOT_NAN:%.*]] = fcmp ord float [[X]], 0.000000e+00
; GFX6-IR-NEXT: [[COND8:%.*]] = select i1 [[NOT_NAN]], float [[COND]], float [[X]]
; GFX6-IR-NEXT: ret float [[COND8]]
; IR-FRACT-LABEL: define float @safe_math_fract_f32_swapped_edge_case_multi_use_fabs(
; IR-FRACT-SAME: float [[X:%.*]], ptr addrspace(1) [[PTR:%.*]]) #[[ATTR1]] {
; IR-FRACT-NEXT: [[COND:%.*]] = call float @llvm.amdgcn.fract.f32(float [[X]])
; IR-FRACT-NEXT: [[X_FABS:%.*]] = call float @llvm.fabs.f32(float [[X]])
; IR-FRACT-NEXT: store float [[X_FABS]], ptr addrspace(1) [[PTR]], align 4
; IR-FRACT-NEXT: [[NOT_INF:%.*]] = fcmp une float [[X_FABS]], 0x7FF0000000000000
; IR-FRACT-NEXT: [[COND8:%.*]] = select i1 [[NOT_INF]], float [[COND]], float 0.000000e+00
; IR-FRACT-NEXT: ret float [[COND8]]
%floor = call float @llvm.floor.f32(float %x) #3
%sub = fsub float %x, %floor
%min = call float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
%x.fabs = call float @llvm.fabs.f32(float %x)
store float %x.fabs, ptr addrspace(1) %ptr
%not.inf = fcmp une float %x.fabs, 0x7FF0000000000000
%cond = select i1 %not.inf, float %min, float 0.0
%not.nan = fcmp ord float %x, 0.0
%cond8 = select i1 %not.nan, float %cond, float %x
ret float %cond8
}
define float @safe_math_fract_f32_swapped_edge_case_wrong_compared(float %x, float %wrong) #0 {
; IR-LABEL: define float @safe_math_fract_f32_swapped_edge_case_wrong_compared(
; IR-SAME: float [[X:%.*]], float [[WRONG:%.*]]) #[[ATTR1:[0-9]+]] {
; IR-NEXT: [[ENTRY:.*:]]
; IR-NEXT: [[FLOOR:%.*]] = call float @llvm.floor.f32(float [[X]])
; IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; IR-NEXT: [[MIN:%.*]] = call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; IR-NEXT: [[WRONG_FABS:%.*]] = call float @llvm.fabs.f32(float [[WRONG]])
; IR-NEXT: [[NOT_INF:%.*]] = fcmp une float [[WRONG_FABS]], 0x7FF0000000000000
; IR-NEXT: [[COND:%.*]] = select i1 [[NOT_INF]], float [[MIN]], float 0.000000e+00
; IR-NEXT: [[NOT_NAN:%.*]] = fcmp ord float [[X]], 0.000000e+00
; IR-NEXT: [[COND8:%.*]] = select i1 [[NOT_NAN]], float [[COND]], float [[X]]
; IR-NEXT: ret float [[COND8]]
;
; GFX6-LABEL: safe_math_fract_f32_swapped_edge_case_wrong_compared:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v2, v0
; GFX6-NEXT: v_sub_f32_e32 v2, v0, v2
; GFX6-NEXT: s_mov_b32 s4, 0x7f800000
; GFX6-NEXT: v_min_f32_e32 v2, 0x3f7fffff, v2
; GFX6-NEXT: v_cmp_neq_f32_e64 vcc, |v1|, s4
; GFX6-NEXT: v_cndmask_b32_e32 v1, 0, v2, vcc
; GFX6-NEXT: v_cmp_o_f32_e32 vcc, v0, v0
; GFX6-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: safe_math_fract_f32_swapped_edge_case_wrong_compared:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_floor_f32_e32 v2, v0
; GFX7-NEXT: v_sub_f32_e32 v2, v0, v2
; GFX7-NEXT: s_mov_b32 s4, 0x7f800000
; GFX7-NEXT: v_min_f32_e32 v2, 0x3f7fffff, v2
; GFX7-NEXT: v_cmp_neq_f32_e64 vcc, |v1|, s4
; GFX7-NEXT: v_cndmask_b32_e32 v1, 0, v2, vcc
; GFX7-NEXT: v_cmp_o_f32_e32 vcc, v0, v0
; GFX7-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: safe_math_fract_f32_swapped_edge_case_wrong_compared:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_floor_f32_e32 v2, v0
; GFX8-NEXT: v_sub_f32_e32 v2, v0, v2
; GFX8-NEXT: s_mov_b32 s4, 0x7f800000
; GFX8-NEXT: v_min_f32_e32 v2, 0x3f7fffff, v2
; GFX8-NEXT: v_cmp_neq_f32_e64 vcc, |v1|, s4
; GFX8-NEXT: v_cndmask_b32_e32 v1, 0, v2, vcc
; GFX8-NEXT: v_cmp_o_f32_e32 vcc, v0, v0
; GFX8-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: safe_math_fract_f32_swapped_edge_case_wrong_compared:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_floor_f32_e32 v2, v0
; GFX11-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v1|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_sub_f32_e32 v2, v0, v2
; GFX11-NEXT: v_min_f32_e32 v2, 0x3f7fffff, v2
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX11-NEXT: v_cndmask_b32_e32 v1, 0, v2, vcc_lo
; GFX11-NEXT: v_cmp_o_f32_e32 vcc_lo, v0, v0
; GFX11-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc_lo
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: safe_math_fract_f32_swapped_edge_case_wrong_compared:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_floor_f32_e32 v2, v0
; GFX12-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v1|
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-NEXT: v_sub_f32_e32 v2, v0, v2
; GFX12-NEXT: v_min_num_f32_e32 v2, 0x3f7fffff, v2
; GFX12-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
; GFX12-NEXT: v_cndmask_b32_e32 v1, 0, v2, vcc_lo
; GFX12-NEXT: v_cmp_o_f32_e32 vcc_lo, v0, v0
; GFX12-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc_lo
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = call float @llvm.floor.f32(float %x) #3
%sub = fsub float %x, %floor
%min = call float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
%wrong.fabs = call float @llvm.fabs.f32(float %wrong)
%not.inf = fcmp une float %wrong.fabs, 0x7FF0000000000000
%cond = select i1 %not.inf, float %min, float 0.0
%not.nan = fcmp ord float %x, 0.0
%cond8 = select i1 %not.nan, float %cond, float %x
ret float %cond8
}
define float @safe_math_fract_f32_swapped_edge_case_commute_inf_check(float %x) #0 {
; IR-LABEL: define float @safe_math_fract_f32_swapped_edge_case_commute_inf_check(
; IR-SAME: float [[X:%.*]]) #[[ATTR1]] {
; IR-NEXT: [[ENTRY:.*:]]
; IR-NEXT: [[FLOOR:%.*]] = call float @llvm.floor.f32(float [[X]])
; IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; IR-NEXT: [[MIN:%.*]] = call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; IR-NEXT: [[X_FABS:%.*]] = call float @llvm.fabs.f32(float [[X]])
; IR-NEXT: [[IS_INF:%.*]] = fcmp oeq float [[X_FABS]], 0x7FF0000000000000
; IR-NEXT: [[COND:%.*]] = select i1 [[IS_INF]], float 0.000000e+00, float [[MIN]]
; IR-NEXT: [[NOT_NAN:%.*]] = fcmp ord float [[X]], 0.000000e+00
; IR-NEXT: [[COND8:%.*]] = select i1 [[NOT_NAN]], float [[COND]], float [[X]]
; IR-NEXT: ret float [[COND8]]
;
; GFX6-LABEL: safe_math_fract_f32_swapped_edge_case_commute_inf_check:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v1, v0, v1
; GFX6-NEXT: s_mov_b32 s4, 0x7f800000
; GFX6-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v1
; GFX6-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX6-NEXT: v_cndmask_b32_e32 v1, 0, v1, vcc
; GFX6-NEXT: v_cmp_o_f32_e32 vcc, v0, v0
; GFX6-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: safe_math_fract_f32_swapped_edge_case_commute_inf_check:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_floor_f32_e32 v1, v0
; GFX7-NEXT: v_sub_f32_e32 v1, v0, v1
; GFX7-NEXT: s_mov_b32 s4, 0x7f800000
; GFX7-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v1
; GFX7-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX7-NEXT: v_cndmask_b32_e32 v1, 0, v1, vcc
; GFX7-NEXT: v_cmp_o_f32_e32 vcc, v0, v0
; GFX7-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: safe_math_fract_f32_swapped_edge_case_commute_inf_check:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_floor_f32_e32 v1, v0
; GFX8-NEXT: v_sub_f32_e32 v1, v0, v1
; GFX8-NEXT: s_mov_b32 s4, 0x7f800000
; GFX8-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v1
; GFX8-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX8-NEXT: v_cndmask_b32_e32 v1, 0, v1, vcc
; GFX8-NEXT: v_cmp_o_f32_e32 vcc, v0, v0
; GFX8-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: safe_math_fract_f32_swapped_edge_case_commute_inf_check:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_floor_f32_e32 v1, v0
; GFX11-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_sub_f32_e32 v1, v0, v1
; GFX11-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v1
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX11-NEXT: v_cndmask_b32_e32 v1, 0, v1, vcc_lo
; GFX11-NEXT: v_cmp_o_f32_e32 vcc_lo, v0, v0
; GFX11-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc_lo
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: safe_math_fract_f32_swapped_edge_case_commute_inf_check:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_floor_f32_e32 v1, v0
; GFX12-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-NEXT: v_sub_f32_e32 v1, v0, v1
; GFX12-NEXT: v_min_num_f32_e32 v1, 0x3f7fffff, v1
; GFX12-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
; GFX12-NEXT: v_cndmask_b32_e32 v1, 0, v1, vcc_lo
; GFX12-NEXT: v_cmp_o_f32_e32 vcc_lo, v0, v0
; GFX12-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc_lo
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = call float @llvm.floor.f32(float %x) #3
%sub = fsub float %x, %floor
%min = call float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
%x.fabs = call float @llvm.fabs.f32(float %x)
%is.inf = fcmp oeq float %x.fabs, 0x7FF0000000000000
%cond = select i1 %is.inf, float 0.0, float %min
%not.nan = fcmp ord float %x, 0.0
%cond8 = select i1 %not.nan, float %cond, float %x
ret float %cond8
}
define float @safe_math_fract_f32_swapped_edge_case_commute_nan_check(float %x) #0 {
; IR-LABEL: define float @safe_math_fract_f32_swapped_edge_case_commute_nan_check(
