6dbc01e801
This is a NFC patch. This patch run a bulk update on CodeGen tests that are impacted by the true16 features. This patch applies: 1. duplicate GFX11plus runlines and apply them with "+mattr=+real-true16" and "+mattr=-real-true16" 2. update the test with the update script For some GISEL runlines, the current CodeGen do not fully support the true16 version. Still update the runlines, but comment out the failing one, and added a "FIXME-TRUE16" comment to that test for easier tracking. These test will be fixed in the following patches. This is in a transition state that we support both "+real-true16/-real-true16" in our code base. We plan to move to "+real-true16" as default, and finally remove "-real-true16" mode and test lines.
4496 lines
202 KiB
LLVM
4496 lines
202 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
|
; RUN: llc -amdgpu-scalarize-global-loads=false -enable-misched=0 -mtriple=amdgcn -mattr=+mad-mac-f32-insts -verify-machineinstrs < %s | FileCheck --check-prefix=SI %s
|
|
; RUN: llc -amdgpu-scalarize-global-loads=false -enable-misched=0 -mtriple=amdgcn -mcpu=bonaire -verify-machineinstrs < %s | FileCheck --check-prefix=CI %s
|
|
; RUN: llc -amdgpu-scalarize-global-loads=false -enable-misched=0 -mtriple=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck --check-prefix=VI %s
|
|
; RUN: llc -amdgpu-scalarize-global-loads=false -enable-misched=0 -mtriple=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck --check-prefix=GFX9 %s
|
|
; RUN: llc -amdgpu-scalarize-global-loads=false -enable-misched=0 -mtriple=amdgcn -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck --check-prefix=GFX10 %s
|
|
; RUN: llc -amdgpu-scalarize-global-loads=false -enable-misched=0 -mtriple=amdgcn -mcpu=gfx1100 -mattr=+real-true16 -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX11,GFX11-TRUE16 %s
|
|
; RUN: llc -amdgpu-scalarize-global-loads=false -enable-misched=0 -mtriple=amdgcn -mcpu=gfx1100 -mattr=-real-true16 -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX11,GFX11-FAKE16 %s
|
|
; RUN: llc -amdgpu-scalarize-global-loads=false -enable-misched=0 -mtriple=amdgcn -mcpu=gfx1150 -verify-machineinstrs < %s | FileCheck --check-prefix=GFX1150 %s
|
|
|
|
define amdgpu_kernel void @frem_f16(ptr addrspace(1) %out, ptr addrspace(1) %in1,
|
|
; SI-LABEL: frem_f16:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9
|
|
; SI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0xd
|
|
; SI-NEXT: s_mov_b32 s11, 0xf000
|
|
; SI-NEXT: s_mov_b32 s10, -1
|
|
; SI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SI-NEXT: s_mov_b32 s8, s0
|
|
; SI-NEXT: s_mov_b32 s9, s1
|
|
; SI-NEXT: s_mov_b32 s0, s2
|
|
; SI-NEXT: s_mov_b32 s1, s3
|
|
; SI-NEXT: s_mov_b32 s2, s10
|
|
; SI-NEXT: s_mov_b32 s3, s11
|
|
; SI-NEXT: s_mov_b32 s6, s10
|
|
; SI-NEXT: s_mov_b32 s7, s11
|
|
; SI-NEXT: buffer_load_ushort v0, off, s[0:3], 0
|
|
; SI-NEXT: s_waitcnt vmcnt(0)
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; SI-NEXT: buffer_load_ushort v1, off, s[4:7], 0 offset:8
|
|
; SI-NEXT: s_waitcnt vmcnt(0)
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v1, v1
|
|
; SI-NEXT: v_div_scale_f32 v2, vcc, v0, v1, v0
|
|
; SI-NEXT: v_div_scale_f32 v3, s[0:1], v1, v1, v0
|
|
; SI-NEXT: v_rcp_f32_e32 v4, v3
|
|
; SI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; SI-NEXT: v_fma_f32 v5, -v3, v4, 1.0
|
|
; SI-NEXT: v_fma_f32 v4, v5, v4, v4
|
|
; SI-NEXT: v_mul_f32_e32 v5, v2, v4
|
|
; SI-NEXT: v_fma_f32 v6, -v3, v5, v2
|
|
; SI-NEXT: v_fma_f32 v5, v6, v4, v5
|
|
; SI-NEXT: v_fma_f32 v2, -v3, v5, v2
|
|
; SI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; SI-NEXT: v_div_fmas_f32 v2, v2, v4, v5
|
|
; SI-NEXT: v_div_fixup_f32 v2, v2, v1, v0
|
|
; SI-NEXT: v_trunc_f32_e32 v2, v2
|
|
; SI-NEXT: v_fma_f32 v0, -v2, v1, v0
|
|
; SI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 0
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; SI-NEXT: buffer_store_short v0, off, s[8:11], 0
|
|
; SI-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: frem_f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9
|
|
; CI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0xd
|
|
; CI-NEXT: s_mov_b32 s11, 0xf000
|
|
; CI-NEXT: s_mov_b32 s10, -1
|
|
; CI-NEXT: s_mov_b32 s6, s10
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_mov_b32 s8, s0
|
|
; CI-NEXT: s_mov_b32 s9, s1
|
|
; CI-NEXT: s_mov_b32 s0, s2
|
|
; CI-NEXT: s_mov_b32 s1, s3
|
|
; CI-NEXT: s_mov_b32 s2, s10
|
|
; CI-NEXT: s_mov_b32 s3, s11
|
|
; CI-NEXT: s_mov_b32 s7, s11
|
|
; CI-NEXT: buffer_load_ushort v0, off, s[0:3], 0
|
|
; CI-NEXT: buffer_load_ushort v1, off, s[4:7], 0 offset:8
|
|
; CI-NEXT: s_waitcnt vmcnt(1)
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; CI-NEXT: s_waitcnt vmcnt(0)
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
|
|
; CI-NEXT: v_div_scale_f32 v3, s[0:1], v1, v1, v0
|
|
; CI-NEXT: v_div_scale_f32 v2, vcc, v0, v1, v0
|
|
; CI-NEXT: v_rcp_f32_e32 v4, v3
|
|
; CI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; CI-NEXT: v_fma_f32 v5, -v3, v4, 1.0
|
|
; CI-NEXT: v_fma_f32 v4, v5, v4, v4
|
|
; CI-NEXT: v_mul_f32_e32 v5, v2, v4
|
|
; CI-NEXT: v_fma_f32 v6, -v3, v5, v2
|
|
; CI-NEXT: v_fma_f32 v5, v6, v4, v5
|
|
; CI-NEXT: v_fma_f32 v2, -v3, v5, v2
|
|
; CI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; CI-NEXT: v_div_fmas_f32 v2, v2, v4, v5
|
|
; CI-NEXT: v_div_fixup_f32 v2, v2, v1, v0
|
|
; CI-NEXT: v_trunc_f32_e32 v2, v2
|
|
; CI-NEXT: v_fma_f32 v0, -v2, v1, v0
|
|
; CI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: buffer_store_short v0, off, s[8:11], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; VI-LABEL: frem_f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; VI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x34
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: s_add_u32 s0, s4, 8
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: v_mov_b32_e32 v2, s2
|
|
; VI-NEXT: v_mov_b32_e32 v3, s3
|
|
; VI-NEXT: s_addc_u32 s1, s5, 0
|
|
; VI-NEXT: flat_load_ushort v4, v[2:3]
|
|
; VI-NEXT: v_mov_b32_e32 v3, s1
|
|
; VI-NEXT: v_mov_b32_e32 v2, s0
|
|
; VI-NEXT: flat_load_ushort v2, v[2:3]
|
|
; VI-NEXT: s_waitcnt vmcnt(1)
|
|
; VI-NEXT: v_cvt_f32_f16_e32 v3, v4
|
|
; VI-NEXT: s_waitcnt vmcnt(0)
|
|
; VI-NEXT: v_cvt_f32_f16_e32 v5, v2
|
|
; VI-NEXT: v_rcp_f32_e32 v6, v5
|
|
; VI-NEXT: v_mul_f32_e32 v7, v3, v6
|
|
; VI-NEXT: v_mad_f32 v8, -v5, v7, v3
|
|
; VI-NEXT: v_mac_f32_e32 v7, v8, v6
|
|
; VI-NEXT: v_mad_f32 v3, -v5, v7, v3
|
|
; VI-NEXT: v_mul_f32_e32 v3, v3, v6
|
|
; VI-NEXT: v_and_b32_e32 v3, 0xff800000, v3
|
|
; VI-NEXT: v_add_f32_e32 v3, v3, v7
|
|
; VI-NEXT: v_cvt_f16_f32_e32 v3, v3
|
|
; VI-NEXT: v_div_fixup_f16 v3, v3, v2, v4
|
|
; VI-NEXT: v_trunc_f16_e32 v3, v3
|
|
; VI-NEXT: v_fma_f16 v2, -v3, v2, v4
|
|
; VI-NEXT: flat_store_short v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: frem_f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; GFX9-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_load_ushort v1, v0, s[2:3]
|
|
; GFX9-NEXT: global_load_ushort v2, v0, s[6:7] offset:8
|
|
; GFX9-NEXT: s_waitcnt vmcnt(1)
|
|
; GFX9-NEXT: v_cvt_f32_f16_e32 v3, v1
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX9-NEXT: v_cvt_f32_f16_e32 v4, v2
|
|
; GFX9-NEXT: v_rcp_f32_e32 v4, v4
|
|
; GFX9-NEXT: v_mul_f32_e32 v3, v3, v4
|
|
; GFX9-NEXT: v_mad_mix_f32 v5, -v2, v3, v1 op_sel_hi:[1,0,1]
|
|
; GFX9-NEXT: v_mac_f32_e32 v3, v5, v4
|
|
; GFX9-NEXT: v_mad_mix_f32 v5, -v2, v3, v1 op_sel_hi:[1,0,1]
|
|
; GFX9-NEXT: v_mul_f32_e32 v4, v5, v4
|
|
; GFX9-NEXT: v_and_b32_e32 v4, 0xff800000, v4
|
|
; GFX9-NEXT: v_add_f32_e32 v3, v4, v3
|
|
; GFX9-NEXT: v_cvt_f16_f32_e32 v3, v3
|
|
; GFX9-NEXT: v_div_fixup_f16 v3, v3, v2, v1
|
|
; GFX9-NEXT: v_trunc_f16_e32 v3, v3
|
|
; GFX9-NEXT: v_fma_f16 v1, -v3, v2, v1
|
|
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; GFX10-LABEL: frem_f16:
|
|
; GFX10: ; %bb.0:
|
|
; GFX10-NEXT: s_clause 0x1
|
|
; GFX10-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; GFX10-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
|
|
; GFX10-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX10-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-NEXT: s_clause 0x1
|
|
; GFX10-NEXT: global_load_ushort v1, v0, s[2:3]
|
|
; GFX10-NEXT: global_load_ushort v2, v0, s[6:7] offset:8
|
|
; GFX10-NEXT: s_waitcnt vmcnt(1)
|
|
; GFX10-NEXT: v_cvt_f32_f16_e32 v3, v1
|
|
; GFX10-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX10-NEXT: v_cvt_f32_f16_e32 v4, v2
|
|
; GFX10-NEXT: v_rcp_f32_e32 v5, v4
|
|
; GFX10-NEXT: v_mul_f32_e32 v6, v3, v5
|
|
; GFX10-NEXT: v_mad_f32 v7, -v4, v6, v3
|
|
; GFX10-NEXT: v_mac_f32_e32 v6, v7, v5
|
|
; GFX10-NEXT: v_mad_f32 v3, -v4, v6, v3
|
|
; GFX10-NEXT: v_mul_f32_e32 v3, v3, v5
|
|
; GFX10-NEXT: v_and_b32_e32 v3, 0xff800000, v3
|
|
; GFX10-NEXT: v_add_f32_e32 v3, v3, v6
|
|
; GFX10-NEXT: v_cvt_f16_f32_e32 v3, v3
|
|
; GFX10-NEXT: v_div_fixup_f16 v3, v3, v2, v1
|
|
; GFX10-NEXT: v_trunc_f16_e32 v3, v3
|
|
; GFX10-NEXT: v_fma_f16 v1, -v3, v2, v1
|
|
; GFX10-NEXT: global_store_short v0, v1, s[0:1]
|
|
; GFX10-NEXT: s_endpgm
|
|
;
|
|
; GFX11-TRUE16-LABEL: frem_f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_clause 0x1
|
|
; GFX11-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX11-TRUE16-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX11-TRUE16-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: s_clause 0x1
|
|
; GFX11-TRUE16-NEXT: global_load_d16_b16 v0, v2, s[2:3]
|
|
; GFX11-TRUE16-NEXT: global_load_d16_b16 v1, v2, s[4:5] offset:8
|
|
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(1)
|
|
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v3, v0.l
|
|
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v4, v1.l
|
|
; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v1.l
|
|
; GFX11-TRUE16-NEXT: v_mov_b16_e32 v6.l, v0.l
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_2) | instid1(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_rcp_f32_e32 v4, v4
|
|
; GFX11-TRUE16-NEXT: s_waitcnt_depctr 0xfff
|
|
; GFX11-TRUE16-NEXT: v_mul_f32_e32 v3, v3, v4
|
|
; GFX11-TRUE16-NEXT: v_fma_mix_f32 v7, -v5, v3, v6 op_sel_hi:[1,0,1]
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_fmac_f32_e32 v3, v7, v4
|
|
; GFX11-TRUE16-NEXT: v_fma_mix_f32 v5, -v5, v3, v6 op_sel_hi:[1,0,1]
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_mul_f32_e32 v4, v5, v4
|
|
; GFX11-TRUE16-NEXT: v_and_b32_e32 v4, 0xff800000, v4
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_add_f32_e32 v3, v4, v3
|
|
; GFX11-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.h, v3
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_div_fixup_f16 v0.h, v0.h, v1.l, v0.l
|
|
; GFX11-TRUE16-NEXT: v_trunc_f16_e32 v0.h, v0.h
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_fma_f16 v0.l, -v0.h, v1.l, v0.l
|
|
; GFX11-TRUE16-NEXT: global_store_b16 v2, v0, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_endpgm
|
|
;
|
|
; GFX11-FAKE16-LABEL: frem_f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_clause 0x1
|
|
; GFX11-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX11-FAKE16-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX11-FAKE16-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: s_clause 0x1
|
|
; GFX11-FAKE16-NEXT: global_load_u16 v1, v0, s[2:3]
|
|
; GFX11-FAKE16-NEXT: global_load_u16 v2, v0, s[4:5] offset:8
|
|
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(1)
|
|
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v3, v1
|
|
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v4, v2
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_rcp_f32_e32 v4, v4
|
|
; GFX11-FAKE16-NEXT: s_waitcnt_depctr 0xfff
|
|
; GFX11-FAKE16-NEXT: v_mul_f32_e32 v3, v3, v4
|
|
; GFX11-FAKE16-NEXT: v_fma_mix_f32 v5, -v2, v3, v1 op_sel_hi:[1,0,1]
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_fmac_f32_e32 v3, v5, v4
|
|
; GFX11-FAKE16-NEXT: v_fma_mix_f32 v5, -v2, v3, v1 op_sel_hi:[1,0,1]
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_mul_f32_e32 v4, v5, v4
|
|
; GFX11-FAKE16-NEXT: v_and_b32_e32 v4, 0xff800000, v4
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_add_f32_e32 v3, v4, v3
|
|
; GFX11-FAKE16-NEXT: v_cvt_f16_f32_e32 v3, v3
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_div_fixup_f16 v3, v3, v2, v1
|
|
; GFX11-FAKE16-NEXT: v_trunc_f16_e32 v3, v3
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_fma_f16 v1, -v3, v2, v1
|
|
; GFX11-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_endpgm
|
|
;
|
|
; GFX1150-LABEL: frem_f16:
|
|
; GFX1150: ; %bb.0:
|
|
; GFX1150-NEXT: s_clause 0x1
|
|
; GFX1150-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX1150-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX1150-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX1150-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX1150-NEXT: s_clause 0x1
|
|
; GFX1150-NEXT: global_load_u16 v1, v0, s[2:3]
|
|
; GFX1150-NEXT: global_load_u16 v2, v0, s[4:5] offset:8
|
|
; GFX1150-NEXT: s_waitcnt vmcnt(1)
|
|
; GFX1150-NEXT: v_cvt_f32_f16_e32 v3, v1
|
|
; GFX1150-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX1150-NEXT: v_cvt_f32_f16_e32 v4, v2
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(TRANS32_DEP_1)
|
|
; GFX1150-NEXT: v_rcp_f32_e32 v4, v4
|
|
; GFX1150-NEXT: v_mul_f32_e32 v3, v3, v4
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_mix_f32 v5, -v2, v3, v1 op_sel_hi:[1,0,1]
|
|
; GFX1150-NEXT: v_fmac_f32_e32 v3, v5, v4
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_mix_f32 v5, -v2, v3, v1 op_sel_hi:[1,0,1]
|
|
; GFX1150-NEXT: v_mul_f32_e32 v4, v5, v4
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_and_b32_e32 v4, 0xff800000, v4
|
|
; GFX1150-NEXT: v_add_f32_e32 v3, v4, v3
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_cvt_f16_f32_e32 v3, v3
|
|
; GFX1150-NEXT: v_div_fixup_f16 v3, v3, v2, v1
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_trunc_f16_e32 v3, v3
|
|
; GFX1150-NEXT: v_xor_b32_e32 v3, 0x8000, v3
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fmac_f16_e32 v1, v3, v2
|
|
; GFX1150-NEXT: global_store_b16 v0, v1, s[0:1]
|
|
; GFX1150-NEXT: s_endpgm
|
|
ptr addrspace(1) %in2) #0 {
|
|
%gep2 = getelementptr half, ptr addrspace(1) %in2, i32 4
|
|
%r0 = load half, ptr addrspace(1) %in1, align 4
|
|
%r1 = load half, ptr addrspace(1) %gep2, align 4
|
|
%r2 = frem half %r0, %r1
|
|
store half %r2, ptr addrspace(1) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @fast_frem_f16(ptr addrspace(1) %out, ptr addrspace(1) %in1,
|
|
; SI-LABEL: fast_frem_f16:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9
|
|
; SI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0xd
|
|
; SI-NEXT: s_mov_b32 s11, 0xf000
|
|
; SI-NEXT: s_mov_b32 s10, -1
|
|
; SI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SI-NEXT: s_mov_b32 s8, s0
|
|
; SI-NEXT: s_mov_b32 s9, s1
|
|
; SI-NEXT: s_mov_b32 s0, s2
|
|
; SI-NEXT: s_mov_b32 s1, s3
|
|
; SI-NEXT: s_mov_b32 s2, s10
|
|
; SI-NEXT: s_mov_b32 s3, s11
|
|
; SI-NEXT: s_mov_b32 s6, s10
|
|
; SI-NEXT: s_mov_b32 s7, s11
|
|
; SI-NEXT: buffer_load_ushort v0, off, s[0:3], 0
|
|
; SI-NEXT: s_waitcnt vmcnt(0)
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; SI-NEXT: buffer_load_ushort v1, off, s[4:7], 0 offset:8
|
|
; SI-NEXT: s_waitcnt vmcnt(0)
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v1, v1
|
|
; SI-NEXT: v_rcp_f32_e32 v2, v1
|
|
; SI-NEXT: v_mul_f32_e32 v2, v0, v2
|
|
; SI-NEXT: v_trunc_f32_e32 v2, v2
|
|
; SI-NEXT: v_fma_f32 v0, -v2, v1, v0
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; SI-NEXT: buffer_store_short v0, off, s[8:11], 0
|
|
; SI-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: fast_frem_f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9
|
|
; CI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0xd
|
|
; CI-NEXT: s_mov_b32 s11, 0xf000
|
|