; IR-SAME: float [[X:%.*]]) #[[ATTR1]] {
; IR-NEXT: [[ENTRY:.*:]]
; IR-NEXT: [[FLOOR:%.*]] = call float @llvm.floor.f32(float [[X]])
; IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; IR-NEXT: [[MIN:%.*]] = call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; IR-NEXT: [[X_FABS:%.*]] = call float @llvm.fabs.f32(float [[X]])
; IR-NEXT: [[NOT_INF:%.*]] = fcmp une float [[X_FABS]], 0x7FF0000000000000
; IR-NEXT: [[COND:%.*]] = select i1 [[NOT_INF]], float [[MIN]], float 0.000000e+00
; IR-NEXT: [[IS_NAN:%.*]] = fcmp uno float [[X]], 0.000000e+00
; IR-NEXT: [[COND8:%.*]] = select i1 [[IS_NAN]], float [[X]], float [[COND]]
; IR-NEXT: ret float [[COND8]]
;
; GFX6-LABEL: safe_math_fract_f32_swapped_edge_case_commute_nan_check:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v1, v0, v1
; GFX6-NEXT: s_mov_b32 s4, 0x7f800000
; GFX6-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v1
; GFX6-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX6-NEXT: v_cndmask_b32_e32 v1, 0, v1, vcc
; GFX6-NEXT: v_cmp_u_f32_e32 vcc, v0, v0
; GFX6-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: safe_math_fract_f32_swapped_edge_case_commute_nan_check:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_floor_f32_e32 v1, v0
; GFX7-NEXT: v_sub_f32_e32 v1, v0, v1
; GFX7-NEXT: s_mov_b32 s4, 0x7f800000
; GFX7-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v1
; GFX7-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX7-NEXT: v_cndmask_b32_e32 v1, 0, v1, vcc
; GFX7-NEXT: v_cmp_u_f32_e32 vcc, v0, v0
; GFX7-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: safe_math_fract_f32_swapped_edge_case_commute_nan_check:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_floor_f32_e32 v1, v0
; GFX8-NEXT: v_sub_f32_e32 v1, v0, v1
; GFX8-NEXT: s_mov_b32 s4, 0x7f800000
; GFX8-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v1
; GFX8-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX8-NEXT: v_cndmask_b32_e32 v1, 0, v1, vcc
; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v0, v0
; GFX8-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: safe_math_fract_f32_swapped_edge_case_commute_nan_check:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_floor_f32_e32 v1, v0
; GFX11-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_sub_f32_e32 v1, v0, v1
; GFX11-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v1
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX11-NEXT: v_cndmask_b32_e32 v1, 0, v1, vcc_lo
; GFX11-NEXT: v_cmp_u_f32_e32 vcc_lo, v0, v0
; GFX11-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc_lo
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: safe_math_fract_f32_swapped_edge_case_commute_nan_check:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_floor_f32_e32 v1, v0
; GFX12-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-NEXT: v_sub_f32_e32 v1, v0, v1
; GFX12-NEXT: v_min_num_f32_e32 v1, 0x3f7fffff, v1
; GFX12-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
; GFX12-NEXT: v_cndmask_b32_e32 v1, 0, v1, vcc_lo
; GFX12-NEXT: v_cmp_u_f32_e32 vcc_lo, v0, v0
; GFX12-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc_lo
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = call float @llvm.floor.f32(float %x) #3
%sub = fsub float %x, %floor
%min = call float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
%x.fabs = call float @llvm.fabs.f32(float %x)
%not.inf = fcmp une float %x.fabs, 0x7FF0000000000000
%cond = select i1 %not.inf, float %min, float 0.0
%is.nan = fcmp uno float %x, 0.0
%cond8 = select i1 %is.nan, float %x, float %cond
ret float %cond8
}
; Cannot match against -inf, must be +inf
define float @safe_math_fract_f32_swapped_edge_case_cmp_neg_inf(float %x) #0 {
; IR-LABEL: define float @safe_math_fract_f32_swapped_edge_case_cmp_neg_inf(
; IR-SAME: float [[X:%.*]]) #[[ATTR1]] {
; IR-NEXT: [[ENTRY:.*:]]
; IR-NEXT: [[FLOOR:%.*]] = call float @llvm.floor.f32(float [[X]])
; IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; IR-NEXT: [[MIN:%.*]] = call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; IR-NEXT: [[X_FABS:%.*]] = call float @llvm.fabs.f32(float [[X]])
; IR-NEXT: [[NOT_INF:%.*]] = fcmp une float [[X_FABS]], 0xFFF0000000000000
; IR-NEXT: [[COND:%.*]] = select i1 [[NOT_INF]], float [[MIN]], float 0.000000e+00
; IR-NEXT: [[NOT_NAN:%.*]] = fcmp ord float [[X]], 0.000000e+00
; IR-NEXT: [[COND8:%.*]] = select i1 [[NOT_NAN]], float [[COND]], float [[X]]
; IR-NEXT: ret float [[COND8]]
;
; GFX6-LABEL: safe_math_fract_f32_swapped_edge_case_cmp_neg_inf:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v1, v0, v1
; GFX6-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v1
; GFX6-NEXT: v_cmp_o_f32_e32 vcc, v0, v0
; GFX6-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: safe_math_fract_f32_swapped_edge_case_cmp_neg_inf:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_fract_f32_e32 v0, v0
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: safe_math_fract_f32_swapped_edge_case_cmp_neg_inf:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_fract_f32_e32 v0, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: safe_math_fract_f32_swapped_edge_case_cmp_neg_inf:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v0, v0
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: safe_math_fract_f32_swapped_edge_case_cmp_neg_inf:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v0, v0
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = call float @llvm.floor.f32(float %x) #3
%sub = fsub float %x, %floor
%min = call float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
%x.fabs = call float @llvm.fabs.f32(float %x)
%not.inf = fcmp une float %x.fabs, 0xFFF0000000000000
%cond = select i1 %not.inf, float %min, float 0.0
%not.nan = fcmp ord float %x, 0.0
%cond8 = select i1 %not.nan, float %cond, float %x
ret float %cond8
}
define float @basic_fract_f32_with_inf_check(float %x) {
; IR-LABEL: define float @basic_fract_f32_with_inf_check(
; IR-SAME: float [[X:%.*]]) #[[ATTR0]] {
; IR-NEXT: [[ENTRY:.*:]]
; IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; IR-NEXT: [[MIN:%.*]] = tail call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; IR-NEXT: [[X_FABS:%.*]] = call float @llvm.fabs.f32(float [[X]])
; IR-NEXT: [[NOT_INF:%.*]] = fcmp une float [[X_FABS]], 0x7FF0000000000000
; IR-NEXT: [[CLAMP_INF_TO_ZERO:%.*]] = select i1 [[NOT_INF]], float [[MIN]], float 0.000000e+00
; IR-NEXT: ret float [[CLAMP_INF_TO_ZERO]]
;
; GFX6-LABEL: basic_fract_f32_with_inf_check:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v1, v0, v1
; GFX6-NEXT: s_mov_b32 s4, 0x7f800000
; GFX6-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v1
; GFX6-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX6-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: basic_fract_f32_with_inf_check:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_floor_f32_e32 v1, v0
; GFX7-NEXT: v_sub_f32_e32 v1, v0, v1
; GFX7-NEXT: s_mov_b32 s4, 0x7f800000
; GFX7-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v1
; GFX7-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX7-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: basic_fract_f32_with_inf_check:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_floor_f32_e32 v1, v0
; GFX8-NEXT: v_sub_f32_e32 v1, v0, v1
; GFX8-NEXT: s_mov_b32 s4, 0x7f800000
; GFX8-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v1
; GFX8-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX8-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: basic_fract_f32_with_inf_check:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_floor_f32_e32 v1, v0
; GFX11-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_sub_f32_e32 v1, v0, v1
; GFX11-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v1
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc_lo
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: basic_fract_f32_with_inf_check:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_floor_f32_e32 v1, v0
; GFX12-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-NEXT: v_sub_f32_e32 v1, v0, v1
; GFX12-NEXT: v_min_num_f32_e32 v1, 0x3f7fffff, v1
; GFX12-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc_lo
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = tail call float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
%x.fabs = call float @llvm.fabs.f32(float %x)
%not.inf = fcmp une float %x.fabs, 0x7FF0000000000000
%clamp.inf.to.zero = select i1 %not.inf, float %min, float 0.0
ret float %clamp.inf.to.zero
}
define float @basic_fract_f32_nonans_with_inf_check(float nofpclass(nan) %x) {
; GFX6-IR-LABEL: define float @basic_fract_f32_nonans_with_inf_check(
; GFX6-IR-SAME: float nofpclass(nan) [[X:%.*]]) #[[ATTR0]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = tail call float @llvm.floor.f32(float [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = tail call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; GFX6-IR-NEXT: [[X_FABS:%.*]] = call float @llvm.fabs.f32(float [[X]])
; GFX6-IR-NEXT: [[NOT_INF:%.*]] = fcmp une float [[X_FABS]], 0x7FF0000000000000
; GFX6-IR-NEXT: [[CLAMP_INF_TO_ZERO:%.*]] = select i1 [[NOT_INF]], float [[MIN]], float 0.000000e+00
; GFX6-IR-NEXT: ret float [[CLAMP_INF_TO_ZERO]]
;
; IR-FRACT-LABEL: define float @basic_fract_f32_nonans_with_inf_check(
; IR-FRACT-SAME: float nofpclass(nan) [[X:%.*]]) #[[ATTR0]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[MIN:%.*]] = call nnan float @llvm.amdgcn.fract.f32(float [[X]])
; IR-FRACT-NEXT: [[X_FABS:%.*]] = call float @llvm.fabs.f32(float [[X]])
; IR-FRACT-NEXT: [[NOT_INF:%.*]] = fcmp une float [[X_FABS]], 0x7FF0000000000000
; IR-FRACT-NEXT: [[CLAMP_INF_TO_ZERO:%.*]] = select i1 [[NOT_INF]], float [[MIN]], float 0.000000e+00
; IR-FRACT-NEXT: ret float [[CLAMP_INF_TO_ZERO]]
;
; GFX6-LABEL: basic_fract_f32_nonans_with_inf_check:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v1, v0, v1
; GFX6-NEXT: s_mov_b32 s4, 0x7f800000
; GFX6-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v1
; GFX6-NEXT: v_cmp_lg_f32_e64 vcc, |v0|, s4
; GFX6-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: basic_fract_f32_nonans_with_inf_check:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: s_mov_b32 s4, 0x7f800000
; GFX7-NEXT: v_fract_f32_e32 v1, v0
; GFX7-NEXT: v_cmp_lg_f32_e64 vcc, |v0|, s4
; GFX7-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: basic_fract_f32_nonans_with_inf_check:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: s_mov_b32 s4, 0x7f800000
; GFX8-NEXT: v_fract_f32_e32 v1, v0
; GFX8-NEXT: v_cmp_lg_f32_e64 vcc, |v0|, s4
; GFX8-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: basic_fract_f32_nonans_with_inf_check:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v1, v0
; GFX11-NEXT: v_cmp_lg_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX11-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc_lo
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: basic_fract_f32_nonans_with_inf_check:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v1, v0
; GFX12-NEXT: v_cmp_lg_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX12-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX12-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc_lo
; GFX12-NEXT: s_setpc_b64 s[30:31]
entry:
%floor = tail call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = tail call float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
%x.fabs = call float @llvm.fabs.f32(float %x)
%not.inf = fcmp une float %x.fabs, 0x7FF0000000000000
%clamp.inf.to.zero = select i1 %not.inf, float %min, float 0.0
ret float %clamp.inf.to.zero
}
@gv = external constant float
; Make sure insert point is safe from non-instructions
define float @safe_math_fract_f32_swapped_edge_case_x_is_const() #0 {
; GFX6-LABEL: safe_math_fract_f32_swapped_edge_case_x_is_const:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v0, gv@abs32@lo
; GFX6-NEXT: s_mov_b32 s4, gv@abs32@lo
; GFX6-NEXT: v_sub_f32_e32 v0, gv@abs32@lo, v0
; GFX6-NEXT: v_mov_b32_e32 v1, 0x7f800000
; GFX6-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX6-NEXT: v_cmp_neq_f32_e64 vcc, |s4|, v1
; GFX6-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc
; GFX6-NEXT: v_mov_b32_e32 v1, gv@abs32@lo
; GFX6-NEXT: v_cmp_o_f32_e64 vcc, s4, s4
; GFX6-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: safe_math_fract_f32_swapped_edge_case_x_is_const:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: s_mov_b32 s4, gv@abs32@lo
; GFX7-NEXT: v_mov_b32_e32 v1, 0x7f800000
; GFX7-NEXT: v_fract_f32_e32 v0, gv@abs32@lo
; GFX7-NEXT: v_cmp_neq_f32_e64 vcc, |s4|, v1
; GFX7-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: safe_math_fract_f32_swapped_edge_case_x_is_const:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: s_mov_b32 s4, gv@abs32@lo
; GFX8-NEXT: v_mov_b32_e32 v1, 0x7f800000
; GFX8-NEXT: v_fract_f32_e32 v0, gv@abs32@lo
; GFX8-NEXT: v_cmp_neq_f32_e64 vcc, |s4|, v1
; GFX8-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: safe_math_fract_f32_swapped_edge_case_x_is_const:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s0, 0x7f800000
; GFX11-NEXT: v_fract_f32_e32 v0, gv@abs32@lo
; GFX11-NEXT: v_cmp_neq_f32_e64 vcc_lo, |gv@abs32@lo|, s0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX11-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc_lo
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: safe_math_fract_f32_swapped_edge_case_x_is_const:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: s_brev_b32 s0, -2
; GFX12-NEXT: v_fract_f32_e32 v0, gv@abs32@lo
; GFX12-NEXT: s_wait_alu depctr_sa_sdst(0)
; GFX12-NEXT: s_and_b32 s0, gv@abs32@lo, s0
; GFX12-NEXT: s_wait_alu depctr_sa_sdst(0)
; GFX12-NEXT: s_cmp_neq_f32 s0, 0x7f800000
; GFX12-NEXT: s_cselect_b32 vcc_lo, -1, 0
; GFX12-NEXT: s_wait_alu depctr_sa_sdst(0)
; GFX12-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc_lo
; GFX12-NEXT: s_setpc_b64 s[30:31]
; GFX6-IR-LABEL: define float @safe_math_fract_f32_swapped_edge_case_x_is_const(
; GFX6-IR-SAME: ) #[[ATTR1]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = call float @llvm.floor.f32(float bitcast (i32 ptrtoint (ptr @gv to i32) to float))
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub float bitcast (i32 ptrtoint (ptr @gv to i32) to float), [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; GFX6-IR-NEXT: [[X_FABS:%.*]] = call float @llvm.fabs.f32(float bitcast (i32 ptrtoint (ptr @gv to i32) to float))
; GFX6-IR-NEXT: [[NOT_INF:%.*]] = fcmp une float [[X_FABS]], 0x7FF0000000000000
; GFX6-IR-NEXT: [[COND:%.*]] = select i1 [[NOT_INF]], float [[MIN]], float 0.000000e+00
; GFX6-IR-NEXT: [[NOT_NAN:%.*]] = fcmp ord float bitcast (i32 ptrtoint (ptr @gv to i32) to float), 0.000000e+00
; GFX6-IR-NEXT: [[COND8:%.*]] = select i1 [[NOT_NAN]], float [[COND]], float bitcast (i32 ptrtoint (ptr @gv to i32) to float)
; GFX6-IR-NEXT: ret float [[COND8]]
; IR-FRACT-LABEL: define float @safe_math_fract_f32_swapped_edge_case_x_is_const(
; IR-FRACT-SAME: ) #[[ATTR1]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[COND:%.*]] = call float @llvm.amdgcn.fract.f32(float bitcast (i32 ptrtoint (ptr @gv to i32) to float))
; IR-FRACT-NEXT: [[X_FABS:%.*]] = call float @llvm.fabs.f32(float bitcast (i32 ptrtoint (ptr @gv to i32) to float))
; IR-FRACT-NEXT: [[NOT_INF:%.*]] = fcmp une float [[X_FABS]], 0x7FF0000000000000
; IR-FRACT-NEXT: [[COND8:%.*]] = select i1 [[NOT_INF]], float [[COND]], float 0.000000e+00
; IR-FRACT-NEXT: ret float [[COND8]]
entry:
%floor = call float @llvm.floor.f32(float bitcast (i32 ptrtoint (ptr @gv to i32) to float))
%sub = fsub float bitcast (i32 ptrtoint (ptr @gv to i32) to float), %floor
%min = call float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
%x.fabs = call float @llvm.fabs.f32(float bitcast (i32 ptrtoint (ptr @gv to i32) to float))
%not.inf = fcmp une float %x.fabs, 0x7FF0000000000000
%cond = select i1 %not.inf, float %min, float 0.000000e+00
%not.nan = fcmp ord float bitcast (i32 ptrtoint (ptr @gv to i32) to float), 0.000000e+00
%cond8 = select i1 %not.nan, float %cond, float bitcast (i32 ptrtoint (ptr @gv to i32) to float)
ret float %cond8
}
define float @safe_math_fract_f32_swapped_edge_case_split_block(float %x, i1 %cond) #0 {
; GFX6-LABEL: safe_math_fract_f32_swapped_edge_case_split_block:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_and_b32_e32 v1, 1, v1
; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, 1, v1
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v1, v0, v1
; GFX6-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v1
; GFX6-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX6-NEXT: ; %bb.1: ; %edge_cases
; GFX6-NEXT: s_mov_b32 s6, 0x7f800000
; GFX6-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s6
; GFX6-NEXT: v_cndmask_b32_e32 v1, 0, v1, vcc
; GFX6-NEXT: v_cmp_o_f32_e32 vcc, v0, v0
; GFX6-NEXT: v_cndmask_b32_e32 v1, v0, v1, vcc
; GFX6-NEXT: ; %bb.2: ; %UnifiedReturnBlock
; GFX6-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX6-NEXT: v_mov_b32_e32 v0, v1
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: safe_math_fract_f32_swapped_edge_case_split_block:
; GFX7: ; %bb.0:
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_and_b32_e32 v1, 1, v1
; GFX7-NEXT: v_cmp_ne_u32_e32 vcc, 1, v1
; GFX7-NEXT: ; implicit-def: $vgpr1
; GFX7-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX7-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GFX7-NEXT: ; %bb.1: ; %ret
; GFX7-NEXT: v_floor_f32_e32 v1, v0
; GFX7-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX7-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v0
; GFX7-NEXT: ; implicit-def: $vgpr0
; GFX7-NEXT: ; %bb.2: ; %Flow
; GFX7-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; GFX7-NEXT: ; %bb.3: ; %edge_cases
; GFX7-NEXT: s_mov_b32 s6, 0x7f800000
; GFX7-NEXT: v_fract_f32_e32 v1, v0
; GFX7-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s6