; CI-NEXT: s_mov_b32 s10, -1
|
|
; CI-NEXT: s_mov_b32 s6, s10
|
|
; CI-NEXT: s_mov_b32 s7, s11
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_load_ushort v1, off, s[4:7], 0 offset:8
|
|
; CI-NEXT: s_mov_b32 s8, s0
|
|
; CI-NEXT: s_mov_b32 s9, s1
|
|
; CI-NEXT: s_mov_b32 s0, s2
|
|
; CI-NEXT: s_mov_b32 s1, s3
|
|
; CI-NEXT: s_mov_b32 s2, s10
|
|
; CI-NEXT: s_mov_b32 s3, s11
|
|
; CI-NEXT: buffer_load_ushort v0, off, s[0:3], 0
|
|
; CI-NEXT: s_waitcnt vmcnt(1)
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
|
|
; CI-NEXT: v_rcp_f32_e32 v2, v1
|
|
; CI-NEXT: s_waitcnt vmcnt(0)
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; CI-NEXT: v_mul_f32_e32 v2, v0, v2
|
|
; CI-NEXT: v_trunc_f32_e32 v2, v2
|
|
; CI-NEXT: v_fma_f32 v0, -v2, v1, v0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: buffer_store_short v0, off, s[8:11], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; VI-LABEL: fast_frem_f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; VI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x34
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: s_add_u32 s0, s4, 8
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: v_mov_b32_e32 v2, s2
|
|
; VI-NEXT: v_mov_b32_e32 v3, s3
|
|
; VI-NEXT: s_addc_u32 s1, s5, 0
|
|
; VI-NEXT: flat_load_ushort v4, v[2:3]
|
|
; VI-NEXT: v_mov_b32_e32 v3, s1
|
|
; VI-NEXT: v_mov_b32_e32 v2, s0
|
|
; VI-NEXT: flat_load_ushort v2, v[2:3]
|
|
; VI-NEXT: s_waitcnt vmcnt(0)
|
|
; VI-NEXT: v_rcp_f16_e32 v3, v2
|
|
; VI-NEXT: v_mul_f16_e32 v3, v4, v3
|
|
; VI-NEXT: v_trunc_f16_e32 v3, v3
|
|
; VI-NEXT: v_fma_f16 v2, -v3, v2, v4
|
|
; VI-NEXT: flat_store_short v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: fast_frem_f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; GFX9-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_load_ushort v1, v0, s[2:3]
|
|
; GFX9-NEXT: global_load_ushort v2, v0, s[6:7] offset:8
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX9-NEXT: v_rcp_f16_e32 v3, v2
|
|
; GFX9-NEXT: v_mul_f16_e32 v3, v1, v3
|
|
; GFX9-NEXT: v_trunc_f16_e32 v3, v3
|
|
; GFX9-NEXT: v_fma_f16 v1, -v3, v2, v1
|
|
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; GFX10-LABEL: fast_frem_f16:
|
|
; GFX10: ; %bb.0:
|
|
; GFX10-NEXT: s_clause 0x1
|
|
; GFX10-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; GFX10-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
|
|
; GFX10-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX10-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-NEXT: s_clause 0x1
|
|
; GFX10-NEXT: global_load_ushort v1, v0, s[2:3]
|
|
; GFX10-NEXT: global_load_ushort v2, v0, s[6:7] offset:8
|
|
; GFX10-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX10-NEXT: v_rcp_f16_e32 v3, v2
|
|
; GFX10-NEXT: v_mul_f16_e32 v3, v1, v3
|
|
; GFX10-NEXT: v_trunc_f16_e32 v3, v3
|
|
; GFX10-NEXT: v_fma_f16 v1, -v3, v2, v1
|
|
; GFX10-NEXT: global_store_short v0, v1, s[0:1]
|
|
; GFX10-NEXT: s_endpgm
|
|
;
|
|
; GFX11-TRUE16-LABEL: fast_frem_f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_clause 0x1
|
|
; GFX11-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX11-TRUE16-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX11-TRUE16-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: s_clause 0x1
|
|
; GFX11-TRUE16-NEXT: global_load_d16_b16 v0, v2, s[2:3]
|
|
; GFX11-TRUE16-NEXT: global_load_d16_hi_b16 v0, v2, s[4:5] offset:8
|
|
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-TRUE16-NEXT: v_rcp_f16_e32 v1.l, v0.h
|
|
; GFX11-TRUE16-NEXT: s_waitcnt_depctr 0xfff
|
|
; GFX11-TRUE16-NEXT: v_mul_f16_e32 v1.l, v0.l, v1.l
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_trunc_f16_e32 v1.l, v1.l
|
|
; GFX11-TRUE16-NEXT: v_fma_f16 v0.l, -v1.l, v0.h, v0.l
|
|
; GFX11-TRUE16-NEXT: global_store_b16 v2, v0, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_endpgm
|
|
;
|
|
; GFX11-FAKE16-LABEL: fast_frem_f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_clause 0x1
|
|
; GFX11-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX11-FAKE16-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX11-FAKE16-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: s_clause 0x1
|
|
; GFX11-FAKE16-NEXT: global_load_u16 v1, v0, s[2:3]
|
|
; GFX11-FAKE16-NEXT: global_load_u16 v2, v0, s[4:5] offset:8
|
|
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-FAKE16-NEXT: v_rcp_f16_e32 v3, v2
|
|
; GFX11-FAKE16-NEXT: s_waitcnt_depctr 0xfff
|
|
; GFX11-FAKE16-NEXT: v_mul_f16_e32 v3, v1, v3
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_trunc_f16_e32 v3, v3
|
|
; GFX11-FAKE16-NEXT: v_fma_f16 v1, -v3, v2, v1
|
|
; GFX11-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_endpgm
|
|
;
|
|
; GFX1150-LABEL: fast_frem_f16:
|
|
; GFX1150: ; %bb.0:
|
|
; GFX1150-NEXT: s_clause 0x1
|
|
; GFX1150-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX1150-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX1150-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX1150-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX1150-NEXT: s_clause 0x1
|
|
; GFX1150-NEXT: global_load_u16 v1, v0, s[2:3]
|
|
; GFX1150-NEXT: global_load_u16 v2, v0, s[4:5] offset:8
|
|
; GFX1150-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX1150-NEXT: v_rcp_f16_e32 v3, v2
|
|
; GFX1150-NEXT: s_delay_alu instid0(TRANS32_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_mul_f16_e32 v3, v1, v3
|
|
; GFX1150-NEXT: v_trunc_f16_e32 v3, v3
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_xor_b32_e32 v3, 0x8000, v3
|
|
; GFX1150-NEXT: v_fmac_f16_e32 v1, v3, v2
|
|
; GFX1150-NEXT: global_store_b16 v0, v1, s[0:1]
|
|
; GFX1150-NEXT: s_endpgm
|
|
ptr addrspace(1) %in2) #0 {
|
|
%gep2 = getelementptr half, ptr addrspace(1) %in2, i32 4
|
|
%r0 = load half, ptr addrspace(1) %in1, align 4
|
|
%r1 = load half, ptr addrspace(1) %gep2, align 4
|
|
%r2 = frem fast half %r0, %r1
|
|
store half %r2, ptr addrspace(1) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @unsafe_frem_f16(ptr addrspace(1) %out, ptr addrspace(1) %in1,
|
|
; SI-LABEL: unsafe_frem_f16:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9
|
|
; SI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0xd
|
|
; SI-NEXT: s_mov_b32 s11, 0xf000
|
|
; SI-NEXT: s_mov_b32 s10, -1
|
|
; SI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SI-NEXT: s_mov_b32 s8, s0
|
|
; SI-NEXT: s_mov_b32 s9, s1
|
|
; SI-NEXT: s_mov_b32 s0, s2
|
|
; SI-NEXT: s_mov_b32 s1, s3
|
|
; SI-NEXT: s_mov_b32 s2, s10
|
|
; SI-NEXT: s_mov_b32 s3, s11
|
|
; SI-NEXT: s_mov_b32 s6, s10
|
|
; SI-NEXT: s_mov_b32 s7, s11
|
|
; SI-NEXT: buffer_load_ushort v0, off, s[0:3], 0
|
|
; SI-NEXT: s_waitcnt vmcnt(0)
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; SI-NEXT: buffer_load_ushort v1, off, s[4:7], 0 offset:8
|
|
; SI-NEXT: s_waitcnt vmcnt(0)
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v1, v1
|
|
; SI-NEXT: v_rcp_f32_e32 v2, v1
|
|
; SI-NEXT: v_mul_f32_e32 v2, v0, v2
|
|
; SI-NEXT: v_trunc_f32_e32 v2, v2
|
|
; SI-NEXT: v_fma_f32 v0, -v2, v1, v0
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; SI-NEXT: buffer_store_short v0, off, s[8:11], 0
|
|
; SI-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: unsafe_frem_f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9
|
|
; CI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0xd
|
|
; CI-NEXT: s_mov_b32 s11, 0xf000
|
|
; CI-NEXT: s_mov_b32 s10, -1
|
|
; CI-NEXT: s_mov_b32 s6, s10
|
|
; CI-NEXT: s_mov_b32 s7, s11
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_load_ushort v1, off, s[4:7], 0 offset:8
|
|
; CI-NEXT: s_mov_b32 s8, s0
|
|
; CI-NEXT: s_mov_b32 s9, s1
|
|
; CI-NEXT: s_mov_b32 s0, s2
|
|
; CI-NEXT: s_mov_b32 s1, s3
|
|
; CI-NEXT: s_mov_b32 s2, s10
|
|
; CI-NEXT: s_mov_b32 s3, s11
|
|
; CI-NEXT: buffer_load_ushort v0, off, s[0:3], 0
|
|
; CI-NEXT: s_waitcnt vmcnt(1)
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
|
|
; CI-NEXT: v_rcp_f32_e32 v2, v1
|
|
; CI-NEXT: s_waitcnt vmcnt(0)
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; CI-NEXT: v_mul_f32_e32 v2, v0, v2
|
|
; CI-NEXT: v_trunc_f32_e32 v2, v2
|
|
; CI-NEXT: v_fma_f32 v0, -v2, v1, v0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: buffer_store_short v0, off, s[8:11], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; VI-LABEL: unsafe_frem_f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; VI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x34
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: s_add_u32 s0, s4, 8
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: v_mov_b32_e32 v2, s2
|
|
; VI-NEXT: v_mov_b32_e32 v3, s3
|
|
; VI-NEXT: s_addc_u32 s1, s5, 0
|
|
; VI-NEXT: flat_load_ushort v4, v[2:3]
|
|
; VI-NEXT: v_mov_b32_e32 v3, s1
|
|
; VI-NEXT: v_mov_b32_e32 v2, s0
|
|
; VI-NEXT: flat_load_ushort v2, v[2:3]
|
|
; VI-NEXT: s_waitcnt vmcnt(0)
|
|
; VI-NEXT: v_rcp_f16_e32 v3, v2
|
|
; VI-NEXT: v_mul_f16_e32 v3, v4, v3
|
|
; VI-NEXT: v_trunc_f16_e32 v3, v3
|
|
; VI-NEXT: v_fma_f16 v2, -v3, v2, v4
|
|
; VI-NEXT: flat_store_short v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: unsafe_frem_f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; GFX9-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_load_ushort v1, v0, s[2:3]
|
|
; GFX9-NEXT: global_load_ushort v2, v0, s[6:7] offset:8
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX9-NEXT: v_rcp_f16_e32 v3, v2
|
|
; GFX9-NEXT: v_mul_f16_e32 v3, v1, v3
|
|
; GFX9-NEXT: v_trunc_f16_e32 v3, v3
|
|
; GFX9-NEXT: v_fma_f16 v1, -v3, v2, v1
|
|
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; GFX10-LABEL: unsafe_frem_f16:
|
|
; GFX10: ; %bb.0:
|
|
; GFX10-NEXT: s_clause 0x1
|
|
; GFX10-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; GFX10-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
|
|
; GFX10-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX10-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-NEXT: s_clause 0x1
|
|
; GFX10-NEXT: global_load_ushort v1, v0, s[2:3]
|
|
; GFX10-NEXT: global_load_ushort v2, v0, s[6:7] offset:8
|
|
; GFX10-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX10-NEXT: v_rcp_f16_e32 v3, v2
|
|
; GFX10-NEXT: v_mul_f16_e32 v3, v1, v3
|
|
; GFX10-NEXT: v_trunc_f16_e32 v3, v3
|
|
; GFX10-NEXT: v_fma_f16 v1, -v3, v2, v1
|
|
; GFX10-NEXT: global_store_short v0, v1, s[0:1]
|
|
; GFX10-NEXT: s_endpgm
|
|
;
|
|
; GFX11-TRUE16-LABEL: unsafe_frem_f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_clause 0x1
|
|
; GFX11-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX11-TRUE16-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX11-TRUE16-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: s_clause 0x1
|
|
; GFX11-TRUE16-NEXT: global_load_d16_b16 v0, v2, s[2:3]
|
|
; GFX11-TRUE16-NEXT: global_load_d16_hi_b16 v0, v2, s[4:5] offset:8
|
|
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-TRUE16-NEXT: v_rcp_f16_e32 v1.l, v0.h
|
|
; GFX11-TRUE16-NEXT: s_waitcnt_depctr 0xfff
|
|
; GFX11-TRUE16-NEXT: v_mul_f16_e32 v1.l, v0.l, v1.l
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_trunc_f16_e32 v1.l, v1.l
|
|
; GFX11-TRUE16-NEXT: v_fma_f16 v0.l, -v1.l, v0.h, v0.l
|
|
; GFX11-TRUE16-NEXT: global_store_b16 v2, v0, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_endpgm
|
|
;
|
|
; GFX11-FAKE16-LABEL: unsafe_frem_f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_clause 0x1
|
|
; GFX11-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX11-FAKE16-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX11-FAKE16-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: s_clause 0x1
|
|
; GFX11-FAKE16-NEXT: global_load_u16 v1, v0, s[2:3]
|
|
; GFX11-FAKE16-NEXT: global_load_u16 v2, v0, s[4:5] offset:8
|
|
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-FAKE16-NEXT: v_rcp_f16_e32 v3, v2
|
|
; GFX11-FAKE16-NEXT: s_waitcnt_depctr 0xfff
|
|
; GFX11-FAKE16-NEXT: v_mul_f16_e32 v3, v1, v3
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_trunc_f16_e32 v3, v3
|
|
; GFX11-FAKE16-NEXT: v_fma_f16 v1, -v3, v2, v1
|
|
; GFX11-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_endpgm
|
|
;
|
|
; GFX1150-LABEL: unsafe_frem_f16:
|
|
; GFX1150: ; %bb.0:
|
|
; GFX1150-NEXT: s_clause 0x1
|
|
; GFX1150-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX1150-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX1150-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX1150-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX1150-NEXT: s_clause 0x1
|
|
; GFX1150-NEXT: global_load_u16 v1, v0, s[2:3]
|
|
; GFX1150-NEXT: global_load_u16 v2, v0, s[4:5] offset:8
|
|
; GFX1150-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX1150-NEXT: v_rcp_f16_e32 v3, v2
|
|
; GFX1150-NEXT: s_delay_alu instid0(TRANS32_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_mul_f16_e32 v3, v1, v3
|
|
; GFX1150-NEXT: v_trunc_f16_e32 v3, v3
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_xor_b32_e32 v3, 0x8000, v3
|
|
; GFX1150-NEXT: v_fmac_f16_e32 v1, v3, v2
|
|
; GFX1150-NEXT: global_store_b16 v0, v1, s[0:1]
|
|
; GFX1150-NEXT: s_endpgm
|
|
ptr addrspace(1) %in2) #1 {
|
|
%gep2 = getelementptr half, ptr addrspace(1) %in2, i32 4
|
|
%r0 = load half, ptr addrspace(1) %in1, align 4
|
|
%r1 = load half, ptr addrspace(1) %gep2, align 4
|
|
%r2 = frem afn half %r0, %r1
|
|
store half %r2, ptr addrspace(1) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @frem_f32(ptr addrspace(1) %out, ptr addrspace(1) %in1,
|
|
; SI-LABEL: frem_f32:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9
|
|
; SI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0xd
|
|
; SI-NEXT: s_mov_b32 s11, 0xf000
|
|
; SI-NEXT: s_mov_b32 s10, -1
|
|
; SI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SI-NEXT: s_mov_b32 s8, s0
|
|
; SI-NEXT: s_mov_b32 s9, s1
|
|
; SI-NEXT: s_mov_b32 s0, s2
|
|
; SI-NEXT: s_mov_b32 s1, s3
|
|
; SI-NEXT: s_mov_b32 s2, s10
|
|
; SI-NEXT: s_mov_b32 s3, s11
|
|
; SI-NEXT: s_mov_b32 s6, s10
|
|
; SI-NEXT: s_mov_b32 s7, s11
|
|
; SI-NEXT: buffer_load_dword v0, off, s[0:3], 0
|
|
; SI-NEXT: buffer_load_dword v1, off, s[4:7], 0 offset:16
|
|
; SI-NEXT: s_waitcnt vmcnt(0)
|
|
; SI-NEXT: v_div_scale_f32 v2, vcc, v0, v1, v0
|
|
; SI-NEXT: v_div_scale_f32 v3, s[0:1], v1, v1, v0
|
|
; SI-NEXT: v_rcp_f32_e32 v4, v3
|
|
; SI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; SI-NEXT: v_fma_f32 v5, -v3, v4, 1.0
|
|
; SI-NEXT: v_fma_f32 v4, v5, v4, v4
|
|
; SI-NEXT: v_mul_f32_e32 v5, v2, v4
|
|
; SI-NEXT: v_fma_f32 v6, -v3, v5, v2
|
|
; SI-NEXT: v_fma_f32 v5, v6, v4, v5
|
|
; SI-NEXT: v_fma_f32 v2, -v3, v5, v2
|
|
; SI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; SI-NEXT: v_div_fmas_f32 v2, v2, v4, v5
|
|
; SI-NEXT: v_div_fixup_f32 v2, v2, v1, v0
|
|
; SI-NEXT: v_trunc_f32_e32 v2, v2
|
|
; SI-NEXT: v_fma_f32 v0, -v2, v1, v0
|
|
; SI-NEXT: buffer_store_dword v0, off, s[8:11], 0
|
|
; SI-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: frem_f32:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9
|
|
; CI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0xd
|
|
; CI-NEXT: s_mov_b32 s11, 0xf000
|
|
; CI-NEXT: s_mov_b32 s10, -1
|
|
; CI-NEXT: s_mov_b32 s6, s10
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_mov_b32 s8, s0
|
|
; CI-NEXT: s_mov_b32 s9, s1
|
|
; CI-NEXT: s_mov_b32 s0, s2
|
|
; CI-NEXT: s_mov_b32 s1, s3
|
|
; CI-NEXT: s_mov_b32 s2, s10
|
|
; CI-NEXT: s_mov_b32 s3, s11
|
|
; CI-NEXT: s_mov_b32 s7, s11
|
|
; CI-NEXT: buffer_load_dword v0, off, s[0:3], 0
|
|
; CI-NEXT: buffer_load_dword v1, off, s[4:7], 0 offset:16
|
|
; CI-NEXT: s_waitcnt vmcnt(0)
|
|
; CI-NEXT: v_div_scale_f32 v3, s[0:1], v1, v1, v0
|
|
; CI-NEXT: v_div_scale_f32 v2, vcc, v0, v1, v0
|
|
; CI-NEXT: v_rcp_f32_e32 v4, v3
|
|
; CI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; CI-NEXT: v_fma_f32 v5, -v3, v4, 1.0
|
|
; CI-NEXT: v_fma_f32 v4, v5, v4, v4
|
|
; CI-NEXT: v_mul_f32_e32 v5, v2, v4
|
|
; CI-NEXT: v_fma_f32 v6, -v3, v5, v2
|
|