; GFX7-NEXT: v_cndmask_b32_e32 v1, 0, v1, vcc
; GFX7-NEXT: ; %bb.4: ; %UnifiedReturnBlock
; GFX7-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX7-NEXT: v_mov_b32_e32 v0, v1
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: safe_math_fract_f32_swapped_edge_case_split_block:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_and_b32_e32 v1, 1, v1
; GFX8-NEXT: v_cmp_ne_u32_e32 vcc, 1, v1
; GFX8-NEXT: ; implicit-def: $vgpr1
; GFX8-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX8-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GFX8-NEXT: ; %bb.1: ; %ret
; GFX8-NEXT: v_floor_f32_e32 v1, v0
; GFX8-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX8-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v0
; GFX8-NEXT: ; implicit-def: $vgpr0
; GFX8-NEXT: ; %bb.2: ; %Flow
; GFX8-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; GFX8-NEXT: ; %bb.3: ; %edge_cases
; GFX8-NEXT: s_mov_b32 s6, 0x7f800000
; GFX8-NEXT: v_fract_f32_e32 v1, v0
; GFX8-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s6
; GFX8-NEXT: v_cndmask_b32_e32 v1, 0, v1, vcc
; GFX8-NEXT: ; %bb.4: ; %UnifiedReturnBlock
; GFX8-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX8-NEXT: v_mov_b32_e32 v0, v1
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: safe_math_fract_f32_swapped_edge_case_split_block:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_mov_b32_e32 v2, v0
; GFX11-NEXT: v_and_b32_e32 v0, 1, v1
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
; GFX11-NEXT: v_cmp_ne_u32_e32 vcc_lo, 1, v0
; GFX11-NEXT: ; implicit-def: $vgpr0
; GFX11-NEXT: s_and_saveexec_b32 s0, vcc_lo
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
; GFX11-NEXT: s_cbranch_execnz .LBB58_3
; GFX11-NEXT: ; %bb.1: ; %Flow
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX11-NEXT: s_cbranch_execnz .LBB58_4
; GFX11-NEXT: .LBB58_2: ; %UnifiedReturnBlock
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
; GFX11-NEXT: .LBB58_3: ; %ret
; GFX11-NEXT: v_floor_f32_e32 v0, v2
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_sub_f32_e32 v0, v2, v0
; GFX11-NEXT: ; implicit-def: $vgpr2
; GFX11-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX11-NEXT: s_cbranch_execz .LBB58_2
; GFX11-NEXT: .LBB58_4: ; %edge_cases
; GFX11-NEXT: v_fract_f32_e32 v0, v2
; GFX11-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v2|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX11-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc_lo
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: safe_math_fract_f32_swapped_edge_case_split_block:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_mov_b32_e32 v2, v0
; GFX12-NEXT: v_and_b32_e32 v0, 1, v1
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT: v_cmp_ne_u32_e32 vcc_lo, 1, v0
; GFX12-NEXT: ; implicit-def: $vgpr0
; GFX12-NEXT: s_and_saveexec_b32 s0, vcc_lo
; GFX12-NEXT: s_wait_alu depctr_sa_sdst(0)
; GFX12-NEXT: s_xor_b32 s0, exec_lo, s0
; GFX12-NEXT: s_cbranch_execnz .LBB58_3
; GFX12-NEXT: ; %bb.1: ; %Flow
; GFX12-NEXT: s_wait_alu depctr_sa_sdst(0)
; GFX12-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX12-NEXT: s_cbranch_execnz .LBB58_4
; GFX12-NEXT: .LBB58_2: ; %UnifiedReturnBlock
; GFX12-NEXT: s_wait_alu depctr_sa_sdst(0)
; GFX12-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX12-NEXT: s_setpc_b64 s[30:31]
; GFX12-NEXT: .LBB58_3: ; %ret
; GFX12-NEXT: v_floor_f32_e32 v0, v2
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-NEXT: v_sub_f32_e32 v0, v2, v0
; GFX12-NEXT: ; implicit-def: $vgpr2
; GFX12-NEXT: v_min_num_f32_e32 v0, 0x3f7fffff, v0
; GFX12-NEXT: s_wait_alu depctr_sa_sdst(0)
; GFX12-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX12-NEXT: s_cbranch_execz .LBB58_2
; GFX12-NEXT: .LBB58_4: ; %edge_cases
; GFX12-NEXT: v_fract_f32_e32 v0, v2
; GFX12-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v2|
; GFX12-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX12-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc_lo
; GFX12-NEXT: s_wait_alu depctr_sa_sdst(0)
; GFX12-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX12-NEXT: s_setpc_b64 s[30:31]
; GFX6-IR-LABEL: define float @safe_math_fract_f32_swapped_edge_case_split_block(
; GFX6-IR-SAME: float [[X:%.*]], i1 [[COND:%.*]]) #[[ATTR1]] {
; GFX6-IR-NEXT: [[FLOOR:%.*]] = call float @llvm.floor.f32(float [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; GFX6-IR-NEXT: [[X_FABS:%.*]] = call float @llvm.fabs.f32(float [[X]])
; GFX6-IR-NEXT: br i1 [[COND]], label %[[EDGE_CASES:.*]], label %[[RET:.*]]
; GFX6-IR: [[EDGE_CASES]]:
; GFX6-IR-NEXT: [[NOT_INF:%.*]] = fcmp une float [[X_FABS]], 0x7FF0000000000000
; GFX6-IR-NEXT: [[CLAMP_INF:%.*]] = select i1 [[NOT_INF]], float [[MIN]], float 0.000000e+00
; GFX6-IR-NEXT: [[NOT_NAN:%.*]] = fcmp ord float [[X]], 0.000000e+00
; GFX6-IR-NEXT: [[COND8:%.*]] = select i1 [[NOT_NAN]], float [[CLAMP_INF]], float [[X]]
; GFX6-IR-NEXT: ret float [[COND8]]
; GFX6-IR: [[RET]]:
; GFX6-IR-NEXT: ret float [[MIN]]
; IR-FRACT-LABEL: define float @safe_math_fract_f32_swapped_edge_case_split_block(
; IR-FRACT-SAME: float [[X:%.*]], i1 [[COND:%.*]]) #[[ATTR1]] {
; IR-FRACT-NEXT: [[FLOOR:%.*]] = call float @llvm.floor.f32(float [[X]])
; IR-FRACT-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; IR-FRACT-NEXT: [[MIN:%.*]] = call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; IR-FRACT-NEXT: [[CLAMP_INF:%.*]] = call float @llvm.amdgcn.fract.f32(float [[X]])
; IR-FRACT-NEXT: [[X_FABS:%.*]] = call float @llvm.fabs.f32(float [[X]])
; IR-FRACT-NEXT: br i1 [[COND]], label %[[EDGE_CASES:.*]], label %[[RET:.*]]
; IR-FRACT: [[EDGE_CASES]]:
; IR-FRACT-NEXT: [[NOT_INF:%.*]] = fcmp une float [[X_FABS]], 0x7FF0000000000000
; IR-FRACT-NEXT: [[COND8:%.*]] = select i1 [[NOT_INF]], float [[CLAMP_INF]], float 0.000000e+00
; IR-FRACT-NEXT: ret float [[COND8]]
; IR-FRACT: [[RET]]:
; IR-FRACT-NEXT: ret float [[MIN]]
%floor = call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = call float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
%x.fabs = call float @llvm.fabs.f32(float %x)
br i1 %cond, label %edge_cases, label %ret
edge_cases:
%not.inf = fcmp une float %x.fabs, 0x7FF0000000000000
%clamp.inf = select i1 %not.inf, float %min, float 0.0
%not.nan = fcmp ord float %x, 0.0
%cond8 = select i1 %not.nan, float %clamp.inf, float %x
ret float %cond8
ret:
ret float %min
}
define <3 x float> @safe_math_fract_f32_swapped_edge_case_vector(<3 x float> %x) #0 {
; GFX6-LABEL: safe_math_fract_f32_swapped_edge_case_vector:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v4, v2
; GFX6-NEXT: v_floor_f32_e32 v3, v0
; GFX6-NEXT: v_sub_f32_e32 v4, v2, v4
; GFX6-NEXT: v_mov_b32_e32 v5, 0x1fb
; GFX6-NEXT: v_sub_f32_e32 v3, v0, v3
; GFX6-NEXT: v_min_f32_e32 v4, 0x3f7fffff, v4
; GFX6-NEXT: v_cmp_class_f32_e32 vcc, v2, v5
; GFX6-NEXT: v_min_f32_e32 v3, 0x3f7fffff, v3
; GFX6-NEXT: v_cndmask_b32_e32 v4, 0, v4, vcc
; GFX6-NEXT: v_cmp_class_f32_e32 vcc, v0, v5
; GFX6-NEXT: v_cndmask_b32_e32 v3, 0, v3, vcc
; GFX6-NEXT: v_cmp_o_f32_e32 vcc, v0, v0
; GFX6-NEXT: v_cndmask_b32_e32 v0, v0, v3, vcc
; GFX6-NEXT: v_cmp_o_f32_e32 vcc, v2, v2
; GFX6-NEXT: v_cndmask_b32_e32 v2, v2, v4, vcc
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: safe_math_fract_f32_swapped_edge_case_vector:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_mov_b32_e32 v5, 0x1fb
; GFX7-NEXT: v_fract_f32_e32 v3, v0
; GFX7-NEXT: v_cmp_class_f32_e32 vcc, v0, v5
; GFX7-NEXT: v_fract_f32_e32 v4, v2
; GFX7-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc
; GFX7-NEXT: v_cmp_class_f32_e32 vcc, v2, v5
; GFX7-NEXT: v_fract_f32_e32 v1, v1
; GFX7-NEXT: v_cndmask_b32_e32 v2, 0, v4, vcc
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: safe_math_fract_f32_swapped_edge_case_vector:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_mov_b32_e32 v5, 0x1fb
; GFX8-NEXT: v_fract_f32_e32 v3, v0
; GFX8-NEXT: v_cmp_class_f32_e32 vcc, v0, v5
; GFX8-NEXT: v_fract_f32_e32 v4, v2
; GFX8-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc
; GFX8-NEXT: v_cmp_class_f32_e32 vcc, v2, v5
; GFX8-NEXT: v_fract_f32_e32 v1, v1
; GFX8-NEXT: v_cndmask_b32_e32 v2, 0, v4, vcc
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: safe_math_fract_f32_swapped_edge_case_vector:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v3, v0
; GFX11-NEXT: v_cmp_class_f32_e64 vcc_lo, v0, 0x1fb
; GFX11-NEXT: v_fract_f32_e32 v4, v2
; GFX11-NEXT: v_fract_f32_e32 v1, v1
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_4)
; GFX11-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc_lo
; GFX11-NEXT: v_cmp_class_f32_e64 vcc_lo, v2, 0x1fb
; GFX11-NEXT: v_cndmask_b32_e32 v2, 0, v4, vcc_lo
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: safe_math_fract_f32_swapped_edge_case_vector:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v3, v0
; GFX12-NEXT: v_cmp_class_f32_e64 vcc_lo, v0, 0x1fb
; GFX12-NEXT: v_fract_f32_e32 v4, v2
; GFX12-NEXT: v_fract_f32_e32 v1, v1
; GFX12-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc_lo
; GFX12-NEXT: v_cmp_class_f32_e64 vcc_lo, v2, 0x1fb
; GFX12-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-NEXT: v_cndmask_b32_e32 v2, 0, v4, vcc_lo
; GFX12-NEXT: s_setpc_b64 s[30:31]
; GFX6-IR-LABEL: define <3 x float> @safe_math_fract_f32_swapped_edge_case_vector(
; GFX6-IR-SAME: <3 x float> [[X:%.*]]) #[[ATTR1]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[FLOOR:%.*]] = call <3 x float> @llvm.floor.v3f32(<3 x float> [[X]])
; GFX6-IR-NEXT: [[SUB:%.*]] = fsub <3 x float> [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = call <3 x float> @llvm.minnum.v3f32(<3 x float> [[SUB]], <3 x float> <float 0x3FEFFFFFE0000000, float poison, float 0x3FEFFFFFE0000000>)
; GFX6-IR-NEXT: [[X_FABS:%.*]] = call <3 x float> @llvm.fabs.v3f32(<3 x float> [[X]])
; GFX6-IR-NEXT: [[NOT_INF:%.*]] = fcmp une <3 x float> [[X_FABS]], <float 0x7FF0000000000000, float poison, float 0x7FF0000000000000>
; GFX6-IR-NEXT: [[COND:%.*]] = select <3 x i1> [[NOT_INF]], <3 x float> [[MIN]], <3 x float> <float 0.000000e+00, float poison, float 0.000000e+00>
; GFX6-IR-NEXT: [[NOT_NAN:%.*]] = fcmp ord <3 x float> [[X]], <float 0.000000e+00, float poison, float 0.000000e+00>
; GFX6-IR-NEXT: [[COND8:%.*]] = select <3 x i1> [[NOT_NAN]], <3 x float> [[COND]], <3 x float> [[X]]
; GFX6-IR-NEXT: ret <3 x float> [[COND8]]
; IR-FRACT-LABEL: define <3 x float> @safe_math_fract_f32_swapped_edge_case_vector(
; IR-FRACT-SAME: <3 x float> [[X:%.*]]) #[[ATTR1]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[TMP0:%.*]] = extractelement <3 x float> [[X]], i64 0
; IR-FRACT-NEXT: [[TMP1:%.*]] = extractelement <3 x float> [[X]], i64 1
; IR-FRACT-NEXT: [[TMP2:%.*]] = extractelement <3 x float> [[X]], i64 2
; IR-FRACT-NEXT: [[TMP3:%.*]] = call float @llvm.amdgcn.fract.f32(float [[TMP0]])
; IR-FRACT-NEXT: [[TMP4:%.*]] = call float @llvm.amdgcn.fract.f32(float [[TMP1]])
; IR-FRACT-NEXT: [[TMP5:%.*]] = call float @llvm.amdgcn.fract.f32(float [[TMP2]])
; IR-FRACT-NEXT: [[TMP6:%.*]] = insertelement <3 x float> poison, float [[TMP3]], i64 0
; IR-FRACT-NEXT: [[TMP7:%.*]] = insertelement <3 x float> [[TMP6]], float [[TMP4]], i64 1
; IR-FRACT-NEXT: [[COND:%.*]] = insertelement <3 x float> [[TMP7]], float [[TMP5]], i64 2
; IR-FRACT-NEXT: [[X_FABS:%.*]] = call <3 x float> @llvm.fabs.v3f32(<3 x float> [[X]])
; IR-FRACT-NEXT: [[NOT_INF:%.*]] = fcmp une <3 x float> [[X_FABS]], <float 0x7FF0000000000000, float poison, float 0x7FF0000000000000>
; IR-FRACT-NEXT: [[COND8:%.*]] = select <3 x i1> [[NOT_INF]], <3 x float> [[COND]], <3 x float> <float 0.000000e+00, float poison, float 0.000000e+00>
; IR-FRACT-NEXT: ret <3 x float> [[COND8]]
entry:
%floor = call <3 x float> @llvm.floor.v3f32(<3 x float> %x)
%sub = fsub <3 x float> %x, %floor
%min = call <3 x float> @llvm.minnum.v3f32(<3 x float> %sub, <3 x float> <float 0x3FEFFFFFE0000000, float poison, float 0x3FEFFFFFE0000000>)
%x.fabs = call <3 x float> @llvm.fabs.v3f32(<3 x float> %x)
%not.inf = fcmp une <3 x float> %x.fabs, <float 0x7FF0000000000000, float poison, float 0x7FF0000000000000>
%cond = select <3 x i1> %not.inf, <3 x float> %min, <3 x float> <float 0.0, float poison, float 0.0>
%not.nan = fcmp ord <3 x float> %x, <float 0.0, float poison, float 0.0>
%cond8 = select <3 x i1> %not.nan, <3 x float> %cond, <3 x float> %x
ret <3 x float> %cond8
}
define float @safe_math_fract_f32_swapped_edge_case_inf_check_wrong_compare(float %x) #0 {
; GFX6-LABEL: safe_math_fract_f32_swapped_edge_case_inf_check_wrong_compare:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v1, v0, v1
; GFX6-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v1
; GFX6-NEXT: v_cmp_o_f32_e32 vcc, v0, v0
; GFX6-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: safe_math_fract_f32_swapped_edge_case_inf_check_wrong_compare:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_fract_f32_e32 v0, v0
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: safe_math_fract_f32_swapped_edge_case_inf_check_wrong_compare:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_fract_f32_e32 v0, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: safe_math_fract_f32_swapped_edge_case_inf_check_wrong_compare:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v0, v0
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: safe_math_fract_f32_swapped_edge_case_inf_check_wrong_compare:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v0, v0
; GFX12-NEXT: s_setpc_b64 s[30:31]
; IR-LABEL: define float @safe_math_fract_f32_swapped_edge_case_inf_check_wrong_compare(
; IR-SAME: float [[X:%.*]]) #[[ATTR1]] {
; IR-NEXT: [[ENTRY:.*:]]
; IR-NEXT: [[FLOOR:%.*]] = call float @llvm.floor.f32(float [[X]])
; IR-NEXT: [[SUB:%.*]] = fsub float [[X]], [[FLOOR]]
; IR-NEXT: [[MIN:%.*]] = call float @llvm.minnum.f32(float [[SUB]], float 0x3FEFFFFFE0000000)
; IR-NEXT: [[X_FABS:%.*]] = call float @llvm.fabs.f32(float [[X]])
; IR-NEXT: [[NOT_NOT_INF:%.*]] = fcmp ule float [[X_FABS]], 0x7FF0000000000000
; IR-NEXT: [[COND:%.*]] = select i1 [[NOT_NOT_INF]], float [[MIN]], float 0.000000e+00
; IR-NEXT: [[NOT_NAN:%.*]] = fcmp ord float [[X]], 0.000000e+00
; IR-NEXT: [[COND8:%.*]] = select i1 [[NOT_NAN]], float [[COND]], float [[X]]
; IR-NEXT: ret float [[COND8]]
entry:
%floor = call float @llvm.floor.f32(float %x)
%sub = fsub float %x, %floor
%min = call float @llvm.minnum.f32(float %sub, float 0x3FEFFFFFE0000000)
%x.fabs = call float @llvm.fabs.f32(float %x)
%not.not.inf = fcmp ule float %x.fabs, 0x7FF0000000000000
%cond = select i1 %not.not.inf, float %min, float 0.0
%not.nan = fcmp ord float %x, 0.0
%cond8 = select i1 %not.nan, float %cond, float %x
ret float %cond8
}
; Use compare and select instead of min intrinsic.
define float @fract_pat_fcmp_oge_select(float %x, ptr addrspace(1) %iptr) #0 {
; GFX6-LABEL: fract_pat_fcmp_oge_select:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: s_mov_b32 s6, 0
; GFX6-NEXT: s_mov_b32 s7, 0xf000
; GFX6-NEXT: s_mov_b32 s4, s6
; GFX6-NEXT: s_mov_b32 s5, s6
; GFX6-NEXT: v_floor_f32_e32 v3, v0
; GFX6-NEXT: buffer_store_dword v3, v[1:2], s[4:7], 0 addr64
; GFX6-NEXT: v_sub_f32_e32 v1, v0, v3
; GFX6-NEXT: s_mov_b32 s4, 0x7f800000
; GFX6-NEXT: v_min_legacy_f32_e32 v1, 0x3f7fffff, v1
; GFX6-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX6-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: fract_pat_fcmp_oge_select:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: s_mov_b32 s6, 0
; GFX7-NEXT: s_mov_b32 s7, 0xf000
; GFX7-NEXT: s_mov_b32 s4, s6
; GFX7-NEXT: s_mov_b32 s5, s6
; GFX7-NEXT: v_floor_f32_e32 v3, v0
; GFX7-NEXT: buffer_store_dword v3, v[1:2], s[4:7], 0 addr64
; GFX7-NEXT: s_mov_b32 s4, 0x7f800000
; GFX7-NEXT: v_fract_f32_e32 v1, v0
; GFX7-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX7-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: fract_pat_fcmp_oge_select:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_floor_f32_e32 v3, v0
; GFX8-NEXT: s_mov_b32 s4, 0x7f800000
; GFX8-NEXT: global_store_dword v[1:2], v3, off
; GFX8-NEXT: v_fract_f32_e32 v1, v0
; GFX8-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX8-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: fract_pat_fcmp_oge_select:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v3, v0
; GFX11-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX11-NEXT: v_floor_f32_e32 v4, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_3)
; GFX11-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc_lo
; GFX11-NEXT: global_store_b32 v[1:2], v4, off
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: fract_pat_fcmp_oge_select:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v3, v0
; GFX12-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX12-NEXT: v_floor_f32_e32 v4, v0
; GFX12-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_3)
; GFX12-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc_lo
; GFX12-NEXT: global_store_b32 v[1:2], v4, off
; GFX12-NEXT: s_setpc_b64 s[30:31]
; GFX6-IR-LABEL: define float @fract_pat_fcmp_oge_select(
; GFX6-IR-SAME: float [[X:%.*]], ptr addrspace(1) [[IPTR:%.*]]) #[[ATTR1]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[CALL:%.*]] = call float @llvm.floor.f32(float [[X]])
; GFX6-IR-NEXT: store float [[CALL]], ptr addrspace(1) [[IPTR]], align 4
; GFX6-IR-NEXT: [[SUB1:%.*]] = fsub float [[X]], [[CALL]]
; GFX6-IR-NEXT: [[CMP:%.*]] = fcmp oge float [[SUB1]], 0x3FEFFFFFE0000000
; GFX6-IR-NEXT: [[COND:%.*]] = select i1 [[CMP]], float 0x3FEFFFFFE0000000, float [[SUB1]]
; GFX6-IR-NEXT: [[FABS_X:%.*]] = call float @llvm.fabs.f32(float [[X]])
; GFX6-IR-NEXT: [[NOT_INF:%.*]] = fcmp une float [[FABS_X]], 0x7FF0000000000000
; GFX6-IR-NEXT: [[COND6:%.*]] = select i1 [[NOT_INF]], float [[COND]], float 0.000000e+00
; GFX6-IR-NEXT: ret float [[COND6]]
; IR-FRACT-LABEL: define float @fract_pat_fcmp_oge_select(
; IR-FRACT-SAME: float [[X:%.*]], ptr addrspace(1) [[IPTR:%.*]]) #[[ATTR1]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[CALL:%.*]] = call float @llvm.floor.f32(float [[X]])