; CI-NEXT: v_fma_f32 v5, v6, v4, v5
|
|
; CI-NEXT: v_fma_f32 v2, -v3, v5, v2
|
|
; CI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; CI-NEXT: v_div_fmas_f32 v2, v2, v4, v5
|
|
; CI-NEXT: v_div_fixup_f32 v2, v2, v1, v0
|
|
; CI-NEXT: v_trunc_f32_e32 v2, v2
|
|
; CI-NEXT: v_fma_f32 v0, -v2, v1, v0
|
|
; CI-NEXT: buffer_store_dword v0, off, s[8:11], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; VI-LABEL: frem_f32:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; VI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x34
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: s_add_u32 s0, s4, 16
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: v_mov_b32_e32 v2, s2
|
|
; VI-NEXT: v_mov_b32_e32 v3, s3
|
|
; VI-NEXT: s_addc_u32 s1, s5, 0
|
|
; VI-NEXT: flat_load_dword v4, v[2:3]
|
|
; VI-NEXT: v_mov_b32_e32 v3, s1
|
|
; VI-NEXT: v_mov_b32_e32 v2, s0
|
|
; VI-NEXT: flat_load_dword v2, v[2:3]
|
|
; VI-NEXT: s_waitcnt vmcnt(0)
|
|
; VI-NEXT: v_div_scale_f32 v5, s[0:1], v2, v2, v4
|
|
; VI-NEXT: v_div_scale_f32 v3, vcc, v4, v2, v4
|
|
; VI-NEXT: v_rcp_f32_e32 v6, v5
|
|
; VI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; VI-NEXT: v_fma_f32 v7, -v5, v6, 1.0
|
|
; VI-NEXT: v_fma_f32 v6, v7, v6, v6
|
|
; VI-NEXT: v_mul_f32_e32 v7, v3, v6
|
|
; VI-NEXT: v_fma_f32 v8, -v5, v7, v3
|
|
; VI-NEXT: v_fma_f32 v7, v8, v6, v7
|
|
; VI-NEXT: v_fma_f32 v3, -v5, v7, v3
|
|
; VI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; VI-NEXT: v_div_fmas_f32 v3, v3, v6, v7
|
|
; VI-NEXT: v_div_fixup_f32 v3, v3, v2, v4
|
|
; VI-NEXT: v_trunc_f32_e32 v3, v3
|
|
; VI-NEXT: v_fma_f32 v2, -v3, v2, v4
|
|
; VI-NEXT: flat_store_dword v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: frem_f32:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; GFX9-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_load_dword v1, v0, s[2:3]
|
|
; GFX9-NEXT: global_load_dword v2, v0, s[6:7] offset:16
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX9-NEXT: v_div_scale_f32 v4, s[2:3], v2, v2, v1
|
|
; GFX9-NEXT: v_div_scale_f32 v3, vcc, v1, v2, v1
|
|
; GFX9-NEXT: v_rcp_f32_e32 v5, v4
|
|
; GFX9-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; GFX9-NEXT: v_fma_f32 v6, -v4, v5, 1.0
|
|
; GFX9-NEXT: v_fma_f32 v5, v6, v5, v5
|
|
; GFX9-NEXT: v_mul_f32_e32 v6, v3, v5
|
|
; GFX9-NEXT: v_fma_f32 v7, -v4, v6, v3
|
|
; GFX9-NEXT: v_fma_f32 v6, v7, v5, v6
|
|
; GFX9-NEXT: v_fma_f32 v3, -v4, v6, v3
|
|
; GFX9-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; GFX9-NEXT: v_div_fmas_f32 v3, v3, v5, v6
|
|
; GFX9-NEXT: v_div_fixup_f32 v3, v3, v2, v1
|
|
; GFX9-NEXT: v_trunc_f32_e32 v3, v3
|
|
; GFX9-NEXT: v_fma_f32 v1, -v3, v2, v1
|
|
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; GFX10-LABEL: frem_f32:
|
|
; GFX10: ; %bb.0:
|
|
; GFX10-NEXT: s_clause 0x1
|
|
; GFX10-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; GFX10-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
|
|
; GFX10-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX10-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-NEXT: s_clause 0x1
|
|
; GFX10-NEXT: global_load_dword v1, v0, s[2:3]
|
|
; GFX10-NEXT: global_load_dword v2, v0, s[6:7] offset:16
|
|
; GFX10-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX10-NEXT: v_div_scale_f32 v4, s2, v2, v2, v1
|
|
; GFX10-NEXT: v_div_scale_f32 v3, vcc_lo, v1, v2, v1
|
|
; GFX10-NEXT: v_rcp_f32_e32 v5, v4
|
|
; GFX10-NEXT: s_denorm_mode 15
|
|
; GFX10-NEXT: v_fma_f32 v6, -v4, v5, 1.0
|
|
; GFX10-NEXT: v_fmac_f32_e32 v5, v6, v5
|
|
; GFX10-NEXT: v_mul_f32_e32 v6, v3, v5
|
|
; GFX10-NEXT: v_fma_f32 v7, -v4, v6, v3
|
|
; GFX10-NEXT: v_fmac_f32_e32 v6, v7, v5
|
|
; GFX10-NEXT: v_fma_f32 v3, -v4, v6, v3
|
|
; GFX10-NEXT: s_denorm_mode 12
|
|
; GFX10-NEXT: v_div_fmas_f32 v3, v3, v5, v6
|
|
; GFX10-NEXT: v_div_fixup_f32 v3, v3, v2, v1
|
|
; GFX10-NEXT: v_trunc_f32_e32 v3, v3
|
|
; GFX10-NEXT: v_fma_f32 v1, -v3, v2, v1
|
|
; GFX10-NEXT: global_store_dword v0, v1, s[0:1]
|
|
; GFX10-NEXT: s_endpgm
|
|
;
|
|
; GFX11-LABEL: frem_f32:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_clause 0x1
|
|
; GFX11-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX11-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX11-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: s_clause 0x1
|
|
; GFX11-NEXT: global_load_b32 v1, v0, s[2:3]
|
|
; GFX11-NEXT: global_load_b32 v2, v0, s[4:5] offset:16
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-NEXT: v_div_scale_f32 v4, null, v2, v2, v1
|
|
; GFX11-NEXT: v_div_scale_f32 v3, vcc_lo, v1, v2, v1
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_3) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_rcp_f32_e32 v5, v4
|
|
; GFX11-NEXT: s_denorm_mode 15
|
|
; GFX11-NEXT: s_waitcnt_depctr 0xfff
|
|
; GFX11-NEXT: v_fma_f32 v6, -v4, v5, 1.0
|
|
; GFX11-NEXT: v_fmac_f32_e32 v5, v6, v5
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_mul_f32_e32 v6, v3, v5
|
|
; GFX11-NEXT: v_fma_f32 v7, -v4, v6, v3
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_fmac_f32_e32 v6, v7, v5
|
|
; GFX11-NEXT: v_fma_f32 v3, -v4, v6, v3
|
|
; GFX11-NEXT: s_denorm_mode 12
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_div_fmas_f32 v3, v3, v5, v6
|
|
; GFX11-NEXT: v_div_fixup_f32 v3, v3, v2, v1
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_trunc_f32_e32 v3, v3
|
|
; GFX11-NEXT: v_fma_f32 v1, -v3, v2, v1
|
|
; GFX11-NEXT: global_store_b32 v0, v1, s[0:1]
|
|
; GFX11-NEXT: s_endpgm
|
|
;
|
|
; GFX1150-LABEL: frem_f32:
|
|
; GFX1150: ; %bb.0:
|
|
; GFX1150-NEXT: s_clause 0x1
|
|
; GFX1150-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX1150-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX1150-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX1150-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX1150-NEXT: s_clause 0x1
|
|
; GFX1150-NEXT: global_load_b32 v1, v0, s[2:3]
|
|
; GFX1150-NEXT: global_load_b32 v2, v0, s[4:5] offset:16
|
|
; GFX1150-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX1150-NEXT: v_div_scale_f32 v4, null, v2, v2, v1
|
|
; GFX1150-NEXT: v_div_scale_f32 v3, vcc_lo, v1, v2, v1
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(TRANS32_DEP_1)
|
|
; GFX1150-NEXT: v_rcp_f32_e32 v5, v4
|
|
; GFX1150-NEXT: s_denorm_mode 15
|
|
; GFX1150-NEXT: v_fma_f32 v6, -v4, v5, 1.0
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fmac_f32_e32 v5, v6, v5
|
|
; GFX1150-NEXT: v_mul_f32_e32 v6, v3, v5
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_f32 v7, -v4, v6, v3
|
|
; GFX1150-NEXT: v_fmac_f32_e32 v6, v7, v5
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_f32 v3, -v4, v6, v3
|
|
; GFX1150-NEXT: s_denorm_mode 12
|
|
; GFX1150-NEXT: v_div_fmas_f32 v3, v3, v5, v6
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_div_fixup_f32 v3, v3, v2, v1
|
|
; GFX1150-NEXT: v_trunc_f32_e32 v3, v3
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_xor_b32_e32 v3, 0x80000000, v3
|
|
; GFX1150-NEXT: v_fmac_f32_e32 v1, v3, v2
|
|
; GFX1150-NEXT: global_store_b32 v0, v1, s[0:1]
|
|
; GFX1150-NEXT: s_endpgm
|
|
ptr addrspace(1) %in2) #0 {
|
|
%gep2 = getelementptr float, ptr addrspace(1) %in2, i32 4
|
|
%r0 = load float, ptr addrspace(1) %in1, align 4
|
|
%r1 = load float, ptr addrspace(1) %gep2, align 4
|
|
%r2 = frem float %r0, %r1
|
|
store float %r2, ptr addrspace(1) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @fast_frem_f32(ptr addrspace(1) %out, ptr addrspace(1) %in1,
|
|
; SI-LABEL: fast_frem_f32:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9
|
|
; SI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0xd
|
|
; SI-NEXT: s_mov_b32 s11, 0xf000
|
|
; SI-NEXT: s_mov_b32 s10, -1
|
|
; SI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SI-NEXT: s_mov_b32 s8, s0
|
|
; SI-NEXT: s_mov_b32 s9, s1
|
|
; SI-NEXT: s_mov_b32 s0, s2
|
|
; SI-NEXT: s_mov_b32 s1, s3
|
|
; SI-NEXT: s_mov_b32 s2, s10
|
|
; SI-NEXT: s_mov_b32 s3, s11
|
|
; SI-NEXT: s_mov_b32 s6, s10
|
|
; SI-NEXT: s_mov_b32 s7, s11
|
|
; SI-NEXT: buffer_load_dword v0, off, s[0:3], 0
|
|
; SI-NEXT: buffer_load_dword v1, off, s[4:7], 0 offset:16
|
|
; SI-NEXT: s_waitcnt vmcnt(0)
|
|
; SI-NEXT: v_rcp_f32_e32 v2, v1
|
|
; SI-NEXT: v_mul_f32_e32 v2, v0, v2
|
|
; SI-NEXT: v_trunc_f32_e32 v2, v2
|
|
; SI-NEXT: v_fma_f32 v0, -v2, v1, v0
|
|
; SI-NEXT: buffer_store_dword v0, off, s[8:11], 0
|
|
; SI-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: fast_frem_f32:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9
|
|
; CI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0xd
|
|
; CI-NEXT: s_mov_b32 s11, 0xf000
|
|
; CI-NEXT: s_mov_b32 s10, -1
|
|
; CI-NEXT: s_mov_b32 s6, s10
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_mov_b32 s8, s0
|
|
; CI-NEXT: s_mov_b32 s9, s1
|
|
; CI-NEXT: s_mov_b32 s0, s2
|
|
; CI-NEXT: s_mov_b32 s1, s3
|
|
; CI-NEXT: s_mov_b32 s2, s10
|
|
; CI-NEXT: s_mov_b32 s3, s11
|
|
; CI-NEXT: s_mov_b32 s7, s11
|
|
; CI-NEXT: buffer_load_dword v0, off, s[0:3], 0
|
|
; CI-NEXT: buffer_load_dword v1, off, s[4:7], 0 offset:16
|
|
; CI-NEXT: s_waitcnt vmcnt(0)
|
|
; CI-NEXT: v_rcp_f32_e32 v2, v1
|
|
; CI-NEXT: v_mul_f32_e32 v2, v0, v2
|
|
; CI-NEXT: v_trunc_f32_e32 v2, v2
|
|
; CI-NEXT: v_fma_f32 v0, -v2, v1, v0
|
|
; CI-NEXT: buffer_store_dword v0, off, s[8:11], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; VI-LABEL: fast_frem_f32:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; VI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x34
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: s_add_u32 s0, s4, 16
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: v_mov_b32_e32 v2, s2
|
|
; VI-NEXT: v_mov_b32_e32 v3, s3
|
|
; VI-NEXT: s_addc_u32 s1, s5, 0
|
|
; VI-NEXT: flat_load_dword v4, v[2:3]
|
|
; VI-NEXT: v_mov_b32_e32 v3, s1
|
|
; VI-NEXT: v_mov_b32_e32 v2, s0
|
|
; VI-NEXT: flat_load_dword v2, v[2:3]
|
|
; VI-NEXT: s_waitcnt vmcnt(0)
|
|
; VI-NEXT: v_rcp_f32_e32 v3, v2
|
|
; VI-NEXT: v_mul_f32_e32 v3, v4, v3
|
|
; VI-NEXT: v_trunc_f32_e32 v3, v3
|
|
; VI-NEXT: v_fma_f32 v2, -v3, v2, v4
|
|
; VI-NEXT: flat_store_dword v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: fast_frem_f32:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; GFX9-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_load_dword v1, v0, s[2:3]
|
|
; GFX9-NEXT: global_load_dword v2, v0, s[6:7] offset:16
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX9-NEXT: v_rcp_f32_e32 v3, v2
|
|
; GFX9-NEXT: v_mul_f32_e32 v3, v1, v3
|
|
; GFX9-NEXT: v_trunc_f32_e32 v3, v3
|
|
; GFX9-NEXT: v_fma_f32 v1, -v3, v2, v1
|
|
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; GFX10-LABEL: fast_frem_f32:
|
|
; GFX10: ; %bb.0:
|
|
; GFX10-NEXT: s_clause 0x1
|
|
; GFX10-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; GFX10-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
|
|
; GFX10-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX10-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-NEXT: s_clause 0x1
|
|
; GFX10-NEXT: global_load_dword v1, v0, s[2:3]
|
|
; GFX10-NEXT: global_load_dword v2, v0, s[6:7] offset:16
|
|
; GFX10-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX10-NEXT: v_rcp_f32_e32 v3, v2
|
|
; GFX10-NEXT: v_mul_f32_e32 v3, v1, v3
|
|
; GFX10-NEXT: v_trunc_f32_e32 v3, v3
|
|
; GFX10-NEXT: v_fma_f32 v1, -v3, v2, v1
|
|
; GFX10-NEXT: global_store_dword v0, v1, s[0:1]
|
|
; GFX10-NEXT: s_endpgm
|
|
;
|
|
; GFX11-LABEL: fast_frem_f32:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_clause 0x1
|
|
; GFX11-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX11-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX11-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: s_clause 0x1
|
|
; GFX11-NEXT: global_load_b32 v1, v0, s[2:3]
|
|
; GFX11-NEXT: global_load_b32 v2, v0, s[4:5] offset:16
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-NEXT: v_rcp_f32_e32 v3, v2
|
|
; GFX11-NEXT: s_waitcnt_depctr 0xfff
|
|
; GFX11-NEXT: v_mul_f32_e32 v3, v1, v3
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_trunc_f32_e32 v3, v3
|
|
; GFX11-NEXT: v_fma_f32 v1, -v3, v2, v1
|
|
; GFX11-NEXT: global_store_b32 v0, v1, s[0:1]
|
|
; GFX11-NEXT: s_endpgm
|
|
;
|
|
; GFX1150-LABEL: fast_frem_f32:
|
|
; GFX1150: ; %bb.0:
|
|
; GFX1150-NEXT: s_clause 0x1
|
|
; GFX1150-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX1150-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX1150-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX1150-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX1150-NEXT: s_clause 0x1
|
|
; GFX1150-NEXT: global_load_b32 v1, v0, s[2:3]
|
|
; GFX1150-NEXT: global_load_b32 v2, v0, s[4:5] offset:16
|
|
; GFX1150-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX1150-NEXT: v_rcp_f32_e32 v3, v2
|
|
; GFX1150-NEXT: s_delay_alu instid0(TRANS32_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_mul_f32_e32 v3, v1, v3
|
|
; GFX1150-NEXT: v_trunc_f32_e32 v3, v3
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_xor_b32_e32 v3, 0x80000000, v3
|
|
; GFX1150-NEXT: v_fmac_f32_e32 v1, v3, v2
|
|
; GFX1150-NEXT: global_store_b32 v0, v1, s[0:1]
|
|
; GFX1150-NEXT: s_endpgm
|
|
ptr addrspace(1) %in2) #0 {
|
|
%gep2 = getelementptr float, ptr addrspace(1) %in2, i32 4
|
|
%r0 = load float, ptr addrspace(1) %in1, align 4
|
|
%r1 = load float, ptr addrspace(1) %gep2, align 4
|
|
%r2 = frem fast float %r0, %r1
|
|
store float %r2, ptr addrspace(1) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @unsafe_frem_f32(ptr addrspace(1) %out, ptr addrspace(1) %in1,
|
|
; SI-LABEL: unsafe_frem_f32:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9
|
|
; SI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0xd
|
|
; SI-NEXT: s_mov_b32 s11, 0xf000
|
|
; SI-NEXT: s_mov_b32 s10, -1
|
|
; SI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SI-NEXT: s_mov_b32 s8, s0
|
|
; SI-NEXT: s_mov_b32 s9, s1
|
|
; SI-NEXT: s_mov_b32 s0, s2
|
|
; SI-NEXT: s_mov_b32 s1, s3
|
|
; SI-NEXT: s_mov_b32 s2, s10
|
|
; SI-NEXT: s_mov_b32 s3, s11
|
|
; SI-NEXT: s_mov_b32 s6, s10
|
|
; SI-NEXT: s_mov_b32 s7, s11
|
|
; SI-NEXT: buffer_load_dword v0, off, s[0:3], 0
|
|
; SI-NEXT: buffer_load_dword v1, off, s[4:7], 0 offset:16
|
|
; SI-NEXT: s_waitcnt vmcnt(0)
|
|
; SI-NEXT: v_rcp_f32_e32 v2, v1
|
|
; SI-NEXT: v_mul_f32_e32 v2, v0, v2
|
|
; SI-NEXT: v_trunc_f32_e32 v2, v2
|
|
; SI-NEXT: v_fma_f32 v0, -v2, v1, v0
|
|
; SI-NEXT: buffer_store_dword v0, off, s[8:11], 0
|
|
; SI-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: unsafe_frem_f32:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9
|
|
; CI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0xd
|
|
; CI-NEXT: s_mov_b32 s11, 0xf000
|
|
; CI-NEXT: s_mov_b32 s10, -1
|
|
; CI-NEXT: s_mov_b32 s6, s10
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_mov_b32 s8, s0
|
|
; CI-NEXT: s_mov_b32 s9, s1
|
|
; CI-NEXT: s_mov_b32 s0, s2
|
|
; CI-NEXT: s_mov_b32 s1, s3
|
|
; CI-NEXT: s_mov_b32 s2, s10
|
|
; CI-NEXT: s_mov_b32 s3, s11
|
|
; CI-NEXT: s_mov_b32 s7, s11
|
|
; CI-NEXT: buffer_load_dword v0, off, s[0:3], 0
|
|
; CI-NEXT: buffer_load_dword v1, off, s[4:7], 0 offset:16
|
|
; CI-NEXT: s_waitcnt vmcnt(0)
|
|
; CI-NEXT: v_rcp_f32_e32 v2, v1
|
|
; CI-NEXT: v_mul_f32_e32 v2, v0, v2
|
|
; CI-NEXT: v_trunc_f32_e32 v2, v2
|
|
; CI-NEXT: v_fma_f32 v0, -v2, v1, v0
|
|
; CI-NEXT: buffer_store_dword v0, off, s[8:11], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; VI-LABEL: unsafe_frem_f32:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; VI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x34
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: s_add_u32 s0, s4, 16
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: v_mov_b32_e32 v2, s2
|
|
; VI-NEXT: v_mov_b32_e32 v3, s3
|
|
; VI-NEXT: s_addc_u32 s1, s5, 0
|
|
; VI-NEXT: flat_load_dword v4, v[2:3]
|
|
; VI-NEXT: v_mov_b32_e32 v3, s1
|
|
; VI-NEXT: v_mov_b32_e32 v2, s0
|
|
; VI-NEXT: flat_load_dword v2, v[2:3]
|
|
; VI-NEXT: s_waitcnt vmcnt(0)
|
|
; VI-NEXT: v_rcp_f32_e32 v3, v2
|
|
; VI-NEXT: v_mul_f32_e32 v3, v4, v3
|
|
; VI-NEXT: v_trunc_f32_e32 v3, v3
|
|
; VI-NEXT: v_fma_f32 v2, -v3, v2, v4
|
|
; VI-NEXT: flat_store_dword v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: unsafe_frem_f32:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; GFX9-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_load_dword v1, v0, s[2:3]
|
|
; GFX9-NEXT: global_load_dword v2, v0, s[6:7] offset:16
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX9-NEXT: v_rcp_f32_e32 v3, v2
|
|
; GFX9-NEXT: v_mul_f32_e32 v3, v1, v3
|
|
; GFX9-NEXT: v_trunc_f32_e32 v3, v3
|
|
; GFX9-NEXT: v_fma_f32 v1, -v3, v2, v1
|
|
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; GFX10-LABEL: unsafe_frem_f32:
|
|
; GFX10: ; %bb.0:
|
|
; GFX10-NEXT: s_clause 0x1
|
|
; GFX10-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; GFX10-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
|
|
; GFX10-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX10-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-NEXT: s_clause 0x1
|
|
; GFX10-NEXT: global_load_dword v1, v0, s[2:3]
|
|
; GFX10-NEXT: global_load_dword v2, v0, s[6:7] offset:16
|
|
; GFX10-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX10-NEXT: v_rcp_f32_e32 v3, v2
|
|
; GFX10-NEXT: v_mul_f32_e32 v3, v1, v3
|
|
; GFX10-NEXT: v_trunc_f32_e32 v3, v3
|
|
; GFX10-NEXT: v_fma_f32 v1, -v3, v2, v1
|
|
; GFX10-NEXT: global_store_dword v0, v1, s[0:1]
|
|
; GFX10-NEXT: s_endpgm
|
|
;
|
|
; GFX11-LABEL: unsafe_frem_f32:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_clause 0x1
|
|
; GFX11-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX11-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX11-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: s_clause 0x1
|
|
; GFX11-NEXT: global_load_b32 v1, v0, s[2:3]
|
|
; GFX11-NEXT: global_load_b32 v2, v0, s[4:5] offset:16
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-NEXT: v_rcp_f32_e32 v3, v2
|
|
; GFX11-NEXT: s_waitcnt_depctr 0xfff
|
|
; GFX11-NEXT: v_mul_f32_e32 v3, v1, v3
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_trunc_f32_e32 v3, v3
|
|
; GFX11-NEXT: v_fma_f32 v1, -v3, v2, v1
|
|
; GFX11-NEXT: global_store_b32 v0, v1, s[0:1]
|
|
; GFX11-NEXT: s_endpgm
|
|
;
|
|
; GFX1150-LABEL: unsafe_frem_f32:
|
|
; GFX1150: ; %bb.0:
|
|
; GFX1150-NEXT: s_clause 0x1
|
|
; GFX1150-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX1150-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX1150-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX1150-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX1150-NEXT: s_clause 0x1
|
|
; GFX1150-NEXT: global_load_b32 v1, v0, s[2:3]
|
|
; GFX1150-NEXT: global_load_b32 v2, v0, s[4:5] offset:16
|
|
; GFX1150-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX1150-NEXT: v_rcp_f32_e32 v3, v2
|
|
; GFX1150-NEXT: s_delay_alu instid0(TRANS32_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_mul_f32_e32 v3, v1, v3
|
|
; GFX1150-NEXT: v_trunc_f32_e32 v3, v3
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_xor_b32_e32 v3, 0x80000000, v3
|
|
; GFX1150-NEXT: v_fmac_f32_e32 v1, v3, v2
|
|
; GFX1150-NEXT: global_store_b32 v0, v1, s[0:1]
|
|
; GFX1150-NEXT: s_endpgm
|
|
ptr addrspace(1) %in2) #1 {
|
|
%gep2 = getelementptr float, ptr addrspace(1) %in2, i32 4
|
|
%r0 = load float, ptr addrspace(1) %in1, align 4
|
|
%r1 = load float, ptr addrspace(1) %gep2, align 4
|
|
%r2 = frem afn float %r0, %r1
|
|
store float %r2, ptr addrspace(1) %out, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @frem_f64(ptr addrspace(1) %out, ptr addrspace(1) %in1,
|
|
; SI-LABEL: frem_f64:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9
|
|
; SI-NEXT: s_load_dwordx2 s[8:9], s[4:5], 0xd
|
|
; SI-NEXT: s_mov_b32 s7, 0xf000
|
|
; SI-NEXT: s_mov_b32 s6, -1
|
|
; SI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SI-NEXT: s_mov_b32 s4, s0
|
|
; SI-NEXT: s_mov_b32 s5, s1
|
|
; SI-NEXT: s_mov_b32 s0, s2
|
|
; SI-NEXT: s_mov_b32 s1, s3
|
|
; SI-NEXT: s_mov_b32 s2, s6
|
|
; SI-NEXT: s_mov_b32 s3, s7
|
|
; SI-NEXT: s_mov_b32 s10, s6
|
|
; SI-NEXT: s_mov_b32 s11, s7
|
|
; SI-NEXT: buffer_load_dwordx2 v[0:1], off, s[0:3], 0
|
|
; SI-NEXT: buffer_load_dwordx2 v[2:3], off, s[8:11], 0
|
|
; SI-NEXT: s_waitcnt vmcnt(0)
|
|
; SI-NEXT: v_div_scale_f64 v[4:5], s[0:1], v[2:3], v[2:3], v[0:1]
|
|
; SI-NEXT: v_rcp_f64_e32 v[6:7], v[4:5]
|
|
; SI-NEXT: v_fma_f64 v[8:9], -v[4:5], v[6:7], 1.0
|
|
; SI-NEXT: v_fma_f64 v[6:7], v[6:7], v[8:9], v[6:7]
|
|
; SI-NEXT: v_fma_f64 v[8:9], -v[4:5], v[6:7], 1.0
|
|
; SI-NEXT: v_fma_f64 v[6:7], v[6:7], v[8:9], v[6:7]
|
|
; SI-NEXT: v_div_scale_f64 v[8:9], s[0:1], v[0:1], v[2:3], v[0:1]
|
|
; SI-NEXT: v_mul_f64 v[10:11], v[8:9], v[6:7]
|
|
; SI-NEXT: v_fma_f64 v[12:13], -v[4:5], v[10:11], v[8:9]
|
|
; SI-NEXT: v_cmp_eq_u32_e32 vcc, v3, v5
|
|
; SI-NEXT: v_cmp_eq_u32_e64 s[0:1], v1, v9
|
|
; SI-NEXT: s_xor_b64 vcc, s[0:1], vcc
|
|
; SI-NEXT: s_nop 1
|
|
; SI-NEXT: v_div_fmas_f64 v[4:5], v[12:13], v[6:7], v[10:11]
|
|
; SI-NEXT: v_div_fixup_f64 v[4:5], v[4:5], v[2:3], v[0:1]
|
|
; SI-NEXT: v_readfirstlane_b32 s2, v5
|
|
; SI-NEXT: s_bfe_u32 s0, s2, 0xb0014
|
|
; SI-NEXT: s_add_i32 s3, s0, 0xfffffc01
|
|
; SI-NEXT: s_mov_b32 s1, 0xfffff
|
|
; SI-NEXT: s_mov_b32 s0, s6
|
|
; SI-NEXT: s_lshr_b64 s[0:1], s[0:1], s3
|
|
; SI-NEXT: v_not_b32_e32 v6, s0
|
|
; SI-NEXT: v_and_b32_e32 v6, v4, v6
|
|
; SI-NEXT: v_not_b32_e32 v7, s1
|
|
; SI-NEXT: v_and_b32_e32 v5, v5, v7
|
|
; SI-NEXT: s_and_b32 s0, s2, 0x80000000
|
|
; SI-NEXT: s_cmp_lt_i32 s3, 0
|
|
; SI-NEXT: s_cselect_b64 vcc, -1, 0
|
|
; SI-NEXT: v_cndmask_b32_e64 v6, v6, 0, vcc
|
|
; SI-NEXT: v_mov_b32_e32 v7, s0
|
|
; SI-NEXT: v_cndmask_b32_e32 v5, v5, v7, vcc
|
|
; SI-NEXT: s_cmp_gt_i32 s3, 51
|
|
; SI-NEXT: s_cselect_b64 vcc, -1, 0
|
|
; SI-NEXT: v_mov_b32_e32 v7, s2
|
|
; SI-NEXT: v_cndmask_b32_e32 v5, v5, v7, vcc
|
|
; SI-NEXT: v_cndmask_b32_e32 v4, v6, v4, vcc
|
|
; SI-NEXT: v_fma_f64 v[0:1], -v[4:5], v[2:3], v[0:1]
|
|
; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0
|
|
; SI-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: frem_f64:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9
|
|
; CI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0xd
|
|
; CI-NEXT: s_mov_b32 s11, 0xf000
|
|
; CI-NEXT: s_mov_b32 s10, -1
|
|
; CI-NEXT: s_mov_b32 s6, s10
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_mov_b32 s8, s0
|
|
; CI-NEXT: s_mov_b32 s9, s1
|
|
; CI-NEXT: s_mov_b32 s0, s2
|
|
; CI-NEXT: s_mov_b32 s1, s3
|
|
; CI-NEXT: s_mov_b32 s2, s10
|
|
; CI-NEXT: s_mov_b32 s3, s11
|
|
; CI-NEXT: s_mov_b32 s7, s11
|
|
; CI-NEXT: buffer_load_dwordx2 v[0:1], off, s[0:3], 0
|
|
; CI-NEXT: buffer_load_dwordx2 v[2:3], off, s[4:7], 0
|
|
; CI-NEXT: s_waitcnt vmcnt(0)
|
|
; CI-NEXT: v_div_scale_f64 v[4:5], s[0:1], v[2:3], v[2:3], v[0:1]
|
|
; CI-NEXT: v_rcp_f64_e32 v[6:7], v[4:5]
|
|
; CI-NEXT: v_fma_f64 v[8:9], -v[4:5], v[6:7], 1.0
|
|
; CI-NEXT: v_fma_f64 v[6:7], v[6:7], v[8:9], v[6:7]
|
|
; CI-NEXT: v_fma_f64 v[8:9], -v[4:5], v[6:7], 1.0
|
|
; CI-NEXT: v_fma_f64 v[6:7], v[6:7], v[8:9], v[6:7]
|
|
; CI-NEXT: v_div_scale_f64 v[8:9], vcc, v[0:1], v[2:3], v[0:1]
|
|
; CI-NEXT: v_mul_f64 v[10:11], v[8:9], v[6:7]
|
|
; CI-NEXT: v_fma_f64 v[4:5], -v[4:5], v[10:11], v[8:9]
|
|
; CI-NEXT: s_nop 1
|
|
; CI-NEXT: v_div_fmas_f64 v[4:5], v[4:5], v[6:7], v[10:11]
|
|
; CI-NEXT: v_div_fixup_f64 v[4:5], v[4:5], v[2:3], v[0:1]
|
|
; CI-NEXT: v_trunc_f64_e32 v[4:5], v[4:5]
|
|
; CI-NEXT: v_fma_f64 v[0:1], -v[4:5], v[2:3], v[0:1]
|
|
; CI-NEXT: buffer_store_dwordx2 v[0:1], off, s[8:11], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; VI-LABEL: frem_f64:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; VI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x34
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v2, s2
|
|
; VI-NEXT: v_mov_b32_e32 v3, s3
|
|
; VI-NEXT: v_mov_b32_e32 v4, s4
|
|
; VI-NEXT: v_mov_b32_e32 v5, s5
|
|
; VI-NEXT: flat_load_dwordx2 v[2:3], v[2:3]
|
|
; VI-NEXT: flat_load_dwordx2 v[4:5], v[4:5]
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: s_waitcnt vmcnt(0)
|
|
; VI-NEXT: v_div_scale_f64 v[6:7], s[0:1], v[4:5], v[4:5], v[2:3]
|
|
; VI-NEXT: v_rcp_f64_e32 v[8:9], v[6:7]
|
|
; VI-NEXT: v_fma_f64 v[10:11], -v[6:7], v[8:9], 1.0
|
|
; VI-NEXT: v_fma_f64 v[8:9], v[8:9], v[10:11], v[8:9]
|
|
; VI-NEXT: v_fma_f64 v[10:11], -v[6:7], v[8:9], 1.0
|
|
; VI-NEXT: v_fma_f64 v[8:9], v[8:9], v[10:11], v[8:9]
|
|
; VI-NEXT: v_div_scale_f64 v[10:11], vcc, v[2:3], v[4:5], v[2:3]
|
|
; VI-NEXT: v_mul_f64 v[12:13], v[10:11], v[8:9]
|
|
; VI-NEXT: v_fma_f64 v[6:7], -v[6:7], v[12:13], v[10:11]
|
|
; VI-NEXT: s_nop 1
|
|
; VI-NEXT: v_div_fmas_f64 v[6:7], v[6:7], v[8:9], v[12:13]
|
|
; VI-NEXT: v_div_fixup_f64 v[6:7], v[6:7], v[4:5], v[2:3]
|
|
; VI-NEXT: v_trunc_f64_e32 v[6:7], v[6:7]
|
|
; VI-NEXT: v_fma_f64 v[2:3], -v[6:7], v[4:5], v[2:3]
|
|
; VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3]
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: frem_f64:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; GFX9-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
|
|
; GFX9-NEXT: v_mov_b32_e32 v12, 0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_load_dwordx2 v[0:1], v12, s[2:3]
|
|
; GFX9-NEXT: global_load_dwordx2 v[2:3], v12, s[6:7]
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX9-NEXT: v_div_scale_f64 v[4:5], s[2:3], v[2:3], v[2:3], v[0:1]
|
|
; GFX9-NEXT: v_rcp_f64_e32 v[6:7], v[4:5]
|
|
; GFX9-NEXT: v_fma_f64 v[8:9], -v[4:5], v[6:7], 1.0
|
|
; GFX9-NEXT: v_fma_f64 v[6:7], v[6:7], v[8:9], v[6:7]
|
|
; GFX9-NEXT: v_fma_f64 v[8:9], -v[4:5], v[6:7], 1.0
|
|
; GFX9-NEXT: v_fma_f64 v[6:7], v[6:7], v[8:9], v[6:7]
|
|
; GFX9-NEXT: v_div_scale_f64 v[8:9], vcc, v[0:1], v[2:3], v[0:1]
|
|
; GFX9-NEXT: v_mul_f64 v[10:11], v[8:9], v[6:7]
|
|
; GFX9-NEXT: v_fma_f64 v[4:5], -v[4:5], v[10:11], v[8:9]
|
|
; GFX9-NEXT: s_nop 1
|
|
; GFX9-NEXT: v_div_fmas_f64 v[4:5], v[4:5], v[6:7], v[10:11]
|
|
; GFX9-NEXT: v_div_fixup_f64 v[4:5], v[4:5], v[2:3], v[0:1]
|
|
; GFX9-NEXT: v_trunc_f64_e32 v[4:5], v[4:5]
|
|
; GFX9-NEXT: v_fma_f64 v[0:1], -v[4:5], v[2:3], v[0:1]
|
|
; GFX9-NEXT: global_store_dwordx2 v12, v[0:1], s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; GFX10-LABEL: frem_f64:
|
|
; GFX10: ; %bb.0:
|
|
; GFX10-NEXT: s_clause 0x1
|
|
; GFX10-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; GFX10-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
|
|
; GFX10-NEXT: v_mov_b32_e32 v12, 0
|
|
; GFX10-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-NEXT: s_clause 0x1
|
|
; GFX10-NEXT: global_load_dwordx2 v[0:1], v12, s[2:3]
|
|
; GFX10-NEXT: global_load_dwordx2 v[2:3], v12, s[6:7]
|
|
; GFX10-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX10-NEXT: v_div_scale_f64 v[4:5], s2, v[2:3], v[2:3], v[0:1]
|
|
; GFX10-NEXT: v_rcp_f64_e32 v[6:7], v[4:5]
|
|
; GFX10-NEXT: v_fma_f64 v[8:9], -v[4:5], v[6:7], 1.0
|
|
; GFX10-NEXT: v_fma_f64 v[6:7], v[6:7], v[8:9], v[6:7]
|
|
; GFX10-NEXT: v_fma_f64 v[8:9], -v[4:5], v[6:7], 1.0
|
|
; GFX10-NEXT: v_fma_f64 v[6:7], v[6:7], v[8:9], v[6:7]
|
|
; GFX10-NEXT: v_div_scale_f64 v[8:9], vcc_lo, v[0:1], v[2:3], v[0:1]
|
|
; GFX10-NEXT: v_mul_f64 v[10:11], v[8:9], v[6:7]
|
|
; GFX10-NEXT: v_fma_f64 v[4:5], -v[4:5], v[10:11], v[8:9]
|
|
; GFX10-NEXT: v_div_fmas_f64 v[4:5], v[4:5], v[6:7], v[10:11]
|
|
; GFX10-NEXT: v_div_fixup_f64 v[4:5], v[4:5], v[2:3], v[0:1]
|
|
; GFX10-NEXT: v_trunc_f64_e32 v[4:5], v[4:5]
|
|
; GFX10-NEXT: v_fma_f64 v[0:1], -v[4:5], v[2:3], v[0:1]
|
|
; GFX10-NEXT: global_store_dwordx2 v12, v[0:1], s[0:1]
|
|
; GFX10-NEXT: s_endpgm
|
|
;
|
|
; GFX11-LABEL: frem_f64:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_clause 0x1
|
|
; GFX11-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX11-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX11-NEXT: v_mov_b32_e32 v12, 0
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: s_clause 0x1
|
|
; GFX11-NEXT: global_load_b64 v[0:1], v12, s[2:3]
|
|
; GFX11-NEXT: global_load_b64 v[2:3], v12, s[4:5]
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-NEXT: v_div_scale_f64 v[4:5], null, v[2:3], v[2:3], v[0:1]
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_rcp_f64_e32 v[6:7], v[4:5]
|
|
; GFX11-NEXT: s_waitcnt_depctr 0xfff
|
|
; GFX11-NEXT: v_fma_f64 v[8:9], -v[4:5], v[6:7], 1.0
|
|
; GFX11-NEXT: v_fma_f64 v[6:7], v[6:7], v[8:9], v[6:7]
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_fma_f64 v[8:9], -v[4:5], v[6:7], 1.0
|
|
; GFX11-NEXT: v_fma_f64 v[6:7], v[6:7], v[8:9], v[6:7]
|
|
; GFX11-NEXT: v_div_scale_f64 v[8:9], vcc_lo, v[0:1], v[2:3], v[0:1]
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_mul_f64 v[10:11], v[8:9], v[6:7]
|
|
; GFX11-NEXT: v_fma_f64 v[4:5], -v[4:5], v[10:11], v[8:9]
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_div_fmas_f64 v[4:5], v[4:5], v[6:7], v[10:11]
|
|
; GFX11-NEXT: v_div_fixup_f64 v[4:5], v[4:5], v[2:3], v[0:1]
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_trunc_f64_e32 v[4:5], v[4:5]
|
|
; GFX11-NEXT: v_fma_f64 v[0:1], -v[4:5], v[2:3], v[0:1]
|
|
; GFX11-NEXT: global_store_b64 v12, v[0:1], s[0:1]
|
|
; GFX11-NEXT: s_endpgm
|
|
;
|
|
; GFX1150-LABEL: frem_f64:
|
|
; GFX1150: ; %bb.0:
|
|
; GFX1150-NEXT: s_clause 0x1
|
|
; GFX1150-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX1150-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX1150-NEXT: v_mov_b32_e32 v12, 0
|
|
; GFX1150-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX1150-NEXT: s_clause 0x1
|
|
; GFX1150-NEXT: global_load_b64 v[0:1], v12, s[2:3]
|
|
; GFX1150-NEXT: global_load_b64 v[2:3], v12, s[4:5]
|
|
; GFX1150-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX1150-NEXT: v_div_scale_f64 v[4:5], null, v[2:3], v[2:3], v[0:1]
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(TRANS32_DEP_1)
|
|
; GFX1150-NEXT: v_rcp_f64_e32 v[6:7], v[4:5]
|
|
; GFX1150-NEXT: v_fma_f64 v[8:9], -v[4:5], v[6:7], 1.0
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_f64 v[6:7], v[6:7], v[8:9], v[6:7]
|
|
; GFX1150-NEXT: v_fma_f64 v[8:9], -v[4:5], v[6:7], 1.0
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_f64 v[6:7], v[6:7], v[8:9], v[6:7]
|
|
; GFX1150-NEXT: v_div_scale_f64 v[8:9], vcc_lo, v[0:1], v[2:3], v[0:1]
|
|
; GFX1150-NEXT: v_mul_f64 v[10:11], v[8:9], v[6:7]
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_f64 v[4:5], -v[4:5], v[10:11], v[8:9]
|
|
; GFX1150-NEXT: v_div_fmas_f64 v[4:5], v[4:5], v[6:7], v[10:11]
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_div_fixup_f64 v[4:5], v[4:5], v[2:3], v[0:1]
|
|
; GFX1150-NEXT: v_trunc_f64_e32 v[4:5], v[4:5]
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_f64 v[0:1], -v[4:5], v[2:3], v[0:1]
|
|
; GFX1150-NEXT: global_store_b64 v12, v[0:1], s[0:1]
|
|
; GFX1150-NEXT: s_endpgm
|
|
ptr addrspace(1) %in2) #0 {
|
|
%r0 = load double, ptr addrspace(1) %in1, align 8
|
|
%r1 = load double, ptr addrspace(1) %in2, align 8
|
|
%r2 = frem double %r0, %r1
|
|
store double %r2, ptr addrspace(1) %out, align 8
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @fast_frem_f64(ptr addrspace(1) %out, ptr addrspace(1) %in1,
|
|
; SI-LABEL: fast_frem_f64:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_load_dwordx4 s[8:11], s[4:5], 0x9
|
|
; SI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0xd
|
|
; SI-NEXT: s_mov_b32 s3, 0xf000
|
|
; SI-NEXT: s_mov_b32 s2, -1
|
|
; SI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SI-NEXT: s_mov_b32 s0, s8
|
|
; SI-NEXT: s_mov_b32 s1, s9
|
|
; SI-NEXT: s_mov_b32 s8, s10
|
|
; SI-NEXT: s_mov_b32 s9, s11
|
|
; SI-NEXT: s_mov_b32 s10, s2
|
|
; SI-NEXT: s_mov_b32 s11, s3
|
|
; SI-NEXT: s_mov_b32 s6, s2
|
|
; SI-NEXT: s_mov_b32 s7, s3
|
|
; SI-NEXT: buffer_load_dwordx2 v[0:1], off, s[8:11], 0
|
|
; SI-NEXT: buffer_load_dwordx2 v[2:3], off, s[4:7], 0
|
|
; SI-NEXT: s_waitcnt vmcnt(0)
|
|