; IR-FRACT-NEXT: store float [[CALL]], ptr addrspace(1) [[IPTR]], align 4
; IR-FRACT-NEXT: [[COND:%.*]] = call float @llvm.amdgcn.fract.f32(float [[X]])
; IR-FRACT-NEXT: [[FABS_X:%.*]] = call float @llvm.fabs.f32(float [[X]])
; IR-FRACT-NEXT: [[NOT_INF:%.*]] = fcmp une float [[FABS_X]], 0x7FF0000000000000
; IR-FRACT-NEXT: [[COND6:%.*]] = select i1 [[NOT_INF]], float [[COND]], float 0.000000e+00
; IR-FRACT-NEXT: ret float [[COND6]]
entry:
%call = call float @llvm.floor.f32(float %x)
store float %call, ptr addrspace(1) %iptr, align 4
%sub1 = fsub float %x, %call
%cmp = fcmp oge float %sub1, 0x3FEFFFFFE0000000
%cond = select i1 %cmp, float 0x3FEFFFFFE0000000, float %sub1
%fabs.x = call float @llvm.fabs.f32(float %x)
%not.inf = fcmp une float %fabs.x, 0x7FF0000000000000
%cond6 = select i1 %not.inf, float %cond, float 0.000000e+00
ret float %cond6
}
define float @fract_pat_fcmp_ogt_select(float %x, ptr addrspace(1) %iptr) #0 {
; GFX6-LABEL: fract_pat_fcmp_ogt_select:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: s_mov_b32 s6, 0
; GFX6-NEXT: s_mov_b32 s7, 0xf000
; GFX6-NEXT: s_mov_b32 s4, s6
; GFX6-NEXT: s_mov_b32 s5, s6
; GFX6-NEXT: v_floor_f32_e32 v3, v0
; GFX6-NEXT: buffer_store_dword v3, v[1:2], s[4:7], 0 addr64
; GFX6-NEXT: v_sub_f32_e32 v1, v0, v3
; GFX6-NEXT: s_mov_b32 s4, 0x7f800000
; GFX6-NEXT: v_min_legacy_f32_e32 v1, 0x3f7fffff, v1
; GFX6-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX6-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: fract_pat_fcmp_ogt_select:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: s_mov_b32 s6, 0
; GFX7-NEXT: s_mov_b32 s7, 0xf000
; GFX7-NEXT: s_mov_b32 s4, s6
; GFX7-NEXT: s_mov_b32 s5, s6
; GFX7-NEXT: v_floor_f32_e32 v3, v0
; GFX7-NEXT: buffer_store_dword v3, v[1:2], s[4:7], 0 addr64
; GFX7-NEXT: s_mov_b32 s4, 0x7f800000
; GFX7-NEXT: v_fract_f32_e32 v1, v0
; GFX7-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX7-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: fract_pat_fcmp_ogt_select:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_floor_f32_e32 v3, v0
; GFX8-NEXT: s_mov_b32 s4, 0x7f800000
; GFX8-NEXT: global_store_dword v[1:2], v3, off
; GFX8-NEXT: v_fract_f32_e32 v1, v0
; GFX8-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX8-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: fract_pat_fcmp_ogt_select:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v3, v0
; GFX11-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX11-NEXT: v_floor_f32_e32 v4, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_3)
; GFX11-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc_lo
; GFX11-NEXT: global_store_b32 v[1:2], v4, off
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: fract_pat_fcmp_ogt_select:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v3, v0
; GFX12-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX12-NEXT: v_floor_f32_e32 v4, v0
; GFX12-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_3)
; GFX12-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc_lo
; GFX12-NEXT: global_store_b32 v[1:2], v4, off
; GFX12-NEXT: s_setpc_b64 s[30:31]
; GFX6-IR-LABEL: define float @fract_pat_fcmp_ogt_select(
; GFX6-IR-SAME: float [[X:%.*]], ptr addrspace(1) [[IPTR:%.*]]) #[[ATTR1]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[CALL:%.*]] = call float @llvm.floor.f32(float [[X]])
; GFX6-IR-NEXT: store float [[CALL]], ptr addrspace(1) [[IPTR]], align 4
; GFX6-IR-NEXT: [[SUB1:%.*]] = fsub float [[X]], [[CALL]]
; GFX6-IR-NEXT: [[CMP:%.*]] = fcmp ogt float [[SUB1]], 0x3FEFFFFFE0000000
; GFX6-IR-NEXT: [[COND:%.*]] = select i1 [[CMP]], float 0x3FEFFFFFE0000000, float [[SUB1]]
; GFX6-IR-NEXT: [[FABS_X:%.*]] = call float @llvm.fabs.f32(float [[X]])
; GFX6-IR-NEXT: [[NOT_INF:%.*]] = fcmp une float [[FABS_X]], 0x7FF0000000000000
; GFX6-IR-NEXT: [[COND6:%.*]] = select i1 [[NOT_INF]], float [[COND]], float 0.000000e+00
; GFX6-IR-NEXT: ret float [[COND6]]
; IR-FRACT-LABEL: define float @fract_pat_fcmp_ogt_select(
; IR-FRACT-SAME: float [[X:%.*]], ptr addrspace(1) [[IPTR:%.*]]) #[[ATTR1]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[CALL:%.*]] = call float @llvm.floor.f32(float [[X]])
; IR-FRACT-NEXT: store float [[CALL]], ptr addrspace(1) [[IPTR]], align 4
; IR-FRACT-NEXT: [[COND:%.*]] = call float @llvm.amdgcn.fract.f32(float [[X]])
; IR-FRACT-NEXT: [[FABS_X:%.*]] = call float @llvm.fabs.f32(float [[X]])
; IR-FRACT-NEXT: [[NOT_INF:%.*]] = fcmp une float [[FABS_X]], 0x7FF0000000000000
; IR-FRACT-NEXT: [[COND6:%.*]] = select i1 [[NOT_INF]], float [[COND]], float 0.000000e+00
; IR-FRACT-NEXT: ret float [[COND6]]
entry:
%call = call float @llvm.floor.f32(float %x)
store float %call, ptr addrspace(1) %iptr, align 4
%sub1 = fsub float %x, %call
%cmp = fcmp ogt float %sub1, 0x3FEFFFFFE0000000
%cond = select i1 %cmp, float 0x3FEFFFFFE0000000, float %sub1
%fabs.x = call float @llvm.fabs.f32(float %x)
%not.inf = fcmp une float %fabs.x, 0x7FF0000000000000
%cond6 = select i1 %not.inf, float %cond, float 0.000000e+00
ret float %cond6
}
; Wrong nan handling
define float @negative_fract_pat_fcmp_olt(float %x, ptr addrspace(1) %iptr) #0 {
; GFX6-LABEL: negative_fract_pat_fcmp_olt:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: s_mov_b32 s6, 0
; GFX6-NEXT: s_mov_b32 s7, 0xf000
; GFX6-NEXT: s_mov_b32 s4, s6
; GFX6-NEXT: s_mov_b32 s5, s6
; GFX6-NEXT: v_floor_f32_e32 v3, v0
; GFX6-NEXT: buffer_store_dword v3, v[1:2], s[4:7], 0 addr64
; GFX6-NEXT: v_sub_f32_e32 v1, v0, v3
; GFX6-NEXT: s_mov_b32 s4, 0x7f800000
; GFX6-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v1
; GFX6-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX6-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: negative_fract_pat_fcmp_olt:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: s_mov_b32 s6, 0
; GFX7-NEXT: s_mov_b32 s7, 0xf000
; GFX7-NEXT: s_mov_b32 s4, s6
; GFX7-NEXT: s_mov_b32 s5, s6
; GFX7-NEXT: v_floor_f32_e32 v3, v0
; GFX7-NEXT: buffer_store_dword v3, v[1:2], s[4:7], 0 addr64
; GFX7-NEXT: v_sub_f32_e32 v1, v0, v3
; GFX7-NEXT: s_mov_b32 s4, 0x7f800000
; GFX7-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v1
; GFX7-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX7-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: negative_fract_pat_fcmp_olt:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_floor_f32_e32 v3, v0
; GFX8-NEXT: global_store_dword v[1:2], v3, off
; GFX8-NEXT: v_sub_f32_e32 v1, v0, v3
; GFX8-NEXT: s_mov_b32 s4, 0x7f800000
; GFX8-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v1
; GFX8-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX8-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: negative_fract_pat_fcmp_olt:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_floor_f32_e32 v3, v0
; GFX11-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_sub_f32_e32 v4, v0, v3
; GFX11-NEXT: global_store_b32 v[1:2], v3, off
; GFX11-NEXT: v_min_f32_e32 v4, 0x3f7fffff, v4
; GFX11-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc_lo
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: negative_fract_pat_fcmp_olt:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_floor_f32_e32 v3, v0
; GFX12-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_3) | instid1(VALU_DEP_1)
; GFX12-NEXT: v_sub_f32_e32 v4, v0, v3
; GFX12-NEXT: global_store_b32 v[1:2], v3, off
; GFX12-NEXT: v_min_num_f32_e32 v4, 0x3f7fffff, v4
; GFX12-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc_lo
; GFX12-NEXT: s_setpc_b64 s[30:31]
; IR-LABEL: define float @negative_fract_pat_fcmp_olt(
; IR-SAME: float [[X:%.*]], ptr addrspace(1) [[IPTR:%.*]]) #[[ATTR1]] {
; IR-NEXT: [[ENTRY:.*:]]
; IR-NEXT: [[CALL:%.*]] = call float @llvm.floor.f32(float noundef [[X]])
; IR-NEXT: store float [[CALL]], ptr addrspace(1) [[IPTR]], align 4
; IR-NEXT: [[SUB1:%.*]] = fsub float [[X]], [[CALL]]
; IR-NEXT: [[CMP:%.*]] = fcmp olt float [[SUB1]], 0x3FEFFFFFE0000000
; IR-NEXT: [[COND:%.*]] = select i1 [[CMP]], float [[SUB1]], float 0x3FEFFFFFE0000000
; IR-NEXT: [[FABS_X:%.*]] = call float @llvm.fabs.f32(float [[X]])
; IR-NEXT: [[NOT_INF:%.*]] = fcmp une float [[FABS_X]], 0x7FF0000000000000
; IR-NEXT: [[COND6:%.*]] = select i1 [[NOT_INF]], float [[COND]], float 0.000000e+00
; IR-NEXT: ret float [[COND6]]
entry:
%call = call float @llvm.floor.f32(float noundef %x)
store float %call, ptr addrspace(1) %iptr, align 4
%sub1 = fsub float %x, %call
%cmp = fcmp olt float %sub1, 0x3FEFFFFFE0000000
%cond = select i1 %cmp, float %sub1, float 0x3FEFFFFFE0000000
%fabs.x = call float @llvm.fabs.f32(float %x)
%not.inf = fcmp une float %fabs.x, 0x7FF0000000000000
%cond6 = select i1 %not.inf, float %cond, float 0.000000e+00
ret float %cond6
}
define float @fract_pat_fcmp_olt_not_nan_src(float nofpclass(nan) %x, ptr addrspace(1) %iptr) #0 {
; GFX6-LABEL: fract_pat_fcmp_olt_not_nan_src:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: s_mov_b32 s6, 0