; SI-NEXT: v_rcp_f64_e32 v[4:5], v[2:3]
|
|
; SI-NEXT: v_fma_f64 v[6:7], -v[2:3], v[4:5], 1.0
|
|
; SI-NEXT: v_fma_f64 v[4:5], v[6:7], v[4:5], v[4:5]
|
|
; SI-NEXT: v_fma_f64 v[6:7], -v[2:3], v[4:5], 1.0
|
|
; SI-NEXT: v_fma_f64 v[4:5], v[6:7], v[4:5], v[4:5]
|
|
; SI-NEXT: v_mul_f64 v[6:7], v[0:1], v[4:5]
|
|
; SI-NEXT: v_fma_f64 v[8:9], -v[2:3], v[6:7], v[0:1]
|
|
; SI-NEXT: v_fma_f64 v[4:5], v[8:9], v[4:5], v[6:7]
|
|
; SI-NEXT: v_readfirstlane_b32 s6, v5
|
|
; SI-NEXT: s_bfe_u32 s4, s6, 0xb0014
|
|
; SI-NEXT: s_add_i32 s7, s4, 0xfffffc01
|
|
; SI-NEXT: s_mov_b32 s5, 0xfffff
|
|
; SI-NEXT: s_mov_b32 s4, s2
|
|
; SI-NEXT: s_lshr_b64 s[4:5], s[4:5], s7
|
|
; SI-NEXT: v_not_b32_e32 v6, s4
|
|
; SI-NEXT: v_and_b32_e32 v6, v4, v6
|
|
; SI-NEXT: v_not_b32_e32 v7, s5
|
|
; SI-NEXT: v_and_b32_e32 v5, v5, v7
|
|
; SI-NEXT: s_and_b32 s4, s6, 0x80000000
|
|
; SI-NEXT: s_cmp_lt_i32 s7, 0
|
|
; SI-NEXT: s_cselect_b64 vcc, -1, 0
|
|
; SI-NEXT: v_cndmask_b32_e64 v6, v6, 0, vcc
|
|
; SI-NEXT: v_mov_b32_e32 v7, s4
|
|
; SI-NEXT: v_cndmask_b32_e32 v5, v5, v7, vcc
|
|
; SI-NEXT: s_cmp_gt_i32 s7, 51
|
|
; SI-NEXT: s_cselect_b64 vcc, -1, 0
|
|
; SI-NEXT: v_mov_b32_e32 v7, s6
|
|
; SI-NEXT: v_cndmask_b32_e32 v5, v5, v7, vcc
|
|
; SI-NEXT: v_cndmask_b32_e32 v4, v6, v4, vcc
|
|
; SI-NEXT: v_fma_f64 v[0:1], -v[4:5], v[2:3], v[0:1]
|
|
; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0
|
|
; SI-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: fast_frem_f64:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9
|
|
; CI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0xd
|
|
; CI-NEXT: s_mov_b32 s11, 0xf000
|
|
; CI-NEXT: s_mov_b32 s10, -1
|
|
; CI-NEXT: s_mov_b32 s6, s10
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_mov_b32 s8, s0
|
|
; CI-NEXT: s_mov_b32 s9, s1
|
|
; CI-NEXT: s_mov_b32 s0, s2
|
|
; CI-NEXT: s_mov_b32 s1, s3
|
|
; CI-NEXT: s_mov_b32 s2, s10
|
|
; CI-NEXT: s_mov_b32 s3, s11
|
|
; CI-NEXT: s_mov_b32 s7, s11
|
|
; CI-NEXT: buffer_load_dwordx2 v[0:1], off, s[0:3], 0
|
|
; CI-NEXT: buffer_load_dwordx2 v[2:3], off, s[4:7], 0
|
|
; CI-NEXT: s_waitcnt vmcnt(0)
|
|
; CI-NEXT: v_rcp_f64_e32 v[4:5], v[2:3]
|
|
; CI-NEXT: v_fma_f64 v[6:7], -v[2:3], v[4:5], 1.0
|
|
; CI-NEXT: v_fma_f64 v[4:5], v[6:7], v[4:5], v[4:5]
|
|
; CI-NEXT: v_fma_f64 v[6:7], -v[2:3], v[4:5], 1.0
|
|
; CI-NEXT: v_fma_f64 v[4:5], v[6:7], v[4:5], v[4:5]
|
|
; CI-NEXT: v_mul_f64 v[6:7], v[0:1], v[4:5]
|
|
; CI-NEXT: v_fma_f64 v[8:9], -v[2:3], v[6:7], v[0:1]
|
|
; CI-NEXT: v_fma_f64 v[4:5], v[8:9], v[4:5], v[6:7]
|
|
; CI-NEXT: v_trunc_f64_e32 v[4:5], v[4:5]
|
|
; CI-NEXT: v_fma_f64 v[0:1], -v[4:5], v[2:3], v[0:1]
|
|
; CI-NEXT: buffer_store_dwordx2 v[0:1], off, s[8:11], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; VI-LABEL: fast_frem_f64:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; VI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x34
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v2, s2
|
|
; VI-NEXT: v_mov_b32_e32 v3, s3
|
|
; VI-NEXT: v_mov_b32_e32 v4, s4
|
|
; VI-NEXT: v_mov_b32_e32 v5, s5
|
|
; VI-NEXT: flat_load_dwordx2 v[2:3], v[2:3]
|
|
; VI-NEXT: flat_load_dwordx2 v[4:5], v[4:5]
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: s_waitcnt vmcnt(0)
|
|
; VI-NEXT: v_rcp_f64_e32 v[6:7], v[4:5]
|
|
; VI-NEXT: v_fma_f64 v[8:9], -v[4:5], v[6:7], 1.0
|
|
; VI-NEXT: v_fma_f64 v[6:7], v[8:9], v[6:7], v[6:7]
|
|
; VI-NEXT: v_fma_f64 v[8:9], -v[4:5], v[6:7], 1.0
|
|
; VI-NEXT: v_fma_f64 v[6:7], v[8:9], v[6:7], v[6:7]
|
|
; VI-NEXT: v_mul_f64 v[8:9], v[2:3], v[6:7]
|
|
; VI-NEXT: v_fma_f64 v[10:11], -v[4:5], v[8:9], v[2:3]
|
|
; VI-NEXT: v_fma_f64 v[6:7], v[10:11], v[6:7], v[8:9]
|
|
; VI-NEXT: v_trunc_f64_e32 v[6:7], v[6:7]
|
|
; VI-NEXT: v_fma_f64 v[2:3], -v[6:7], v[4:5], v[2:3]
|
|
; VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3]
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: fast_frem_f64:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; GFX9-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
|
|
; GFX9-NEXT: v_mov_b32_e32 v10, 0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_load_dwordx2 v[0:1], v10, s[2:3]
|
|
; GFX9-NEXT: global_load_dwordx2 v[2:3], v10, s[6:7]
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX9-NEXT: v_rcp_f64_e32 v[4:5], v[2:3]
|
|
; GFX9-NEXT: v_fma_f64 v[6:7], -v[2:3], v[4:5], 1.0
|
|
; GFX9-NEXT: v_fma_f64 v[4:5], v[6:7], v[4:5], v[4:5]
|
|
; GFX9-NEXT: v_fma_f64 v[6:7], -v[2:3], v[4:5], 1.0
|
|
; GFX9-NEXT: v_fma_f64 v[4:5], v[6:7], v[4:5], v[4:5]
|
|
; GFX9-NEXT: v_mul_f64 v[6:7], v[0:1], v[4:5]
|
|
; GFX9-NEXT: v_fma_f64 v[8:9], -v[2:3], v[6:7], v[0:1]
|
|
; GFX9-NEXT: v_fma_f64 v[4:5], v[8:9], v[4:5], v[6:7]
|
|
; GFX9-NEXT: v_trunc_f64_e32 v[4:5], v[4:5]
|
|
; GFX9-NEXT: v_fma_f64 v[0:1], -v[4:5], v[2:3], v[0:1]
|
|
; GFX9-NEXT: global_store_dwordx2 v10, v[0:1], s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; GFX10-LABEL: fast_frem_f64:
|
|
; GFX10: ; %bb.0:
|
|
; GFX10-NEXT: s_clause 0x1
|
|
; GFX10-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; GFX10-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
|
|
; GFX10-NEXT: v_mov_b32_e32 v10, 0
|
|
; GFX10-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-NEXT: s_clause 0x1
|
|
; GFX10-NEXT: global_load_dwordx2 v[0:1], v10, s[2:3]
|
|
; GFX10-NEXT: global_load_dwordx2 v[2:3], v10, s[6:7]
|
|
; GFX10-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX10-NEXT: v_rcp_f64_e32 v[4:5], v[2:3]
|
|
; GFX10-NEXT: v_fma_f64 v[6:7], -v[2:3], v[4:5], 1.0
|
|
; GFX10-NEXT: v_fma_f64 v[4:5], v[6:7], v[4:5], v[4:5]
|
|
; GFX10-NEXT: v_fma_f64 v[6:7], -v[2:3], v[4:5], 1.0
|
|
; GFX10-NEXT: v_fma_f64 v[4:5], v[6:7], v[4:5], v[4:5]
|
|
; GFX10-NEXT: v_mul_f64 v[6:7], v[0:1], v[4:5]
|
|
; GFX10-NEXT: v_fma_f64 v[8:9], -v[2:3], v[6:7], v[0:1]
|
|
; GFX10-NEXT: v_fma_f64 v[4:5], v[8:9], v[4:5], v[6:7]
|
|
; GFX10-NEXT: v_trunc_f64_e32 v[4:5], v[4:5]
|
|
; GFX10-NEXT: v_fma_f64 v[0:1], -v[4:5], v[2:3], v[0:1]
|
|
; GFX10-NEXT: global_store_dwordx2 v10, v[0:1], s[0:1]
|
|
; GFX10-NEXT: s_endpgm
|
|
;
|
|
; GFX11-LABEL: fast_frem_f64:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_clause 0x1
|
|
; GFX11-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX11-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX11-NEXT: v_mov_b32_e32 v10, 0
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: s_clause 0x1
|
|
; GFX11-NEXT: global_load_b64 v[0:1], v10, s[2:3]
|
|
; GFX11-NEXT: global_load_b64 v[2:3], v10, s[4:5]
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-NEXT: v_rcp_f64_e32 v[4:5], v[2:3]
|
|
; GFX11-NEXT: s_waitcnt_depctr 0xfff
|
|
; GFX11-NEXT: v_fma_f64 v[6:7], -v[2:3], v[4:5], 1.0
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_fma_f64 v[4:5], v[6:7], v[4:5], v[4:5]
|
|
; GFX11-NEXT: v_fma_f64 v[6:7], -v[2:3], v[4:5], 1.0
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_fma_f64 v[4:5], v[6:7], v[4:5], v[4:5]
|
|
; GFX11-NEXT: v_mul_f64 v[6:7], v[0:1], v[4:5]
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_fma_f64 v[8:9], -v[2:3], v[6:7], v[0:1]
|
|
; GFX11-NEXT: v_fma_f64 v[4:5], v[8:9], v[4:5], v[6:7]
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_trunc_f64_e32 v[4:5], v[4:5]
|
|
; GFX11-NEXT: v_fma_f64 v[0:1], -v[4:5], v[2:3], v[0:1]
|
|
; GFX11-NEXT: global_store_b64 v10, v[0:1], s[0:1]
|
|
; GFX11-NEXT: s_endpgm
|
|
;
|
|
; GFX1150-LABEL: fast_frem_f64:
|
|
; GFX1150: ; %bb.0:
|
|
; GFX1150-NEXT: s_clause 0x1
|
|
; GFX1150-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX1150-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX1150-NEXT: v_mov_b32_e32 v10, 0
|
|
; GFX1150-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX1150-NEXT: s_clause 0x1
|
|
; GFX1150-NEXT: global_load_b64 v[0:1], v10, s[2:3]
|
|
; GFX1150-NEXT: global_load_b64 v[2:3], v10, s[4:5]
|
|
; GFX1150-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX1150-NEXT: v_rcp_f64_e32 v[4:5], v[2:3]
|
|
; GFX1150-NEXT: s_delay_alu instid0(TRANS32_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_f64 v[6:7], -v[2:3], v[4:5], 1.0
|
|
; GFX1150-NEXT: v_fma_f64 v[4:5], v[6:7], v[4:5], v[4:5]
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_f64 v[6:7], -v[2:3], v[4:5], 1.0
|
|
; GFX1150-NEXT: v_fma_f64 v[4:5], v[6:7], v[4:5], v[4:5]
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_mul_f64 v[6:7], v[0:1], v[4:5]
|
|
; GFX1150-NEXT: v_fma_f64 v[8:9], -v[2:3], v[6:7], v[0:1]
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_f64 v[4:5], v[8:9], v[4:5], v[6:7]
|
|
; GFX1150-NEXT: v_trunc_f64_e32 v[4:5], v[4:5]
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_f64 v[0:1], -v[4:5], v[2:3], v[0:1]
|
|
; GFX1150-NEXT: global_store_b64 v10, v[0:1], s[0:1]
|
|
; GFX1150-NEXT: s_endpgm
|
|
ptr addrspace(1) %in2) #0 {
|
|
%r0 = load double, ptr addrspace(1) %in1, align 8
|
|
%r1 = load double, ptr addrspace(1) %in2, align 8
|
|
%r2 = frem fast double %r0, %r1
|
|
store double %r2, ptr addrspace(1) %out, align 8
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @unsafe_frem_f64(ptr addrspace(1) %out, ptr addrspace(1) %in1,
|
|
; SI-LABEL: unsafe_frem_f64:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_load_dwordx4 s[8:11], s[4:5], 0x9
|
|
; SI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0xd
|
|
; SI-NEXT: s_mov_b32 s3, 0xf000
|
|
; SI-NEXT: s_mov_b32 s2, -1
|
|
; SI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SI-NEXT: s_mov_b32 s0, s8
|
|
; SI-NEXT: s_mov_b32 s1, s9
|
|
; SI-NEXT: s_mov_b32 s8, s10
|
|
; SI-NEXT: s_mov_b32 s9, s11
|
|
; SI-NEXT: s_mov_b32 s10, s2
|
|
; SI-NEXT: s_mov_b32 s11, s3
|
|
; SI-NEXT: s_mov_b32 s6, s2
|
|
; SI-NEXT: s_mov_b32 s7, s3
|
|
; SI-NEXT: buffer_load_dwordx2 v[0:1], off, s[8:11], 0
|
|
; SI-NEXT: buffer_load_dwordx2 v[2:3], off, s[4:7], 0
|
|
; SI-NEXT: s_waitcnt vmcnt(0)
|
|
; SI-NEXT: v_rcp_f64_e32 v[4:5], v[2:3]
|
|
; SI-NEXT: v_fma_f64 v[6:7], -v[2:3], v[4:5], 1.0
|
|
; SI-NEXT: v_fma_f64 v[4:5], v[6:7], v[4:5], v[4:5]
|
|
; SI-NEXT: v_fma_f64 v[6:7], -v[2:3], v[4:5], 1.0
|
|
; SI-NEXT: v_fma_f64 v[4:5], v[6:7], v[4:5], v[4:5]
|
|
; SI-NEXT: v_mul_f64 v[6:7], v[0:1], v[4:5]
|
|
; SI-NEXT: v_fma_f64 v[8:9], -v[2:3], v[6:7], v[0:1]
|
|
; SI-NEXT: v_fma_f64 v[4:5], v[8:9], v[4:5], v[6:7]
|
|
; SI-NEXT: v_readfirstlane_b32 s6, v5
|
|
; SI-NEXT: s_bfe_u32 s4, s6, 0xb0014
|
|
; SI-NEXT: s_add_i32 s7, s4, 0xfffffc01
|
|
; SI-NEXT: s_mov_b32 s5, 0xfffff
|
|
; SI-NEXT: s_mov_b32 s4, s2
|
|
; SI-NEXT: s_lshr_b64 s[4:5], s[4:5], s7
|
|
; SI-NEXT: v_not_b32_e32 v6, s4
|
|
; SI-NEXT: v_and_b32_e32 v6, v4, v6
|
|
; SI-NEXT: v_not_b32_e32 v7, s5
|
|
; SI-NEXT: v_and_b32_e32 v5, v5, v7
|
|
; SI-NEXT: s_and_b32 s4, s6, 0x80000000
|
|
; SI-NEXT: s_cmp_lt_i32 s7, 0
|
|
; SI-NEXT: s_cselect_b64 vcc, -1, 0
|
|
; SI-NEXT: v_cndmask_b32_e64 v6, v6, 0, vcc
|
|
; SI-NEXT: v_mov_b32_e32 v7, s4
|
|
; SI-NEXT: v_cndmask_b32_e32 v5, v5, v7, vcc
|
|
; SI-NEXT: s_cmp_gt_i32 s7, 51
|
|
; SI-NEXT: s_cselect_b64 vcc, -1, 0
|
|
; SI-NEXT: v_mov_b32_e32 v7, s6
|
|
; SI-NEXT: v_cndmask_b32_e32 v5, v5, v7, vcc
|
|
; SI-NEXT: v_cndmask_b32_e32 v4, v6, v4, vcc
|
|
; SI-NEXT: v_fma_f64 v[0:1], -v[4:5], v[2:3], v[0:1]
|
|
; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0
|
|
; SI-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: unsafe_frem_f64:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9
|
|
; CI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0xd
|
|
; CI-NEXT: s_mov_b32 s11, 0xf000
|
|
; CI-NEXT: s_mov_b32 s10, -1
|
|
; CI-NEXT: s_mov_b32 s6, s10
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_mov_b32 s8, s0
|
|
; CI-NEXT: s_mov_b32 s9, s1
|
|
; CI-NEXT: s_mov_b32 s0, s2
|
|
; CI-NEXT: s_mov_b32 s1, s3
|
|
; CI-NEXT: s_mov_b32 s2, s10
|
|
; CI-NEXT: s_mov_b32 s3, s11
|
|
; CI-NEXT: s_mov_b32 s7, s11
|
|
; CI-NEXT: buffer_load_dwordx2 v[0:1], off, s[0:3], 0
|
|
; CI-NEXT: buffer_load_dwordx2 v[2:3], off, s[4:7], 0
|
|
; CI-NEXT: s_waitcnt vmcnt(0)
|
|
; CI-NEXT: v_rcp_f64_e32 v[4:5], v[2:3]
|
|
; CI-NEXT: v_fma_f64 v[6:7], -v[2:3], v[4:5], 1.0
|
|
; CI-NEXT: v_fma_f64 v[4:5], v[6:7], v[4:5], v[4:5]
|
|
; CI-NEXT: v_fma_f64 v[6:7], -v[2:3], v[4:5], 1.0
|
|
; CI-NEXT: v_fma_f64 v[4:5], v[6:7], v[4:5], v[4:5]
|
|
; CI-NEXT: v_mul_f64 v[6:7], v[0:1], v[4:5]
|
|
; CI-NEXT: v_fma_f64 v[8:9], -v[2:3], v[6:7], v[0:1]
|
|
; CI-NEXT: v_fma_f64 v[4:5], v[8:9], v[4:5], v[6:7]
|
|
; CI-NEXT: v_trunc_f64_e32 v[4:5], v[4:5]
|
|
; CI-NEXT: v_fma_f64 v[0:1], -v[4:5], v[2:3], v[0:1]
|
|
; CI-NEXT: buffer_store_dwordx2 v[0:1], off, s[8:11], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; VI-LABEL: unsafe_frem_f64:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; VI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x34
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v2, s2
|
|
; VI-NEXT: v_mov_b32_e32 v3, s3
|
|
; VI-NEXT: v_mov_b32_e32 v4, s4
|
|
; VI-NEXT: v_mov_b32_e32 v5, s5
|
|
; VI-NEXT: flat_load_dwordx2 v[2:3], v[2:3]
|
|
; VI-NEXT: flat_load_dwordx2 v[4:5], v[4:5]
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: s_waitcnt vmcnt(0)
|
|
; VI-NEXT: v_rcp_f64_e32 v[6:7], v[4:5]
|
|
; VI-NEXT: v_fma_f64 v[8:9], -v[4:5], v[6:7], 1.0
|
|
; VI-NEXT: v_fma_f64 v[6:7], v[8:9], v[6:7], v[6:7]
|
|
; VI-NEXT: v_fma_f64 v[8:9], -v[4:5], v[6:7], 1.0
|
|
; VI-NEXT: v_fma_f64 v[6:7], v[8:9], v[6:7], v[6:7]
|
|
; VI-NEXT: v_mul_f64 v[8:9], v[2:3], v[6:7]
|
|
; VI-NEXT: v_fma_f64 v[10:11], -v[4:5], v[8:9], v[2:3]
|
|
; VI-NEXT: v_fma_f64 v[6:7], v[10:11], v[6:7], v[8:9]
|
|
; VI-NEXT: v_trunc_f64_e32 v[6:7], v[6:7]
|
|
; VI-NEXT: v_fma_f64 v[2:3], -v[6:7], v[4:5], v[2:3]
|
|
; VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3]
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: unsafe_frem_f64:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; GFX9-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
|
|
; GFX9-NEXT: v_mov_b32_e32 v10, 0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_load_dwordx2 v[0:1], v10, s[2:3]
|
|
; GFX9-NEXT: global_load_dwordx2 v[2:3], v10, s[6:7]
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX9-NEXT: v_rcp_f64_e32 v[4:5], v[2:3]
|
|
; GFX9-NEXT: v_fma_f64 v[6:7], -v[2:3], v[4:5], 1.0
|
|
; GFX9-NEXT: v_fma_f64 v[4:5], v[6:7], v[4:5], v[4:5]
|
|
; GFX9-NEXT: v_fma_f64 v[6:7], -v[2:3], v[4:5], 1.0
|
|
; GFX9-NEXT: v_fma_f64 v[4:5], v[6:7], v[4:5], v[4:5]
|
|
; GFX9-NEXT: v_mul_f64 v[6:7], v[0:1], v[4:5]
|
|
; GFX9-NEXT: v_fma_f64 v[8:9], -v[2:3], v[6:7], v[0:1]
|
|
; GFX9-NEXT: v_fma_f64 v[4:5], v[8:9], v[4:5], v[6:7]
|
|
; GFX9-NEXT: v_trunc_f64_e32 v[4:5], v[4:5]
|
|
; GFX9-NEXT: v_fma_f64 v[0:1], -v[4:5], v[2:3], v[0:1]
|
|
; GFX9-NEXT: global_store_dwordx2 v10, v[0:1], s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; GFX10-LABEL: unsafe_frem_f64:
|
|
; GFX10: ; %bb.0:
|
|
; GFX10-NEXT: s_clause 0x1
|
|
; GFX10-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; GFX10-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
|
|
; GFX10-NEXT: v_mov_b32_e32 v10, 0
|
|
; GFX10-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-NEXT: s_clause 0x1
|
|
; GFX10-NEXT: global_load_dwordx2 v[0:1], v10, s[2:3]
|
|
; GFX10-NEXT: global_load_dwordx2 v[2:3], v10, s[6:7]
|
|
; GFX10-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX10-NEXT: v_rcp_f64_e32 v[4:5], v[2:3]
|
|
; GFX10-NEXT: v_fma_f64 v[6:7], -v[2:3], v[4:5], 1.0
|
|
; GFX10-NEXT: v_fma_f64 v[4:5], v[6:7], v[4:5], v[4:5]
|
|
; GFX10-NEXT: v_fma_f64 v[6:7], -v[2:3], v[4:5], 1.0
|
|
; GFX10-NEXT: v_fma_f64 v[4:5], v[6:7], v[4:5], v[4:5]
|
|
; GFX10-NEXT: v_mul_f64 v[6:7], v[0:1], v[4:5]
|
|
; GFX10-NEXT: v_fma_f64 v[8:9], -v[2:3], v[6:7], v[0:1]
|
|
; GFX10-NEXT: v_fma_f64 v[4:5], v[8:9], v[4:5], v[6:7]
|
|
; GFX10-NEXT: v_trunc_f64_e32 v[4:5], v[4:5]
|
|
; GFX10-NEXT: v_fma_f64 v[0:1], -v[4:5], v[2:3], v[0:1]
|
|
; GFX10-NEXT: global_store_dwordx2 v10, v[0:1], s[0:1]
|
|
; GFX10-NEXT: s_endpgm
|
|
;
|
|
; GFX11-LABEL: unsafe_frem_f64:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_clause 0x1
|
|
; GFX11-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX11-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX11-NEXT: v_mov_b32_e32 v10, 0
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: s_clause 0x1
|
|
; GFX11-NEXT: global_load_b64 v[0:1], v10, s[2:3]
|
|
; GFX11-NEXT: global_load_b64 v[2:3], v10, s[4:5]
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-NEXT: v_rcp_f64_e32 v[4:5], v[2:3]
|
|
; GFX11-NEXT: s_waitcnt_depctr 0xfff
|
|
; GFX11-NEXT: v_fma_f64 v[6:7], -v[2:3], v[4:5], 1.0
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_fma_f64 v[4:5], v[6:7], v[4:5], v[4:5]
|
|
; GFX11-NEXT: v_fma_f64 v[6:7], -v[2:3], v[4:5], 1.0
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_fma_f64 v[4:5], v[6:7], v[4:5], v[4:5]
|
|
; GFX11-NEXT: v_mul_f64 v[6:7], v[0:1], v[4:5]
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_fma_f64 v[8:9], -v[2:3], v[6:7], v[0:1]
|
|
; GFX11-NEXT: v_fma_f64 v[4:5], v[8:9], v[4:5], v[6:7]
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_trunc_f64_e32 v[4:5], v[4:5]
|
|
; GFX11-NEXT: v_fma_f64 v[0:1], -v[4:5], v[2:3], v[0:1]
|
|
; GFX11-NEXT: global_store_b64 v10, v[0:1], s[0:1]
|
|
; GFX11-NEXT: s_endpgm
|
|
;
|
|
; GFX1150-LABEL: unsafe_frem_f64:
|
|
; GFX1150: ; %bb.0:
|
|
; GFX1150-NEXT: s_clause 0x1
|
|
; GFX1150-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX1150-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX1150-NEXT: v_mov_b32_e32 v10, 0
|
|
; GFX1150-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX1150-NEXT: s_clause 0x1
|
|
; GFX1150-NEXT: global_load_b64 v[0:1], v10, s[2:3]
|
|
; GFX1150-NEXT: global_load_b64 v[2:3], v10, s[4:5]
|
|
; GFX1150-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX1150-NEXT: v_rcp_f64_e32 v[4:5], v[2:3]
|
|
; GFX1150-NEXT: s_delay_alu instid0(TRANS32_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_f64 v[6:7], -v[2:3], v[4:5], 1.0
|
|
; GFX1150-NEXT: v_fma_f64 v[4:5], v[6:7], v[4:5], v[4:5]
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_f64 v[6:7], -v[2:3], v[4:5], 1.0
|
|
; GFX1150-NEXT: v_fma_f64 v[4:5], v[6:7], v[4:5], v[4:5]
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_mul_f64 v[6:7], v[0:1], v[4:5]
|
|
; GFX1150-NEXT: v_fma_f64 v[8:9], -v[2:3], v[6:7], v[0:1]
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_f64 v[4:5], v[8:9], v[4:5], v[6:7]
|
|
; GFX1150-NEXT: v_trunc_f64_e32 v[4:5], v[4:5]
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_f64 v[0:1], -v[4:5], v[2:3], v[0:1]
|
|
; GFX1150-NEXT: global_store_b64 v10, v[0:1], s[0:1]
|
|
; GFX1150-NEXT: s_endpgm
|
|
ptr addrspace(1) %in2) #1 {
|
|
%r0 = load double, ptr addrspace(1) %in1, align 8
|
|
%r1 = load double, ptr addrspace(1) %in2, align 8
|
|
%r2 = frem afn double %r0, %r1
|
|
store double %r2, ptr addrspace(1) %out, align 8
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @frem_v2f16(ptr addrspace(1) %out, ptr addrspace(1) %in1,
|
|
; SI-LABEL: frem_v2f16:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_load_dwordx4 s[8:11], s[4:5], 0x9
|
|
; SI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0xd
|
|
; SI-NEXT: s_mov_b32 s3, 0xf000
|
|
; SI-NEXT: s_mov_b32 s2, -1
|
|
; SI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SI-NEXT: s_mov_b32 s0, s8
|
|
; SI-NEXT: s_mov_b32 s1, s9
|
|
; SI-NEXT: s_mov_b32 s8, s10
|
|
; SI-NEXT: s_mov_b32 s9, s11
|
|
; SI-NEXT: s_mov_b32 s10, s2
|
|
; SI-NEXT: s_mov_b32 s11, s3
|
|
; SI-NEXT: s_mov_b32 s6, s2
|
|
; SI-NEXT: s_mov_b32 s7, s3
|
|
; SI-NEXT: buffer_load_dword v0, off, s[8:11], 0
|
|
; SI-NEXT: s_waitcnt vmcnt(0)
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v1, v0
|
|
; SI-NEXT: v_lshrrev_b32_e32 v0, 16, v0
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; SI-NEXT: buffer_load_dword v2, off, s[4:7], 0 offset:16
|
|
; SI-NEXT: s_waitcnt vmcnt(0)
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v3, v2
|
|
; SI-NEXT: v_lshrrev_b32_e32 v2, 16, v2
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v2, v2
|
|
; SI-NEXT: v_div_scale_f32 v4, vcc, v0, v2, v0
|
|
; SI-NEXT: v_div_scale_f32 v5, s[4:5], v2, v2, v0
|
|
; SI-NEXT: v_rcp_f32_e32 v6, v5
|
|
; SI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; SI-NEXT: v_fma_f32 v7, -v5, v6, 1.0
|
|
; SI-NEXT: v_fma_f32 v6, v7, v6, v6
|
|
; SI-NEXT: v_mul_f32_e32 v7, v4, v6
|
|
; SI-NEXT: v_fma_f32 v8, -v5, v7, v4
|
|
; SI-NEXT: v_fma_f32 v7, v8, v6, v7
|
|
; SI-NEXT: v_fma_f32 v4, -v5, v7, v4
|
|
; SI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; SI-NEXT: v_div_fmas_f32 v4, v4, v6, v7
|
|
; SI-NEXT: v_div_fixup_f32 v4, v4, v2, v0
|
|
; SI-NEXT: v_trunc_f32_e32 v4, v4
|
|
; SI-NEXT: v_fma_f32 v0, -v4, v2, v0
|
|
; SI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 0
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; SI-NEXT: v_lshlrev_b32_e32 v0, 16, v0
|
|
; SI-NEXT: v_div_scale_f32 v2, vcc, v1, v3, v1
|
|
; SI-NEXT: v_div_scale_f32 v4, s[4:5], v3, v3, v1
|
|
; SI-NEXT: v_rcp_f32_e32 v5, v4
|
|
; SI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; SI-NEXT: v_fma_f32 v6, -v4, v5, 1.0
|
|
; SI-NEXT: v_fma_f32 v5, v6, v5, v5
|
|
; SI-NEXT: v_mul_f32_e32 v6, v2, v5
|
|
; SI-NEXT: v_fma_f32 v7, -v4, v6, v2
|
|
; SI-NEXT: v_fma_f32 v6, v7, v5, v6
|
|
; SI-NEXT: v_fma_f32 v2, -v4, v6, v2
|
|
; SI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; SI-NEXT: v_div_fmas_f32 v2, v2, v5, v6
|
|
; SI-NEXT: v_div_fixup_f32 v2, v2, v3, v1
|
|
; SI-NEXT: v_trunc_f32_e32 v2, v2
|
|
; SI-NEXT: v_fma_f32 v1, -v2, v3, v1
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v1, v1
|
|
; SI-NEXT: v_or_b32_e32 v0, v1, v0
|
|
; SI-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; SI-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: frem_v2f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx4 s[8:11], s[4:5], 0x9
|
|
; CI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0xd
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: s_mov_b32 s6, s2
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_mov_b32 s0, s8
|
|
; CI-NEXT: s_mov_b32 s1, s9
|
|
; CI-NEXT: s_mov_b32 s8, s10
|
|
; CI-NEXT: s_mov_b32 s9, s11
|
|
; CI-NEXT: s_mov_b32 s10, s2
|
|
; CI-NEXT: s_mov_b32 s11, s3
|
|
; CI-NEXT: s_mov_b32 s7, s3
|
|
; CI-NEXT: buffer_load_dword v0, off, s[8:11], 0
|
|
; CI-NEXT: buffer_load_dword v2, off, s[4:7], 0 offset:16
|
|
; CI-NEXT: s_waitcnt vmcnt(1)
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v1, v0
|
|
; CI-NEXT: v_lshrrev_b32_e32 v0, 16, v0
|
|
; CI-NEXT: s_waitcnt vmcnt(0)
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v3, v2
|
|
; CI-NEXT: v_lshrrev_b32_e32 v2, 16, v2
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v2, v2
|
|
; CI-NEXT: v_div_scale_f32 v5, s[4:5], v2, v2, v0
|
|
; CI-NEXT: v_div_scale_f32 v4, vcc, v0, v2, v0
|
|
; CI-NEXT: v_rcp_f32_e32 v6, v5
|
|
; CI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; CI-NEXT: v_fma_f32 v7, -v5, v6, 1.0
|
|
; CI-NEXT: v_fma_f32 v6, v7, v6, v6
|
|
; CI-NEXT: v_mul_f32_e32 v7, v4, v6
|
|
; CI-NEXT: v_fma_f32 v8, -v5, v7, v4
|
|
; CI-NEXT: v_fma_f32 v7, v8, v6, v7
|
|
; CI-NEXT: v_fma_f32 v4, -v5, v7, v4
|
|
; CI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; CI-NEXT: v_div_fmas_f32 v4, v4, v6, v7
|
|
; CI-NEXT: v_div_fixup_f32 v4, v4, v2, v0
|
|
; CI-NEXT: v_trunc_f32_e32 v4, v4
|
|
; CI-NEXT: v_fma_f32 v0, -v4, v2, v0
|
|
; CI-NEXT: v_div_scale_f32 v4, s[4:5], v3, v3, v1
|
|
; CI-NEXT: v_div_scale_f32 v2, vcc, v1, v3, v1
|
|
; CI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: v_lshlrev_b32_e32 v0, 16, v0
|
|
; CI-NEXT: v_rcp_f32_e32 v5, v4
|
|
; CI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; CI-NEXT: v_fma_f32 v6, -v4, v5, 1.0
|
|
; CI-NEXT: v_fma_f32 v5, v6, v5, v5
|
|
; CI-NEXT: v_mul_f32_e32 v6, v2, v5
|
|
; CI-NEXT: v_fma_f32 v7, -v4, v6, v2
|
|
; CI-NEXT: v_fma_f32 v6, v7, v5, v6
|
|
; CI-NEXT: v_fma_f32 v2, -v4, v6, v2
|
|
; CI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; CI-NEXT: v_div_fmas_f32 v2, v2, v5, v6
|
|
; CI-NEXT: v_div_fixup_f32 v2, v2, v3, v1
|
|
; CI-NEXT: v_trunc_f32_e32 v2, v2
|
|
; CI-NEXT: v_fma_f32 v1, -v2, v3, v1
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
|
|
; CI-NEXT: v_or_b32_e32 v0, v1, v0
|
|
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; VI-LABEL: frem_v2f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; VI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x34
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: s_add_u32 s0, s4, 16
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: v_mov_b32_e32 v2, s2
|
|
; VI-NEXT: v_mov_b32_e32 v3, s3
|
|
; VI-NEXT: s_addc_u32 s1, s5, 0
|
|
; VI-NEXT: flat_load_dword v4, v[2:3]
|
|
; VI-NEXT: v_mov_b32_e32 v3, s1
|
|
; VI-NEXT: v_mov_b32_e32 v2, s0
|
|
; VI-NEXT: flat_load_dword v2, v[2:3]
|
|
; VI-NEXT: s_waitcnt vmcnt(1)
|
|
; VI-NEXT: v_lshrrev_b32_e32 v3, 16, v4
|
|
; VI-NEXT: v_cvt_f32_f16_e32 v5, v3
|
|
; VI-NEXT: s_waitcnt vmcnt(0)
|
|
; VI-NEXT: v_lshrrev_b32_e32 v6, 16, v2
|
|
; VI-NEXT: v_cvt_f32_f16_e32 v7, v6
|
|
; VI-NEXT: v_rcp_f32_e32 v8, v7
|
|
; VI-NEXT: v_mul_f32_e32 v9, v5, v8
|
|
; VI-NEXT: v_mad_f32 v10, -v7, v9, v5
|
|
; VI-NEXT: v_mac_f32_e32 v9, v10, v8
|
|
; VI-NEXT: v_mad_f32 v5, -v7, v9, v5
|
|
; VI-NEXT: v_mul_f32_e32 v5, v5, v8
|
|
; VI-NEXT: v_and_b32_e32 v5, 0xff800000, v5
|
|
; VI-NEXT: v_add_f32_e32 v5, v5, v9
|
|
; VI-NEXT: v_cvt_f16_f32_e32 v5, v5
|
|
; VI-NEXT: v_div_fixup_f16 v5, v5, v6, v3
|
|
; VI-NEXT: v_trunc_f16_e32 v5, v5
|
|
; VI-NEXT: v_fma_f16 v3, -v5, v6, v3
|
|
; VI-NEXT: v_cvt_f32_f16_e32 v6, v2
|
|
; VI-NEXT: v_cvt_f32_f16_e32 v5, v4
|
|
; VI-NEXT: v_lshlrev_b32_e32 v3, 16, v3
|
|
; VI-NEXT: v_rcp_f32_e32 v7, v6
|
|
; VI-NEXT: v_mul_f32_e32 v8, v5, v7
|
|
; VI-NEXT: v_mad_f32 v9, -v6, v8, v5
|
|
; VI-NEXT: v_mac_f32_e32 v8, v9, v7
|
|
; VI-NEXT: v_mad_f32 v5, -v6, v8, v5
|
|
; VI-NEXT: v_mul_f32_e32 v5, v5, v7
|
|
; VI-NEXT: v_and_b32_e32 v5, 0xff800000, v5
|
|
; VI-NEXT: v_add_f32_e32 v5, v5, v8
|
|
; VI-NEXT: v_cvt_f16_f32_e32 v5, v5
|
|
; VI-NEXT: v_div_fixup_f16 v5, v5, v2, v4
|
|
; VI-NEXT: v_trunc_f16_e32 v5, v5
|
|
; VI-NEXT: v_fma_f16 v2, -v5, v2, v4
|
|
; VI-NEXT: v_or_b32_e32 v2, v2, v3
|
|
; VI-NEXT: flat_store_dword v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: frem_v2f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; GFX9-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_load_dword v1, v0, s[2:3]
|
|
; GFX9-NEXT: global_load_dword v2, v0, s[6:7] offset:16
|
|
; GFX9-NEXT: s_waitcnt vmcnt(1)
|
|
; GFX9-NEXT: v_cvt_f32_f16_e32 v3, v1
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX9-NEXT: v_cvt_f32_f16_e32 v4, v2
|
|
; GFX9-NEXT: v_lshrrev_b32_e32 v6, 16, v2
|
|
; GFX9-NEXT: v_cvt_f32_f16_e32 v7, v6
|
|
; GFX9-NEXT: v_rcp_f32_e32 v4, v4
|
|
; GFX9-NEXT: v_rcp_f32_e32 v7, v7
|
|
; GFX9-NEXT: v_mul_f32_e32 v3, v3, v4
|
|
; GFX9-NEXT: v_mad_mix_f32 v5, -v2, v3, v1 op_sel_hi:[1,0,1]
|
|
; GFX9-NEXT: v_mac_f32_e32 v3, v5, v4
|
|
; GFX9-NEXT: v_mad_mix_f32 v5, -v2, v3, v1 op_sel_hi:[1,0,1]
|
|
; GFX9-NEXT: v_mul_f32_e32 v4, v5, v4
|
|
; GFX9-NEXT: v_and_b32_e32 v4, 0xff800000, v4
|
|
; GFX9-NEXT: v_add_f32_e32 v3, v4, v3
|
|
; GFX9-NEXT: v_lshrrev_b32_e32 v4, 16, v1
|
|
; GFX9-NEXT: v_cvt_f32_f16_e32 v5, v4
|
|
; GFX9-NEXT: v_cvt_f16_f32_e32 v3, v3
|
|
; GFX9-NEXT: v_mul_f32_e32 v5, v5, v7
|
|
; GFX9-NEXT: v_div_fixup_f16 v3, v3, v2, v1
|
|
; GFX9-NEXT: v_mad_mix_f32 v8, -v2, v5, v1 op_sel:[1,0,1] op_sel_hi:[1,0,1]
|
|
; GFX9-NEXT: v_trunc_f16_e32 v3, v3
|
|
; GFX9-NEXT: v_mac_f32_e32 v5, v8, v7
|
|
; GFX9-NEXT: v_fma_f16 v3, -v3, v2, v1
|
|
; GFX9-NEXT: v_mad_mix_f32 v1, -v2, v5, v1 op_sel:[1,0,1] op_sel_hi:[1,0,1]
|
|
; GFX9-NEXT: v_mul_f32_e32 v1, v1, v7
|
|
; GFX9-NEXT: v_and_b32_e32 v1, 0xff800000, v1
|
|
; GFX9-NEXT: v_add_f32_e32 v1, v1, v5
|
|
; GFX9-NEXT: v_cvt_f16_f32_e32 v1, v1
|
|
; GFX9-NEXT: v_div_fixup_f16 v1, v1, v6, v4
|
|
; GFX9-NEXT: v_trunc_f16_e32 v1, v1
|
|
; GFX9-NEXT: v_fma_f16 v1, -v1, v6, v4
|
|
; GFX9-NEXT: v_pack_b32_f16 v1, v3, v1
|
|
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; GFX10-LABEL: frem_v2f16:
|
|
; GFX10: ; %bb.0:
|
|
; GFX10-NEXT: s_clause 0x1
|
|
; GFX10-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; GFX10-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
|
|
; GFX10-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX10-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-NEXT: s_clause 0x1
|
|
; GFX10-NEXT: global_load_dword v1, v0, s[2:3]
|
|
; GFX10-NEXT: global_load_dword v2, v0, s[6:7] offset:16
|
|
; GFX10-NEXT: s_waitcnt vmcnt(1)
|
|
; GFX10-NEXT: v_cvt_f32_f16_e32 v3, v1
|
|
; GFX10-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX10-NEXT: v_cvt_f32_f16_e32 v4, v2
|
|
; GFX10-NEXT: v_rcp_f32_e32 v5, v4
|
|
; GFX10-NEXT: v_mul_f32_e32 v6, v3, v5
|
|
; GFX10-NEXT: v_mad_f32 v7, -v4, v6, v3
|
|
; GFX10-NEXT: v_mac_f32_e32 v6, v7, v5
|
|
; GFX10-NEXT: v_mad_f32 v3, -v4, v6, v3
|
|
; GFX10-NEXT: v_mul_f32_e32 v3, v3, v5
|
|
; GFX10-NEXT: v_and_b32_e32 v3, 0xff800000, v3
|
|
; GFX10-NEXT: v_add_f32_e32 v3, v3, v6
|
|
; GFX10-NEXT: v_cvt_f16_f32_e32 v3, v3
|
|
; GFX10-NEXT: v_div_fixup_f16 v3, v3, v2, v1
|
|
; GFX10-NEXT: v_trunc_f16_e32 v3, v3
|
|
; GFX10-NEXT: v_fma_f16 v3, -v3, v2, v1
|
|
; GFX10-NEXT: v_lshrrev_b32_e32 v2, 16, v2
|
|
; GFX10-NEXT: v_lshrrev_b32_e32 v1, 16, v1
|
|
; GFX10-NEXT: v_cvt_f32_f16_e32 v5, v2
|
|
; GFX10-NEXT: v_cvt_f32_f16_e32 v4, v1
|
|
; GFX10-NEXT: v_rcp_f32_e32 v6, v5
|
|
; GFX10-NEXT: v_mul_f32_e32 v7, v4, v6
|
|
; GFX10-NEXT: v_mad_f32 v8, -v5, v7, v4
|
|
; GFX10-NEXT: v_mac_f32_e32 v7, v8, v6
|
|
; GFX10-NEXT: v_mad_f32 v4, -v5, v7, v4
|
|
; GFX10-NEXT: v_mul_f32_e32 v4, v4, v6
|
|
; GFX10-NEXT: v_and_b32_e32 v4, 0xff800000, v4
|
|
; GFX10-NEXT: v_add_f32_e32 v4, v4, v7
|
|
; GFX10-NEXT: v_cvt_f16_f32_e32 v4, v4
|
|
; GFX10-NEXT: v_div_fixup_f16 v4, v4, v2, v1
|
|
; GFX10-NEXT: v_trunc_f16_e32 v4, v4
|
|
; GFX10-NEXT: v_fma_f16 v1, -v4, v2, v1
|
|
; GFX10-NEXT: v_pack_b32_f16 v1, v3, v1
|
|
; GFX10-NEXT: global_store_dword v0, v1, s[0:1]
|
|
; GFX10-NEXT: s_endpgm
|
|
;
|
|
; GFX11-TRUE16-LABEL: frem_v2f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_clause 0x1
|
|
; GFX11-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX11-TRUE16-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: s_clause 0x1
|
|
; GFX11-TRUE16-NEXT: global_load_b32 v2, v1, s[2:3]
|
|
; GFX11-TRUE16-NEXT: global_load_b32 v3, v1, s[4:5] offset:16
|
|
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(1)
|
|
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v0, v2.l
|
|
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v4, v3.l
|
|
; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v3.l
|
|
; GFX11-TRUE16-NEXT: v_mov_b16_e32 v6.l, v2.l
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_2) | instid1(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_rcp_f32_e32 v4, v4
|
|
; GFX11-TRUE16-NEXT: s_waitcnt_depctr 0xfff
|
|
; GFX11-TRUE16-NEXT: v_mul_f32_e32 v0, v0, v4
|
|
; GFX11-TRUE16-NEXT: v_fma_mix_f32 v7, -v5, v0, v6 op_sel_hi:[1,0,1]
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_fmac_f32_e32 v0, v7, v4
|
|
; GFX11-TRUE16-NEXT: v_fma_mix_f32 v5, -v5, v0, v6 op_sel_hi:[1,0,1]
|
|
; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v6, 16, v3
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v7, v6.l
|
|
; GFX11-TRUE16-NEXT: v_rcp_f32_e32 v7, v7
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_mul_f32_e32 v4, v5, v4
|
|
; GFX11-TRUE16-NEXT: v_and_b32_e32 v4, 0xff800000, v4
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
|
|
; GFX11-TRUE16-NEXT: v_add_f32_e32 v0, v4, v0
|
|
; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2
|
|
; GFX11-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v0
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
|
|