; GFX6-NEXT: s_mov_b32 s7, 0xf000
; GFX6-NEXT: s_mov_b32 s4, s6
; GFX6-NEXT: s_mov_b32 s5, s6
; GFX6-NEXT: v_floor_f32_e32 v3, v0
; GFX6-NEXT: buffer_store_dword v3, v[1:2], s[4:7], 0 addr64
; GFX6-NEXT: v_sub_f32_e32 v1, v0, v3
; GFX6-NEXT: s_mov_b32 s4, 0x7f800000
; GFX6-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v1
; GFX6-NEXT: v_cmp_lg_f32_e64 vcc, |v0|, s4
; GFX6-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: fract_pat_fcmp_olt_not_nan_src:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: s_mov_b32 s6, 0
; GFX7-NEXT: s_mov_b32 s7, 0xf000
; GFX7-NEXT: s_mov_b32 s4, s6
; GFX7-NEXT: s_mov_b32 s5, s6
; GFX7-NEXT: v_floor_f32_e32 v3, v0
; GFX7-NEXT: buffer_store_dword v3, v[1:2], s[4:7], 0 addr64
; GFX7-NEXT: v_sub_f32_e32 v1, v0, v3
; GFX7-NEXT: s_mov_b32 s4, 0x7f800000
; GFX7-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v1
; GFX7-NEXT: v_cmp_lg_f32_e64 vcc, |v0|, s4
; GFX7-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: fract_pat_fcmp_olt_not_nan_src:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_floor_f32_e32 v3, v0
; GFX8-NEXT: global_store_dword v[1:2], v3, off
; GFX8-NEXT: v_sub_f32_e32 v1, v0, v3
; GFX8-NEXT: s_mov_b32 s4, 0x7f800000
; GFX8-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v1
; GFX8-NEXT: v_cmp_lg_f32_e64 vcc, |v0|, s4
; GFX8-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: fract_pat_fcmp_olt_not_nan_src:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_floor_f32_e32 v3, v0
; GFX11-NEXT: v_cmp_lg_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_sub_f32_e32 v4, v0, v3
; GFX11-NEXT: global_store_b32 v[1:2], v3, off
; GFX11-NEXT: v_min_f32_e32 v4, 0x3f7fffff, v4
; GFX11-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc_lo
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: fract_pat_fcmp_olt_not_nan_src:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_floor_f32_e32 v3, v0
; GFX12-NEXT: v_cmp_lg_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_3) | instid1(VALU_DEP_1)
; GFX12-NEXT: v_sub_f32_e32 v4, v0, v3
; GFX12-NEXT: global_store_b32 v[1:2], v3, off
; GFX12-NEXT: v_min_num_f32_e32 v4, 0x3f7fffff, v4
; GFX12-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc_lo
; GFX12-NEXT: s_setpc_b64 s[30:31]
; IR-LABEL: define float @fract_pat_fcmp_olt_not_nan_src(
; IR-SAME: float nofpclass(nan) [[X:%.*]], ptr addrspace(1) [[IPTR:%.*]]) #[[ATTR1]] {
; IR-NEXT: [[ENTRY:.*:]]
; IR-NEXT: [[CALL:%.*]] = call float @llvm.floor.f32(float noundef [[X]])
; IR-NEXT: store float [[CALL]], ptr addrspace(1) [[IPTR]], align 4
; IR-NEXT: [[SUB1:%.*]] = fsub float [[X]], [[CALL]]
; IR-NEXT: [[CMP:%.*]] = fcmp olt float [[SUB1]], 0x3FEFFFFFE0000000
; IR-NEXT: [[COND:%.*]] = select i1 [[CMP]], float [[SUB1]], float 0x3FEFFFFFE0000000
; IR-NEXT: [[FABS_X:%.*]] = call float @llvm.fabs.f32(float [[X]])
; IR-NEXT: [[NOT_INF:%.*]] = fcmp une float [[FABS_X]], 0x7FF0000000000000
; IR-NEXT: [[COND6:%.*]] = select i1 [[NOT_INF]], float [[COND]], float 0.000000e+00
; IR-NEXT: ret float [[COND6]]
entry:
%call = call float @llvm.floor.f32(float noundef %x)
store float %call, ptr addrspace(1) %iptr, align 4
%sub1 = fsub float %x, %call
%cmp = fcmp olt float %sub1, 0x3FEFFFFFE0000000
%cond = select i1 %cmp, float %sub1, float 0x3FEFFFFFE0000000
%fabs.x = call float @llvm.fabs.f32(float %x)
%not.inf = fcmp une float %fabs.x, 0x7FF0000000000000
%cond6 = select i1 %not.inf, float %cond, float 0.000000e+00
ret float %cond6
}
define float @fract_pat_minimum(float %x, ptr addrspace(1) %iptr) #0 {
; GFX6-LABEL: fract_pat_minimum:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: s_mov_b32 s6, 0
; GFX6-NEXT: s_mov_b32 s7, 0xf000
; GFX6-NEXT: s_mov_b32 s4, s6
; GFX6-NEXT: s_mov_b32 s5, s6
; GFX6-NEXT: v_floor_f32_e32 v3, v0
; GFX6-NEXT: buffer_store_dword v3, v[1:2], s[4:7], 0 addr64
; GFX6-NEXT: v_sub_f32_e32 v1, v0, v3
; GFX6-NEXT: v_min_f32_e32 v2, 0x3f7fffff, v1
; GFX6-NEXT: s_waitcnt expcnt(0)
; GFX6-NEXT: v_mov_b32_e32 v3, 0x7fc00000
; GFX6-NEXT: v_cmp_o_f32_e32 vcc, v1, v1
; GFX6-NEXT: s_mov_b32 s4, 0x7f800000
; GFX6-NEXT: v_cndmask_b32_e32 v1, v3, v2, vcc
; GFX6-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX6-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc
; GFX6-NEXT: s_waitcnt vmcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: fract_pat_minimum:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: s_mov_b32 s6, 0
; GFX7-NEXT: s_mov_b32 s7, 0xf000
; GFX7-NEXT: s_mov_b32 s4, s6
; GFX7-NEXT: s_mov_b32 s5, s6
; GFX7-NEXT: v_floor_f32_e32 v3, v0
; GFX7-NEXT: buffer_store_dword v3, v[1:2], s[4:7], 0 addr64
; GFX7-NEXT: s_mov_b32 s4, 0x7f800000
; GFX7-NEXT: v_fract_f32_e32 v1, v0
; GFX7-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX7-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: fract_pat_minimum:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_floor_f32_e32 v3, v0
; GFX8-NEXT: s_mov_b32 s4, 0x7f800000
; GFX8-NEXT: global_store_dword v[1:2], v3, off
; GFX8-NEXT: v_fract_f32_e32 v1, v0
; GFX8-NEXT: v_cmp_neq_f32_e64 vcc, |v0|, s4
; GFX8-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: fract_pat_minimum:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v3, v0
; GFX11-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX11-NEXT: v_floor_f32_e32 v4, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_3)
; GFX11-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc_lo
; GFX11-NEXT: global_store_b32 v[1:2], v4, off
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: fract_pat_minimum:
; GFX12: ; %bb.0: ; %entry
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v3, v0
; GFX12-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0|
; GFX12-NEXT: v_floor_f32_e32 v4, v0
; GFX12-NEXT: s_wait_alu depctr_va_vcc(0)
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_3)
; GFX12-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc_lo
; GFX12-NEXT: global_store_b32 v[1:2], v4, off
; GFX12-NEXT: s_setpc_b64 s[30:31]
; GFX6-IR-LABEL: define float @fract_pat_minimum(
; GFX6-IR-SAME: float [[X:%.*]], ptr addrspace(1) [[IPTR:%.*]]) #[[ATTR1]] {
; GFX6-IR-NEXT: [[ENTRY:.*:]]
; GFX6-IR-NEXT: [[CALL:%.*]] = call float @llvm.floor.f32(float [[X]])
; GFX6-IR-NEXT: store float [[CALL]], ptr addrspace(1) [[IPTR]], align 4
; GFX6-IR-NEXT: [[SUB1:%.*]] = fsub float [[X]], [[CALL]]
; GFX6-IR-NEXT: [[MIN:%.*]] = call float @llvm.minimum.f32(float [[SUB1]], float 0x3FEFFFFFE0000000)
; GFX6-IR-NEXT: [[FABS_X:%.*]] = call float @llvm.fabs.f32(float [[X]])
; GFX6-IR-NEXT: [[NOT_INF:%.*]] = fcmp une float [[FABS_X]], 0x7FF0000000000000
; GFX6-IR-NEXT: [[COND6:%.*]] = select i1 [[NOT_INF]], float [[MIN]], float 0.000000e+00
; GFX6-IR-NEXT: ret float [[COND6]]
; IR-FRACT-LABEL: define float @fract_pat_minimum(
; IR-FRACT-SAME: float [[X:%.*]], ptr addrspace(1) [[IPTR:%.*]]) #[[ATTR1]] {
; IR-FRACT-NEXT: [[ENTRY:.*:]]
; IR-FRACT-NEXT: [[CALL:%.*]] = call float @llvm.floor.f32(float [[X]])
; IR-FRACT-NEXT: store float [[CALL]], ptr addrspace(1) [[IPTR]], align 4
; IR-FRACT-NEXT: [[MIN:%.*]] = call nnan float @llvm.amdgcn.fract.f32(float [[X]])
; IR-FRACT-NEXT: [[FABS_X:%.*]] = call float @llvm.fabs.f32(float [[X]])
; IR-FRACT-NEXT: [[NOT_INF:%.*]] = fcmp une float [[FABS_X]], 0x7FF0000000000000
; IR-FRACT-NEXT: [[COND6:%.*]] = select i1 [[NOT_INF]], float [[MIN]], float 0.000000e+00
; IR-FRACT-NEXT: ret float [[COND6]]
entry:
%call = call float @llvm.floor.f32(float %x)
store float %call, ptr addrspace(1) %iptr, align 4
%sub1 = fsub float %x, %call
%min = call float @llvm.minimum.f32(float %sub1, float 0x3FEFFFFFE0000000)
%fabs.x = call float @llvm.fabs.f32(float %x)
%not.inf = fcmp une float %fabs.x, 0x7FF0000000000000
%cond6 = select i1 %not.inf, float %min, float 0.0
ret float %cond6
}
define float @core_fract_pat_fcmp_oge_select(float %x) #0 {
; GFX6-LABEL: core_fract_pat_fcmp_oge_select:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX6-NEXT: v_min_legacy_f32_e32 v0, 0x3f7fffff, v0
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: core_fract_pat_fcmp_oge_select:
; GFX7: ; %bb.0:
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_fract_f32_e32 v0, v0
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: core_fract_pat_fcmp_oge_select:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_fract_f32_e32 v0, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: core_fract_pat_fcmp_oge_select:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v0, v0
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: core_fract_pat_fcmp_oge_select:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v0, v0
; GFX12-NEXT: s_setpc_b64 s[30:31]
; GFX6-IR-LABEL: define float @core_fract_pat_fcmp_oge_select(
; GFX6-IR-SAME: float [[X:%.*]]) #[[ATTR1]] {
; GFX6-IR-NEXT: [[FLOOR:%.*]] = call float @llvm.floor.f32(float [[X]])
; GFX6-IR-NEXT: [[SUB_FLOOR:%.*]] = fsub float [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[OGE_MIN_CONST:%.*]] = fcmp oge float [[SUB_FLOOR]], 0x3FEFFFFFE0000000