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v5, v4.l
|
|
; GFX11-TRUE16-NEXT: v_div_fixup_f16 v0.l, v0.l, v3.l, v2.l
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(TRANS32_DEP_1) | instid1(VALU_DEP_2)
|
|
; GFX11-TRUE16-NEXT: v_mul_f32_e32 v5, v5, v7
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
|
|
; GFX11-TRUE16-NEXT: v_trunc_f16_e32 v0.l, v0.l
|
|
; GFX11-TRUE16-NEXT: v_fma_mix_f32 v8, -v3, v5, v2 op_sel:[1,0,1] op_sel_hi:[1,0,1]
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
|
|
; GFX11-TRUE16-NEXT: v_fma_f16 v0.l, -v0.l, v3.l, v2.l
|
|
; GFX11-TRUE16-NEXT: v_fmac_f32_e32 v5, v8, v7
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_fma_mix_f32 v2, -v3, v5, v2 op_sel:[1,0,1] op_sel_hi:[1,0,1]
|
|
; GFX11-TRUE16-NEXT: v_mul_f32_e32 v2, v2, v7
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_and_b32_e32 v2, 0xff800000, v2
|
|
; GFX11-TRUE16-NEXT: v_add_f32_e32 v2, v2, v5
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.h, v2
|
|
; GFX11-TRUE16-NEXT: v_div_fixup_f16 v0.h, v0.h, v6.l, v4.l
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_trunc_f16_e32 v0.h, v0.h
|
|
; GFX11-TRUE16-NEXT: v_fma_f16 v0.h, -v0.h, v6.l, v4.l
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_pack_b32_f16 v0, v0.l, v0.h
|
|
; GFX11-TRUE16-NEXT: global_store_b32 v1, v0, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_endpgm
|
|
;
|
|
; GFX11-FAKE16-LABEL: frem_v2f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_clause 0x1
|
|
; GFX11-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX11-FAKE16-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX11-FAKE16-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: s_clause 0x1
|
|
; GFX11-FAKE16-NEXT: global_load_b32 v1, v0, s[2:3]
|
|
; GFX11-FAKE16-NEXT: global_load_b32 v2, v0, s[4:5] offset:16
|
|
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(1)
|
|
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v3, v1
|
|
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v4, v2
|
|
; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v2
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_rcp_f32_e32 v4, v4
|
|
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v7, v6
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_rcp_f32_e32 v7, v7
|
|
; GFX11-FAKE16-NEXT: s_waitcnt_depctr 0xfff
|
|
; GFX11-FAKE16-NEXT: v_mul_f32_e32 v3, v3, v4
|
|
; GFX11-FAKE16-NEXT: v_fma_mix_f32 v5, -v2, v3, v1 op_sel_hi:[1,0,1]
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_fmac_f32_e32 v3, v5, v4
|
|
; GFX11-FAKE16-NEXT: v_fma_mix_f32 v5, -v2, v3, v1 op_sel_hi:[1,0,1]
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_mul_f32_e32 v4, v5, v4
|
|
; GFX11-FAKE16-NEXT: v_and_b32_e32 v4, 0xff800000, v4
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
|
|
; GFX11-FAKE16-NEXT: v_add_f32_e32 v3, v4, v3
|
|
; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v1
|
|
; GFX11-FAKE16-NEXT: v_cvt_f16_f32_e32 v3, v3
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
|
|
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v5, v4
|
|
; GFX11-FAKE16-NEXT: v_div_fixup_f16 v3, v3, v2, v1
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
|
|
; GFX11-FAKE16-NEXT: v_mul_f32_e32 v5, v5, v7
|
|
; GFX11-FAKE16-NEXT: v_trunc_f16_e32 v3, v3
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
|
|
; GFX11-FAKE16-NEXT: v_fma_mix_f32 v8, -v2, v5, v1 op_sel:[1,0,1] op_sel_hi:[1,0,1]
|
|
; GFX11-FAKE16-NEXT: v_fma_f16 v3, -v3, v2, v1
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_fmac_f32_e32 v5, v8, v7
|
|
; GFX11-FAKE16-NEXT: v_fma_mix_f32 v1, -v2, v5, v1 op_sel:[1,0,1] op_sel_hi:[1,0,1]
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_mul_f32_e32 v1, v1, v7
|
|
; GFX11-FAKE16-NEXT: v_and_b32_e32 v1, 0xff800000, v1
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_add_f32_e32 v1, v1, v5
|
|
; GFX11-FAKE16-NEXT: v_cvt_f16_f32_e32 v1, v1
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_div_fixup_f16 v1, v1, v6, v4
|
|
; GFX11-FAKE16-NEXT: v_trunc_f16_e32 v1, v1
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_fma_f16 v1, -v1, v6, v4
|
|
; GFX11-FAKE16-NEXT: v_pack_b32_f16 v1, v3, v1
|
|
; GFX11-FAKE16-NEXT: global_store_b32 v0, v1, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_endpgm
|
|
;
|
|
; GFX1150-LABEL: frem_v2f16:
|
|
; GFX1150: ; %bb.0:
|
|
; GFX1150-NEXT: s_clause 0x1
|
|
; GFX1150-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX1150-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX1150-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX1150-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX1150-NEXT: s_clause 0x1
|
|
; GFX1150-NEXT: global_load_b32 v1, v0, s[2:3]
|
|
; GFX1150-NEXT: global_load_b32 v2, v0, s[4:5] offset:16
|
|
; GFX1150-NEXT: s_waitcnt vmcnt(1)
|
|
; GFX1150-NEXT: v_lshrrev_b32_e32 v3, 16, v1
|
|
; GFX1150-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX1150-NEXT: v_lshrrev_b32_e32 v5, 16, v2
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
|
|
; GFX1150-NEXT: v_cvt_f32_f16_e32 v4, v3
|
|
; GFX1150-NEXT: v_cvt_f32_f16_e32 v6, v5
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(TRANS32_DEP_1)
|
|
; GFX1150-NEXT: v_rcp_f32_e32 v6, v6
|
|
; GFX1150-NEXT: v_mul_f32_e32 v4, v4, v6
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_mix_f32 v7, -v2, v4, v1 op_sel:[1,0,1] op_sel_hi:[1,0,1]
|
|
; GFX1150-NEXT: v_fmac_f32_e32 v4, v7, v6
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_mix_f32 v7, -v2, v4, v1 op_sel:[1,0,1] op_sel_hi:[1,0,1]
|
|
; GFX1150-NEXT: v_mul_f32_e32 v6, v7, v6
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_and_b32_e32 v6, 0xff800000, v6
|
|
; GFX1150-NEXT: v_add_f32_e32 v4, v6, v4
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_cvt_f16_f32_e32 v4, v4
|
|
; GFX1150-NEXT: v_div_fixup_f16 v4, v4, v5, v3
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_trunc_f16_e32 v4, v4
|
|
; GFX1150-NEXT: v_xor_b32_e32 v4, 0x8000, v4
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
|
|
; GFX1150-NEXT: v_fmac_f16_e32 v3, v4, v5
|
|
; GFX1150-NEXT: v_cvt_f32_f16_e32 v5, v2
|
|
; GFX1150-NEXT: v_cvt_f32_f16_e32 v4, v1
|
|
; GFX1150-NEXT: v_rcp_f32_e32 v5, v5
|
|
; GFX1150-NEXT: s_delay_alu instid0(TRANS32_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_mul_f32_e32 v4, v4, v5
|
|
; GFX1150-NEXT: v_fma_mix_f32 v6, -v2, v4, v1 op_sel_hi:[1,0,1]
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fmac_f32_e32 v4, v6, v5
|
|
; GFX1150-NEXT: v_fma_mix_f32 v6, -v2, v4, v1 op_sel_hi:[1,0,1]
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_mul_f32_e32 v5, v6, v5
|
|
; GFX1150-NEXT: v_and_b32_e32 v5, 0xff800000, v5
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_add_f32_e32 v4, v5, v4
|
|
; GFX1150-NEXT: v_cvt_f16_f32_e32 v4, v4
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_div_fixup_f16 v4, v4, v2, v1
|
|
; GFX1150-NEXT: v_trunc_f16_e32 v4, v4
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_xor_b32_e32 v4, 0x8000, v4
|
|
; GFX1150-NEXT: v_fmac_f16_e32 v1, v4, v2
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_pack_b32_f16 v1, v1, v3
|
|
; GFX1150-NEXT: global_store_b32 v0, v1, s[0:1]
|
|
; GFX1150-NEXT: s_endpgm
|
|
ptr addrspace(1) %in2) #0 {
|
|
%gep2 = getelementptr <2 x half>, ptr addrspace(1) %in2, i32 4
|
|
%r0 = load <2 x half>, ptr addrspace(1) %in1, align 8
|
|
%r1 = load <2 x half>, ptr addrspace(1) %gep2, align 8
|
|
%r2 = frem <2 x half> %r0, %r1
|
|
store <2 x half> %r2, ptr addrspace(1) %out, align 8
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @frem_v4f16(ptr addrspace(1) %out, ptr addrspace(1) %in1,
|
|
; SI-LABEL: frem_v4f16:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_load_dwordx4 s[8:11], s[4:5], 0x9
|
|
; SI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0xd
|
|
; SI-NEXT: s_mov_b32 s3, 0xf000
|
|
; SI-NEXT: s_mov_b32 s2, -1
|
|
; SI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SI-NEXT: s_mov_b32 s0, s8
|
|
; SI-NEXT: s_mov_b32 s1, s9
|
|
; SI-NEXT: s_mov_b32 s8, s10
|
|
; SI-NEXT: s_mov_b32 s9, s11
|
|
; SI-NEXT: s_mov_b32 s10, s2
|
|
; SI-NEXT: s_mov_b32 s11, s3
|
|
; SI-NEXT: s_mov_b32 s6, s2
|
|
; SI-NEXT: s_mov_b32 s7, s3
|
|
; SI-NEXT: buffer_load_dwordx2 v[0:1], off, s[8:11], 0
|
|
; SI-NEXT: s_waitcnt vmcnt(0)
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v2, v0
|
|
; SI-NEXT: v_lshrrev_b32_e32 v0, 16, v0
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v3, v0
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v4, v1
|
|
; SI-NEXT: v_lshrrev_b32_e32 v0, 16, v1
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v5, v0
|
|
; SI-NEXT: buffer_load_dwordx2 v[0:1], off, s[4:7], 0 offset:32
|
|
; SI-NEXT: s_waitcnt vmcnt(0)
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v6, v0
|
|
; SI-NEXT: v_lshrrev_b32_e32 v0, 16, v0
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v7, v1
|
|
; SI-NEXT: v_lshrrev_b32_e32 v1, 16, v1
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v1, v1
|
|
; SI-NEXT: v_div_scale_f32 v8, vcc, v5, v1, v5
|
|
; SI-NEXT: v_div_scale_f32 v9, s[4:5], v1, v1, v5
|
|
; SI-NEXT: v_rcp_f32_e32 v10, v9
|
|
; SI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; SI-NEXT: v_fma_f32 v11, -v9, v10, 1.0
|
|
; SI-NEXT: v_fma_f32 v10, v11, v10, v10
|
|
; SI-NEXT: v_mul_f32_e32 v11, v8, v10
|
|
; SI-NEXT: v_fma_f32 v12, -v9, v11, v8
|
|
; SI-NEXT: v_fma_f32 v11, v12, v10, v11
|
|
; SI-NEXT: v_fma_f32 v8, -v9, v11, v8
|
|
; SI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; SI-NEXT: v_div_fmas_f32 v8, v8, v10, v11
|
|
; SI-NEXT: v_div_fixup_f32 v8, v8, v1, v5
|
|
; SI-NEXT: v_trunc_f32_e32 v8, v8
|
|
; SI-NEXT: v_fma_f32 v1, -v8, v1, v5
|
|
; SI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 0
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v1, v1
|
|
; SI-NEXT: v_lshlrev_b32_e32 v1, 16, v1
|
|
; SI-NEXT: v_div_scale_f32 v5, vcc, v4, v7, v4
|
|
; SI-NEXT: v_div_scale_f32 v8, s[4:5], v7, v7, v4
|
|
; SI-NEXT: v_rcp_f32_e32 v9, v8
|
|
; SI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; SI-NEXT: v_fma_f32 v10, -v8, v9, 1.0
|
|
; SI-NEXT: v_fma_f32 v9, v10, v9, v9
|
|
; SI-NEXT: v_mul_f32_e32 v10, v5, v9
|
|
; SI-NEXT: v_fma_f32 v11, -v8, v10, v5
|
|
; SI-NEXT: v_fma_f32 v10, v11, v9, v10
|
|
; SI-NEXT: v_fma_f32 v5, -v8, v10, v5
|
|
; SI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; SI-NEXT: v_div_fmas_f32 v5, v5, v9, v10
|
|
; SI-NEXT: v_div_fixup_f32 v5, v5, v7, v4
|
|
; SI-NEXT: v_trunc_f32_e32 v5, v5
|
|
; SI-NEXT: v_fma_f32 v4, -v5, v7, v4
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v4, v4
|
|
; SI-NEXT: v_or_b32_e32 v1, v4, v1
|
|
; SI-NEXT: v_div_scale_f32 v4, vcc, v3, v0, v3
|
|
; SI-NEXT: v_div_scale_f32 v5, s[4:5], v0, v0, v3
|
|
; SI-NEXT: v_rcp_f32_e32 v7, v5
|
|
; SI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; SI-NEXT: v_fma_f32 v8, -v5, v7, 1.0
|
|
; SI-NEXT: v_fma_f32 v7, v8, v7, v7
|
|
; SI-NEXT: v_mul_f32_e32 v8, v4, v7
|
|
; SI-NEXT: v_fma_f32 v9, -v5, v8, v4
|
|
; SI-NEXT: v_fma_f32 v8, v9, v7, v8
|
|
; SI-NEXT: v_fma_f32 v4, -v5, v8, v4
|
|
; SI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; SI-NEXT: v_div_fmas_f32 v4, v4, v7, v8
|
|
; SI-NEXT: v_div_fixup_f32 v4, v4, v0, v3
|
|
; SI-NEXT: v_trunc_f32_e32 v4, v4
|
|
; SI-NEXT: v_fma_f32 v0, -v4, v0, v3
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; SI-NEXT: v_lshlrev_b32_e32 v0, 16, v0
|
|
; SI-NEXT: v_div_scale_f32 v3, vcc, v2, v6, v2
|
|
; SI-NEXT: v_div_scale_f32 v4, s[4:5], v6, v6, v2
|
|
; SI-NEXT: v_rcp_f32_e32 v5, v4
|
|
; SI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; SI-NEXT: v_fma_f32 v7, -v4, v5, 1.0
|
|
; SI-NEXT: v_fma_f32 v5, v7, v5, v5
|
|
; SI-NEXT: v_mul_f32_e32 v7, v3, v5
|
|
; SI-NEXT: v_fma_f32 v8, -v4, v7, v3
|
|
; SI-NEXT: v_fma_f32 v7, v8, v5, v7
|
|
; SI-NEXT: v_fma_f32 v3, -v4, v7, v3
|
|
; SI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; SI-NEXT: v_div_fmas_f32 v3, v3, v5, v7
|
|
; SI-NEXT: v_div_fixup_f32 v3, v3, v6, v2
|
|
; SI-NEXT: v_trunc_f32_e32 v3, v3
|
|
; SI-NEXT: v_fma_f32 v2, -v3, v6, v2
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v2, v2
|
|
; SI-NEXT: v_or_b32_e32 v0, v2, v0
|
|
; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0
|
|
; SI-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: frem_v4f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx4 s[8:11], s[4:5], 0x9
|
|
; CI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0xd
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: s_mov_b32 s6, s2
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_mov_b32 s0, s8
|
|
; CI-NEXT: s_mov_b32 s1, s9
|
|
; CI-NEXT: s_mov_b32 s8, s10
|
|
; CI-NEXT: s_mov_b32 s9, s11
|
|
; CI-NEXT: s_mov_b32 s10, s2
|
|
; CI-NEXT: s_mov_b32 s11, s3
|
|
; CI-NEXT: buffer_load_dwordx2 v[0:1], off, s[8:11], 0
|
|
; CI-NEXT: s_mov_b32 s7, s3
|
|
; CI-NEXT: s_waitcnt vmcnt(0)
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v2, v0
|
|
; CI-NEXT: v_lshrrev_b32_e32 v0, 16, v0
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v3, v0
|
|
; CI-NEXT: v_lshrrev_b32_e32 v0, 16, v1
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v4, v1
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v5, v0
|
|
; CI-NEXT: buffer_load_dwordx2 v[0:1], off, s[4:7], 0 offset:32
|
|
; CI-NEXT: s_waitcnt vmcnt(0)
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v7, v1
|
|
; CI-NEXT: v_lshrrev_b32_e32 v1, 16, v1
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v6, v0
|
|
; CI-NEXT: v_lshrrev_b32_e32 v0, 16, v0
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; CI-NEXT: v_div_scale_f32 v9, s[4:5], v1, v1, v5
|
|
; CI-NEXT: v_div_scale_f32 v8, vcc, v5, v1, v5
|
|
; CI-NEXT: v_rcp_f32_e32 v10, v9
|
|
; CI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; CI-NEXT: v_fma_f32 v11, -v9, v10, 1.0
|
|
; CI-NEXT: v_fma_f32 v10, v11, v10, v10
|
|
; CI-NEXT: v_mul_f32_e32 v11, v8, v10
|
|
; CI-NEXT: v_fma_f32 v12, -v9, v11, v8
|
|
; CI-NEXT: v_fma_f32 v11, v12, v10, v11
|
|
; CI-NEXT: v_fma_f32 v8, -v9, v11, v8
|
|
; CI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; CI-NEXT: v_div_fmas_f32 v8, v8, v10, v11
|
|
; CI-NEXT: v_div_fixup_f32 v8, v8, v1, v5
|
|
; CI-NEXT: v_trunc_f32_e32 v8, v8
|
|
; CI-NEXT: v_fma_f32 v1, -v8, v1, v5
|
|
; CI-NEXT: v_div_scale_f32 v8, s[4:5], v7, v7, v4
|
|
; CI-NEXT: v_div_scale_f32 v5, vcc, v4, v7, v4
|
|
; CI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
|
|
; CI-NEXT: v_lshlrev_b32_e32 v1, 16, v1
|
|
; CI-NEXT: v_rcp_f32_e32 v9, v8
|
|
; CI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; CI-NEXT: v_fma_f32 v10, -v8, v9, 1.0
|
|
; CI-NEXT: v_fma_f32 v9, v10, v9, v9
|
|
; CI-NEXT: v_mul_f32_e32 v10, v5, v9
|
|
; CI-NEXT: v_fma_f32 v11, -v8, v10, v5
|
|
; CI-NEXT: v_fma_f32 v10, v11, v9, v10
|
|
; CI-NEXT: v_fma_f32 v5, -v8, v10, v5
|
|
; CI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; CI-NEXT: v_div_fmas_f32 v5, v5, v9, v10
|
|
; CI-NEXT: v_div_fixup_f32 v5, v5, v7, v4
|
|
; CI-NEXT: v_trunc_f32_e32 v5, v5
|
|
; CI-NEXT: v_fma_f32 v4, -v5, v7, v4
|
|
; CI-NEXT: v_div_scale_f32 v5, s[4:5], v0, v0, v3
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v4, v4
|
|
; CI-NEXT: v_or_b32_e32 v1, v4, v1
|
|
; CI-NEXT: v_div_scale_f32 v4, vcc, v3, v0, v3
|
|
; CI-NEXT: v_rcp_f32_e32 v7, v5
|
|
; CI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; CI-NEXT: v_fma_f32 v8, -v5, v7, 1.0
|
|
; CI-NEXT: v_fma_f32 v7, v8, v7, v7
|
|
; CI-NEXT: v_mul_f32_e32 v8, v4, v7
|
|
; CI-NEXT: v_fma_f32 v9, -v5, v8, v4
|
|
; CI-NEXT: v_fma_f32 v8, v9, v7, v8
|
|
; CI-NEXT: v_fma_f32 v4, -v5, v8, v4
|
|
; CI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; CI-NEXT: v_div_fmas_f32 v4, v4, v7, v8
|
|
; CI-NEXT: v_div_fixup_f32 v4, v4, v0, v3
|
|
; CI-NEXT: v_trunc_f32_e32 v4, v4
|
|
; CI-NEXT: v_fma_f32 v0, -v4, v0, v3
|
|
; CI-NEXT: v_div_scale_f32 v4, s[4:5], v6, v6, v2
|
|
; CI-NEXT: v_div_scale_f32 v3, vcc, v2, v6, v2
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: v_lshlrev_b32_e32 v0, 16, v0
|
|
; CI-NEXT: v_rcp_f32_e32 v5, v4
|
|
; CI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; CI-NEXT: v_fma_f32 v7, -v4, v5, 1.0
|
|
; CI-NEXT: v_fma_f32 v5, v7, v5, v5
|
|
; CI-NEXT: v_mul_f32_e32 v7, v3, v5
|
|
; CI-NEXT: v_fma_f32 v8, -v4, v7, v3
|
|
; CI-NEXT: v_fma_f32 v7, v8, v5, v7
|
|
; CI-NEXT: v_fma_f32 v3, -v4, v7, v3
|
|
; CI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; CI-NEXT: v_div_fmas_f32 v3, v3, v5, v7
|
|
; CI-NEXT: v_div_fixup_f32 v3, v3, v6, v2
|
|
; CI-NEXT: v_trunc_f32_e32 v3, v3
|
|
; CI-NEXT: v_fma_f32 v2, -v3, v6, v2
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v2, v2
|
|
; CI-NEXT: v_or_b32_e32 v0, v2, v0
|
|
; CI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; VI-LABEL: frem_v4f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; VI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x34
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: s_add_u32 s0, s4, 32
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: s_addc_u32 s1, s5, 0
|
|
; VI-NEXT: v_mov_b32_e32 v5, s1
|
|
; VI-NEXT: v_mov_b32_e32 v4, s0
|
|
; VI-NEXT: flat_load_dwordx2 v[4:5], v[4:5]
|
|
; VI-NEXT: v_mov_b32_e32 v2, s2
|
|
; VI-NEXT: v_mov_b32_e32 v3, s3
|
|
; VI-NEXT: flat_load_dwordx2 v[2:3], v[2:3]
|
|
; VI-NEXT: s_waitcnt vmcnt(1)
|
|
; VI-NEXT: v_lshrrev_b32_e32 v8, 16, v5
|
|
; VI-NEXT: v_cvt_f32_f16_e32 v9, v8
|
|
; VI-NEXT: s_waitcnt vmcnt(0)
|
|
; VI-NEXT: v_lshrrev_b32_e32 v6, 16, v3
|
|
; VI-NEXT: v_cvt_f32_f16_e32 v7, v6
|
|
; VI-NEXT: v_rcp_f32_e32 v10, v9
|
|
; VI-NEXT: v_mul_f32_e32 v11, v7, v10
|
|
; VI-NEXT: v_mad_f32 v12, -v9, v11, v7
|
|
; VI-NEXT: v_mac_f32_e32 v11, v12, v10
|
|
; VI-NEXT: v_mad_f32 v7, -v9, v11, v7
|
|
; VI-NEXT: v_mul_f32_e32 v7, v7, v10
|
|
; VI-NEXT: v_and_b32_e32 v7, 0xff800000, v7
|
|
; VI-NEXT: v_add_f32_e32 v7, v7, v11
|
|
; VI-NEXT: v_cvt_f16_f32_e32 v7, v7
|
|
; VI-NEXT: v_div_fixup_f16 v7, v7, v8, v6
|
|
; VI-NEXT: v_trunc_f16_e32 v7, v7
|
|
; VI-NEXT: v_fma_f16 v6, -v7, v8, v6
|
|
; VI-NEXT: v_cvt_f32_f16_e32 v8, v5
|
|
; VI-NEXT: v_cvt_f32_f16_e32 v7, v3
|
|
; VI-NEXT: v_lshlrev_b32_e32 v6, 16, v6
|
|
; VI-NEXT: v_rcp_f32_e32 v9, v8
|
|
; VI-NEXT: v_mul_f32_e32 v10, v7, v9
|
|
; VI-NEXT: v_mad_f32 v11, -v8, v10, v7
|
|
; VI-NEXT: v_mac_f32_e32 v10, v11, v9
|
|
; VI-NEXT: v_mad_f32 v7, -v8, v10, v7
|
|
; VI-NEXT: v_mul_f32_e32 v7, v7, v9
|
|
; VI-NEXT: v_and_b32_e32 v7, 0xff800000, v7
|
|
; VI-NEXT: v_add_f32_e32 v7, v7, v10
|
|
; VI-NEXT: v_cvt_f16_f32_e32 v7, v7
|
|
; VI-NEXT: v_div_fixup_f16 v7, v7, v5, v3
|
|
; VI-NEXT: v_trunc_f16_e32 v7, v7
|
|
; VI-NEXT: v_fma_f16 v3, -v7, v5, v3
|
|
; VI-NEXT: v_lshrrev_b32_e32 v7, 16, v4
|
|
; VI-NEXT: v_cvt_f32_f16_e32 v8, v7
|
|
; VI-NEXT: v_lshrrev_b32_e32 v5, 16, v2
|
|
; VI-NEXT: v_or_b32_e32 v3, v3, v6
|
|
; VI-NEXT: v_cvt_f32_f16_e32 v6, v5
|
|
; VI-NEXT: v_rcp_f32_e32 v9, v8
|
|
; VI-NEXT: v_mul_f32_e32 v10, v6, v9
|
|
; VI-NEXT: v_mad_f32 v11, -v8, v10, v6
|
|
; VI-NEXT: v_mac_f32_e32 v10, v11, v9
|
|
; VI-NEXT: v_mad_f32 v6, -v8, v10, v6
|
|
; VI-NEXT: v_mul_f32_e32 v6, v6, v9
|
|
; VI-NEXT: v_and_b32_e32 v6, 0xff800000, v6
|
|
; VI-NEXT: v_add_f32_e32 v6, v6, v10
|
|
; VI-NEXT: v_cvt_f16_f32_e32 v6, v6
|
|
; VI-NEXT: v_div_fixup_f16 v6, v6, v7, v5
|
|
; VI-NEXT: v_trunc_f16_e32 v6, v6
|
|
; VI-NEXT: v_fma_f16 v5, -v6, v7, v5
|
|
; VI-NEXT: v_cvt_f32_f16_e32 v7, v4
|
|
; VI-NEXT: v_cvt_f32_f16_e32 v6, v2
|
|
; VI-NEXT: v_lshlrev_b32_e32 v5, 16, v5
|
|
; VI-NEXT: v_rcp_f32_e32 v8, v7
|
|
; VI-NEXT: v_mul_f32_e32 v9, v6, v8
|
|
; VI-NEXT: v_mad_f32 v10, -v7, v9, v6
|
|
; VI-NEXT: v_mac_f32_e32 v9, v10, v8
|
|
; VI-NEXT: v_mad_f32 v6, -v7, v9, v6
|
|
; VI-NEXT: v_mul_f32_e32 v6, v6, v8
|
|
; VI-NEXT: v_and_b32_e32 v6, 0xff800000, v6
|
|
; VI-NEXT: v_add_f32_e32 v6, v6, v9
|
|
; VI-NEXT: v_cvt_f16_f32_e32 v6, v6
|
|
; VI-NEXT: v_div_fixup_f16 v6, v6, v4, v2
|
|
; VI-NEXT: v_trunc_f16_e32 v6, v6
|
|
; VI-NEXT: v_fma_f16 v2, -v6, v4, v2
|
|
; VI-NEXT: v_or_b32_e32 v2, v2, v5
|
|
; VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3]
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: frem_v4f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; GFX9-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
|
|
; GFX9-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_load_dwordx2 v[0:1], v4, s[2:3]
|
|
; GFX9-NEXT: global_load_dwordx2 v[2:3], v4, s[6:7] offset:32
|
|
; GFX9-NEXT: s_waitcnt vmcnt(1)
|
|
; GFX9-NEXT: v_cvt_f32_f16_e32 v5, v1
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX9-NEXT: v_cvt_f32_f16_e32 v6, v3
|
|
; GFX9-NEXT: v_lshrrev_b32_e32 v8, 16, v3
|
|
; GFX9-NEXT: v_cvt_f32_f16_e32 v9, v8
|
|
; GFX9-NEXT: v_rcp_f32_e32 v6, v6
|
|
; GFX9-NEXT: v_rcp_f32_e32 v9, v9
|
|
; GFX9-NEXT: v_mul_f32_e32 v5, v5, v6
|
|
; GFX9-NEXT: v_mad_mix_f32 v7, -v3, v5, v1 op_sel_hi:[1,0,1]
|
|
; GFX9-NEXT: v_mac_f32_e32 v5, v7, v6
|
|
; GFX9-NEXT: v_mad_mix_f32 v7, -v3, v5, v1 op_sel_hi:[1,0,1]
|
|
; GFX9-NEXT: v_mul_f32_e32 v6, v7, v6
|
|
; GFX9-NEXT: v_and_b32_e32 v6, 0xff800000, v6
|
|
; GFX9-NEXT: v_add_f32_e32 v5, v6, v5
|
|
; GFX9-NEXT: v_lshrrev_b32_e32 v6, 16, v1
|
|
; GFX9-NEXT: v_cvt_f32_f16_e32 v7, v6
|
|
; GFX9-NEXT: v_cvt_f16_f32_e32 v5, v5
|
|
; GFX9-NEXT: v_mul_f32_e32 v7, v7, v9
|
|
; GFX9-NEXT: v_div_fixup_f16 v5, v5, v3, v1
|
|
; GFX9-NEXT: v_mad_mix_f32 v10, -v3, v7, v1 op_sel:[1,0,1] op_sel_hi:[1,0,1]
|
|
; GFX9-NEXT: v_trunc_f16_e32 v5, v5
|
|
; GFX9-NEXT: v_mac_f32_e32 v7, v10, v9
|
|
; GFX9-NEXT: v_fma_f16 v5, -v5, v3, v1
|
|
; GFX9-NEXT: v_mad_mix_f32 v1, -v3, v7, v1 op_sel:[1,0,1] op_sel_hi:[1,0,1]
|
|
; GFX9-NEXT: v_mul_f32_e32 v1, v1, v9
|
|
; GFX9-NEXT: v_and_b32_e32 v1, 0xff800000, v1
|
|
; GFX9-NEXT: v_add_f32_e32 v1, v1, v7
|
|
; GFX9-NEXT: v_cvt_f16_f32_e32 v1, v1
|
|
; GFX9-NEXT: v_cvt_f32_f16_e32 v3, v0
|
|
; GFX9-NEXT: v_lshrrev_b32_e32 v7, 16, v2
|
|
; GFX9-NEXT: v_div_fixup_f16 v1, v1, v8, v6
|
|
; GFX9-NEXT: v_trunc_f16_e32 v1, v1
|
|
; GFX9-NEXT: v_fma_f16 v1, -v1, v8, v6
|
|
; GFX9-NEXT: v_pack_b32_f16 v1, v5, v1
|
|
; GFX9-NEXT: v_cvt_f32_f16_e32 v5, v2
|
|
; GFX9-NEXT: v_cvt_f32_f16_e32 v8, v7
|
|
; GFX9-NEXT: v_rcp_f32_e32 v5, v5
|
|
; GFX9-NEXT: v_rcp_f32_e32 v8, v8
|
|
; GFX9-NEXT: v_mul_f32_e32 v3, v3, v5
|
|
; GFX9-NEXT: v_mad_mix_f32 v6, -v2, v3, v0 op_sel_hi:[1,0,1]
|
|
; GFX9-NEXT: v_mac_f32_e32 v3, v6, v5
|
|
; GFX9-NEXT: v_mad_mix_f32 v6, -v2, v3, v0 op_sel_hi:[1,0,1]
|
|
; GFX9-NEXT: v_mul_f32_e32 v5, v6, v5
|
|
; GFX9-NEXT: v_and_b32_e32 v5, 0xff800000, v5
|
|
; GFX9-NEXT: v_add_f32_e32 v3, v5, v3
|
|
; GFX9-NEXT: v_lshrrev_b32_e32 v5, 16, v0
|
|
; GFX9-NEXT: v_cvt_f32_f16_e32 v6, v5
|
|
; GFX9-NEXT: v_cvt_f16_f32_e32 v3, v3
|
|
; GFX9-NEXT: v_mul_f32_e32 v6, v6, v8
|
|
; GFX9-NEXT: v_div_fixup_f16 v3, v3, v2, v0
|
|
; GFX9-NEXT: v_mad_mix_f32 v9, -v2, v6, v0 op_sel:[1,0,1] op_sel_hi:[1,0,1]
|
|
; GFX9-NEXT: v_trunc_f16_e32 v3, v3
|
|
; GFX9-NEXT: v_mac_f32_e32 v6, v9, v8
|
|
; GFX9-NEXT: v_fma_f16 v3, -v3, v2, v0
|
|
; GFX9-NEXT: v_mad_mix_f32 v0, -v2, v6, v0 op_sel:[1,0,1] op_sel_hi:[1,0,1]
|
|
; GFX9-NEXT: v_mul_f32_e32 v0, v0, v8
|
|
; GFX9-NEXT: v_and_b32_e32 v0, 0xff800000, v0
|
|
; GFX9-NEXT: v_add_f32_e32 v0, v0, v6
|
|
; GFX9-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; GFX9-NEXT: v_div_fixup_f16 v0, v0, v7, v5
|
|
; GFX9-NEXT: v_trunc_f16_e32 v0, v0
|
|
; GFX9-NEXT: v_fma_f16 v0, -v0, v7, v5
|
|
; GFX9-NEXT: v_pack_b32_f16 v0, v3, v0
|
|
; GFX9-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; GFX10-LABEL: frem_v4f16:
|
|
; GFX10: ; %bb.0:
|
|
; GFX10-NEXT: s_clause 0x1
|
|
; GFX10-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; GFX10-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
|
|
; GFX10-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX10-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-NEXT: s_clause 0x1
|
|
; GFX10-NEXT: global_load_dwordx2 v[0:1], v4, s[2:3]
|
|
; GFX10-NEXT: global_load_dwordx2 v[2:3], v4, s[6:7] offset:32
|
|
; GFX10-NEXT: s_waitcnt vmcnt(1)
|
|
; GFX10-NEXT: v_cvt_f32_f16_e32 v5, v1
|
|
; GFX10-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX10-NEXT: v_cvt_f32_f16_e32 v6, v3
|
|
; GFX10-NEXT: v_rcp_f32_e32 v7, v6
|
|
; GFX10-NEXT: v_mul_f32_e32 v8, v5, v7
|
|
; GFX10-NEXT: v_mad_f32 v9, -v6, v8, v5
|
|
; GFX10-NEXT: v_mac_f32_e32 v8, v9, v7
|
|
; GFX10-NEXT: v_mad_f32 v5, -v6, v8, v5
|
|
; GFX10-NEXT: v_mul_f32_e32 v5, v5, v7
|
|
; GFX10-NEXT: v_and_b32_e32 v5, 0xff800000, v5
|
|
; GFX10-NEXT: v_add_f32_e32 v5, v5, v8
|
|
; GFX10-NEXT: v_cvt_f16_f32_e32 v5, v5
|
|
; GFX10-NEXT: v_div_fixup_f16 v5, v5, v3, v1
|
|
; GFX10-NEXT: v_trunc_f16_e32 v5, v5
|
|
; GFX10-NEXT: v_fma_f16 v5, -v5, v3, v1
|
|
; GFX10-NEXT: v_lshrrev_b32_e32 v3, 16, v3
|
|
; GFX10-NEXT: v_lshrrev_b32_e32 v1, 16, v1
|
|
; GFX10-NEXT: v_cvt_f32_f16_e32 v7, v3
|
|
; GFX10-NEXT: v_cvt_f32_f16_e32 v6, v1
|
|
; GFX10-NEXT: v_rcp_f32_e32 v8, v7
|
|
; GFX10-NEXT: v_mul_f32_e32 v9, v6, v8
|
|
; GFX10-NEXT: v_mad_f32 v10, -v7, v9, v6
|
|
; GFX10-NEXT: v_mac_f32_e32 v9, v10, v8
|
|
; GFX10-NEXT: v_mad_f32 v6, -v7, v9, v6
|
|
; GFX10-NEXT: v_mul_f32_e32 v6, v6, v8
|
|
; GFX10-NEXT: v_and_b32_e32 v6, 0xff800000, v6
|
|
; GFX10-NEXT: v_add_f32_e32 v6, v6, v9
|
|
; GFX10-NEXT: v_cvt_f16_f32_e32 v6, v6
|
|
; GFX10-NEXT: v_div_fixup_f16 v6, v6, v3, v1
|
|
; GFX10-NEXT: v_trunc_f16_e32 v6, v6
|
|
; GFX10-NEXT: v_fma_f16 v1, -v6, v3, v1
|
|
; GFX10-NEXT: v_cvt_f32_f16_e32 v3, v0
|
|
; GFX10-NEXT: v_pack_b32_f16 v1, v5, v1
|
|
; GFX10-NEXT: v_cvt_f32_f16_e32 v5, v2
|
|
; GFX10-NEXT: v_rcp_f32_e32 v6, v5
|
|
; GFX10-NEXT: v_mul_f32_e32 v7, v3, v6
|
|
; GFX10-NEXT: v_mad_f32 v8, -v5, v7, v3
|
|
; GFX10-NEXT: v_mac_f32_e32 v7, v8, v6
|
|
; GFX10-NEXT: v_mad_f32 v3, -v5, v7, v3
|
|
; GFX10-NEXT: v_mul_f32_e32 v3, v3, v6
|
|
; GFX10-NEXT: v_and_b32_e32 v3, 0xff800000, v3
|
|
; GFX10-NEXT: v_add_f32_e32 v3, v3, v7
|
|
; GFX10-NEXT: v_cvt_f16_f32_e32 v3, v3
|
|
; GFX10-NEXT: v_div_fixup_f16 v3, v3, v2, v0
|
|
; GFX10-NEXT: v_trunc_f16_e32 v3, v3
|
|
; GFX10-NEXT: v_fma_f16 v3, -v3, v2, v0
|
|
; GFX10-NEXT: v_lshrrev_b32_e32 v2, 16, v2
|
|
; GFX10-NEXT: v_lshrrev_b32_e32 v0, 16, v0
|
|
; GFX10-NEXT: v_cvt_f32_f16_e32 v6, v2
|
|
; GFX10-NEXT: v_cvt_f32_f16_e32 v5, v0
|
|
; GFX10-NEXT: v_rcp_f32_e32 v7, v6
|
|
; GFX10-NEXT: v_mul_f32_e32 v8, v5, v7
|
|
; GFX10-NEXT: v_mad_f32 v9, -v6, v8, v5
|
|
; GFX10-NEXT: v_mac_f32_e32 v8, v9, v7
|
|
; GFX10-NEXT: v_mad_f32 v5, -v6, v8, v5
|
|
; GFX10-NEXT: v_mul_f32_e32 v5, v5, v7
|
|
; GFX10-NEXT: v_and_b32_e32 v5, 0xff800000, v5
|
|
; GFX10-NEXT: v_add_f32_e32 v5, v5, v8
|
|
; GFX10-NEXT: v_cvt_f16_f32_e32 v5, v5
|
|
; GFX10-NEXT: v_div_fixup_f16 v5, v5, v2, v0
|
|
; GFX10-NEXT: v_trunc_f16_e32 v5, v5
|
|
; GFX10-NEXT: v_fma_f16 v0, -v5, v2, v0
|
|
; GFX10-NEXT: v_pack_b32_f16 v0, v3, v0
|
|
; GFX10-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX10-NEXT: s_endpgm
|
|
;
|
|
; GFX11-TRUE16-LABEL: frem_v4f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_clause 0x1
|
|
; GFX11-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX11-TRUE16-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX11-TRUE16-NEXT: v_mov_b32_e32 v5, 0
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: s_clause 0x1
|
|
; GFX11-TRUE16-NEXT: global_load_b64 v[1:2], v5, s[2:3]
|
|
; GFX11-TRUE16-NEXT: global_load_b64 v[3:4], v5, s[4:5] offset:32
|
|
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(1)
|
|
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v0, v2.l
|
|
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v6, v4.l
|
|
; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v8, 16, v4
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_rcp_f32_e32 v6, v6
|
|
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v9, v8.l
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_rcp_f32_e32 v9, v9
|
|
; GFX11-TRUE16-NEXT: s_waitcnt_depctr 0xfff
|
|
; GFX11-TRUE16-NEXT: v_mul_f32_e32 v0, v0, v6
|
|
; GFX11-TRUE16-NEXT: v_fma_mix_f32 v7, -v4, v0, v2 op_sel_hi:[1,0,1]
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_fmac_f32_e32 v0, v7, v6
|
|
; GFX11-TRUE16-NEXT: v_fma_mix_f32 v7, -v4, v0, v2 op_sel_hi:[1,0,1]
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_mul_f32_e32 v6, v7, v6
|
|
; GFX11-TRUE16-NEXT: v_and_b32_e32 v6, 0xff800000, v6
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
|
|
; GFX11-TRUE16-NEXT: v_add_f32_e32 v0, v6, v0
|
|
; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v6, 16, v2
|
|
; GFX11-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v0
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
|
|
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v7, v6.l
|
|
; GFX11-TRUE16-NEXT: v_div_fixup_f16 v0.l, v0.l, v4.l, v2.l
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
|
|
; GFX11-TRUE16-NEXT: v_mul_f32_e32 v7, v7, v9
|
|
; GFX11-TRUE16-NEXT: v_trunc_f16_e32 v0.l, v0.l
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
|
|
; GFX11-TRUE16-NEXT: v_fma_mix_f32 v10, -v4, v7, v2 op_sel:[1,0,1] op_sel_hi:[1,0,1]
|
|
; GFX11-TRUE16-NEXT: v_fma_f16 v0.l, -v0.l, v4.l, v2.l
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_fmac_f32_e32 v7, v10, v9
|
|
; GFX11-TRUE16-NEXT: v_fma_mix_f32 v2, -v4, v7, v2 op_sel:[1,0,1] op_sel_hi:[1,0,1]
|
|
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v4, v3.l
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
|
|
; GFX11-TRUE16-NEXT: v_mul_f32_e32 v2, v2, v9
|
|
; GFX11-TRUE16-NEXT: v_rcp_f32_e32 v4, v4
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_and_b32_e32 v2, 0xff800000, v2
|
|
; GFX11-TRUE16-NEXT: v_add_f32_e32 v2, v2, v7
|
|
; GFX11-TRUE16-NEXT: v_mov_b16_e32 v7.l, v1.l
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.h, v2
|
|
; GFX11-TRUE16-NEXT: v_div_fixup_f16 v0.h, v0.h, v8.l, v6.l
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_trunc_f16_e32 v0.h, v0.h
|
|
; GFX11-TRUE16-NEXT: v_fma_f16 v0.h, -v0.h, v8.l, v6.l
|
|
; GFX11-TRUE16-NEXT: v_mov_b16_e32 v6.l, v3.l
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_pack_b32_f16 v2, v0.l, v0.h
|
|
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v0, v1.l
|
|
; GFX11-TRUE16-NEXT: v_mul_f32_e32 v0, v0, v4
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_fma_mix_f32 v8, -v6, v0, v7 op_sel_hi:[1,0,1]
|
|
; GFX11-TRUE16-NEXT: v_fmac_f32_e32 v0, v8, v4
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
|
|
; GFX11-TRUE16-NEXT: v_fma_mix_f32 v6, -v6, v0, v7 op_sel_hi:[1,0,1]
|
|
; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v7, 16, v3
|
|
; GFX11-TRUE16-NEXT: v_mul_f32_e32 v4, v6, v4
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
|
|
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v8, v7.l
|
|
; GFX11-TRUE16-NEXT: v_and_b32_e32 v4, 0xff800000, v4
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_rcp_f32_e32 v8, v8
|
|
; GFX11-TRUE16-NEXT: v_add_f32_e32 v0, v4, v0
|
|
; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v4, 16, v1
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
|
|
; GFX11-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v0
|
|
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v6, v4.l
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_3) | instid1(VALU_DEP_2)
|
|
; GFX11-TRUE16-NEXT: v_div_fixup_f16 v0.l, v0.l, v3.l, v1.l
|
|
; GFX11-TRUE16-NEXT: s_waitcnt_depctr 0xfff
|
|
; GFX11-TRUE16-NEXT: v_mul_f32_e32 v6, v6, v8
|
|
; GFX11-TRUE16-NEXT: v_trunc_f16_e32 v0.l, v0.l
|
|
; GFX11-TRUE16-NEXT: v_fma_mix_f32 v9, -v3, v6, v1 op_sel:[1,0,1] op_sel_hi:[1,0,1]
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
|
|
; GFX11-TRUE16-NEXT: v_fma_f16 v0.l, -v0.l, v3.l, v1.l
|
|
; GFX11-TRUE16-NEXT: v_fmac_f32_e32 v6, v9, v8
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_fma_mix_f32 v1, -v3, v6, v1 op_sel:[1,0,1] op_sel_hi:[1,0,1]
|
|
; GFX11-TRUE16-NEXT: v_mul_f32_e32 v1, v1, v8