; GFX6-IR-NEXT: [[SELECT:%.*]] = select i1 [[OGE_MIN_CONST]], float 0x3FEFFFFFE0000000, float [[SUB_FLOOR]]
; GFX6-IR-NEXT: ret float [[SELECT]]
; IR-FRACT-LABEL: define float @core_fract_pat_fcmp_oge_select(
; IR-FRACT-SAME: float [[X:%.*]]) #[[ATTR1]] {
; IR-FRACT-NEXT: [[SELECT:%.*]] = call float @llvm.amdgcn.fract.f32(float [[X]])
; IR-FRACT-NEXT: ret float [[SELECT]]
%floor = call float @llvm.floor.f32(float %x)
%sub.floor = fsub float %x, %floor
%oge.min.const = fcmp oge float %sub.floor, 0x3FEFFFFFE0000000
%select = select i1 %oge.min.const, float 0x3FEFFFFFE0000000, float %sub.floor
ret float %select
}
define float @core_fract_pat_minimum(float %x) #0 {
; GFX6-LABEL: core_fract_pat_minimum:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX6-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v0
; GFX6-NEXT: v_mov_b32_e32 v2, 0x7fc00000
; GFX6-NEXT: v_cmp_o_f32_e32 vcc, v0, v0
; GFX6-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: core_fract_pat_minimum:
; GFX7: ; %bb.0:
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_fract_f32_e32 v0, v0
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: core_fract_pat_minimum:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_fract_f32_e32 v0, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: core_fract_pat_minimum:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v0, v0
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: core_fract_pat_minimum:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v0, v0
; GFX12-NEXT: s_setpc_b64 s[30:31]
; GFX6-IR-LABEL: define float @core_fract_pat_minimum(
; GFX6-IR-SAME: float [[X:%.*]]) #[[ATTR1]] {
; GFX6-IR-NEXT: [[FLOOR:%.*]] = call float @llvm.floor.f32(float [[X]])
; GFX6-IR-NEXT: [[SUB_FLOOR:%.*]] = fsub float [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = call float @llvm.minimum.f32(float [[SUB_FLOOR]], float 0x3FEFFFFFE0000000)
; GFX6-IR-NEXT: ret float [[MIN]]
; IR-FRACT-LABEL: define float @core_fract_pat_minimum(
; IR-FRACT-SAME: float [[X:%.*]]) #[[ATTR1]] {
; IR-FRACT-NEXT: [[MIN:%.*]] = call nnan float @llvm.amdgcn.fract.f32(float [[X]])
; IR-FRACT-NEXT: ret float [[MIN]]
%floor = call float @llvm.floor.f32(float %x)
%sub.floor = fsub float %x, %floor
%min = call float @llvm.minimum.f32(float %sub.floor, float 0x3FEFFFFFE0000000)
ret float %min
}
define <2 x float> @core_fract_pat_fcmp_oge_select_v2f32(<2 x float> %x) #0 {
; GFX6-LABEL: core_fract_pat_fcmp_oge_select_v2f32:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v2, v0
; GFX6-NEXT: v_floor_f32_e32 v3, v1
; GFX6-NEXT: v_sub_f32_e32 v0, v0, v2
; GFX6-NEXT: v_sub_f32_e32 v1, v1, v3
; GFX6-NEXT: v_min_legacy_f32_e32 v0, 0x3f7fffff, v0
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: core_fract_pat_fcmp_oge_select_v2f32:
; GFX7: ; %bb.0:
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_fract_f32_e32 v0, v0
; GFX7-NEXT: v_fract_f32_e32 v1, v1
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: core_fract_pat_fcmp_oge_select_v2f32:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_fract_f32_e32 v0, v0
; GFX8-NEXT: v_fract_f32_e32 v1, v1
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: core_fract_pat_fcmp_oge_select_v2f32:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v0, v0
; GFX11-NEXT: v_fract_f32_e32 v1, v1
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: core_fract_pat_fcmp_oge_select_v2f32:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v0, v0
; GFX12-NEXT: v_fract_f32_e32 v1, v1
; GFX12-NEXT: s_setpc_b64 s[30:31]
; GFX6-IR-LABEL: define <2 x float> @core_fract_pat_fcmp_oge_select_v2f32(
; GFX6-IR-SAME: <2 x float> [[X:%.*]]) #[[ATTR1]] {
; GFX6-IR-NEXT: [[FLOOR:%.*]] = call <2 x float> @llvm.floor.v2f32(<2 x float> [[X]])
; GFX6-IR-NEXT: [[SUB_FLOOR:%.*]] = fsub <2 x float> [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[OGE_MIN_CONST:%.*]] = fcmp oge <2 x float> [[SUB_FLOOR]], <float 0x3FEFFFFFE0000000, float poison>
; GFX6-IR-NEXT: [[SELECT:%.*]] = select <2 x i1> [[OGE_MIN_CONST]], <2 x float> <float 0x3FEFFFFFE0000000, float poison>, <2 x float> [[SUB_FLOOR]]
; GFX6-IR-NEXT: ret <2 x float> [[SELECT]]
; IR-FRACT-LABEL: define <2 x float> @core_fract_pat_fcmp_oge_select_v2f32(
; IR-FRACT-SAME: <2 x float> [[X:%.*]]) #[[ATTR1]] {
; IR-FRACT-NEXT: [[TMP1:%.*]] = extractelement <2 x float> [[X]], i64 0
; IR-FRACT-NEXT: [[TMP2:%.*]] = extractelement <2 x float> [[X]], i64 1
; IR-FRACT-NEXT: [[TMP3:%.*]] = call float @llvm.amdgcn.fract.f32(float [[TMP1]])
; IR-FRACT-NEXT: [[TMP4:%.*]] = call float @llvm.amdgcn.fract.f32(float [[TMP2]])
; IR-FRACT-NEXT: [[TMP5:%.*]] = insertelement <2 x float> poison, float [[TMP3]], i64 0
; IR-FRACT-NEXT: [[SELECT:%.*]] = insertelement <2 x float> [[TMP5]], float [[TMP4]], i64 1
; IR-FRACT-NEXT: ret <2 x float> [[SELECT]]
%floor = call <2 x float> @llvm.floor.v2f32(<2 x float> %x)
%sub.floor = fsub <2 x float> %x, %floor
%oge.min.const = fcmp oge <2 x float> %sub.floor, <float 0x3FEFFFFFE0000000, float poison>
%select = select <2 x i1> %oge.min.const, <2 x float> <float 0x3FEFFFFFE0000000, float poison>, <2 x float> %sub.floor
ret <2 x float> %select
}
define <2 x float> @core_fract_pat_minimum_v2f32(<2 x float> %x) #0 {
; GFX6-LABEL: core_fract_pat_minimum_v2f32:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_floor_f32_e32 v1, v0
; GFX6-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX6-NEXT: v_min_f32_e32 v1, 0x3f7fffff, v0
; GFX6-NEXT: v_mov_b32_e32 v2, 0x7fc00000
; GFX6-NEXT: v_cmp_o_f32_e32 vcc, v0, v0
; GFX6-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc
; GFX6-NEXT: v_mov_b32_e32 v1, 0x7fc00000
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: core_fract_pat_minimum_v2f32:
; GFX7: ; %bb.0:
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: v_fract_f32_e32 v0, v0
; GFX7-NEXT: v_fract_f32_e32 v1, v1
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: core_fract_pat_minimum_v2f32:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_fract_f32_e32 v0, v0
; GFX8-NEXT: v_fract_f32_e32 v1, v1
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: core_fract_pat_minimum_v2f32:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_fract_f32_e32 v0, v0
; GFX11-NEXT: v_fract_f32_e32 v1, v1
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: core_fract_pat_minimum_v2f32:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_fract_f32_e32 v0, v0
; GFX12-NEXT: v_fract_f32_e32 v1, v1
; GFX12-NEXT: s_setpc_b64 s[30:31]
; GFX6-IR-LABEL: define <2 x float> @core_fract_pat_minimum_v2f32(
; GFX6-IR-SAME: <2 x float> [[X:%.*]]) #[[ATTR1]] {
; GFX6-IR-NEXT: [[FLOOR:%.*]] = call <2 x float> @llvm.floor.v2f32(<2 x float> [[X]])
; GFX6-IR-NEXT: [[SUB_FLOOR:%.*]] = fsub <2 x float> [[X]], [[FLOOR]]
; GFX6-IR-NEXT: [[MIN:%.*]] = call <2 x float> @llvm.minimum.v2f32(<2 x float> [[SUB_FLOOR]], <2 x float> <float 0x3FEFFFFFE0000000, float poison>)
; GFX6-IR-NEXT: ret <2 x float> [[MIN]]
; IR-FRACT-LABEL: define <2 x float> @core_fract_pat_minimum_v2f32(
; IR-FRACT-SAME: <2 x float> [[X:%.*]]) #[[ATTR1]] {
; IR-FRACT-NEXT: [[TMP1:%.*]] = extractelement <2 x float> [[X]], i64 0
; IR-FRACT-NEXT: [[TMP2:%.*]] = extractelement <2 x float> [[X]], i64 1
; IR-FRACT-NEXT: [[TMP3:%.*]] = call nnan float @llvm.amdgcn.fract.f32(float [[TMP1]])
; IR-FRACT-NEXT: [[TMP4:%.*]] = call nnan float @llvm.amdgcn.fract.f32(float [[TMP2]])
; IR-FRACT-NEXT: [[TMP5:%.*]] = insertelement <2 x float> poison, float [[TMP3]], i64 0
; IR-FRACT-NEXT: [[MIN:%.*]] = insertelement <2 x float> [[TMP5]], float [[TMP4]], i64 1
; IR-FRACT-NEXT: ret <2 x float> [[MIN]]
%floor = call <2 x float> @llvm.floor.v2f32(<2 x float> %x)
%sub.floor = fsub <2 x float> %x, %floor
%min = call <2 x float> @llvm.minimum.v2f32(<2 x float> %sub.floor, <2 x float> <float 0x3FEFFFFFE0000000, float poison>)
ret <2 x float> %min
}
declare half @llvm.floor.f16(half) #0
declare float @llvm.floor.f32(float) #0
declare double @llvm.floor.f64(double) #0
declare <2 x double> @llvm.floor.v2f64(<2 x double>) #0
declare <2 x float> @llvm.floor.v2f32(<2 x float>) #0
declare <2 x half> @llvm.floor.v2f16(<2 x half>) #0
declare float @llvm.trunc.f32(float) #0
declare float @llvm.minnum.f32(float, float) #0
declare half @llvm.minnum.f16(half, half) #0
declare double @llvm.minnum.f64(double, double) #0
declare <2 x float> @llvm.minnum.v2f32(<2 x float>, <2 x float>) #0
declare <2 x half> @llvm.minnum.v2f16(<2 x half>, <2 x half>) #0
declare <2 x double> @llvm.minnum.v2f64(<2 x double>, <2 x double>) #0
declare float @llvm.maxnum.f32(float, float) #0
declare float @llvm.fabs.f32(float) #0
declare double @llvm.fabs.f64(double) #0
declare <2 x float> @llvm.fabs.v2f32(<2 x float>) #0
declare half @llvm.fabs.f16(half) #0
declare <2 x half> @llvm.fabs.v2f16(<2 x half>) #0
declare <2 x double> @llvm.fabs.v2f64(<2 x double>) #0
attributes #0 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }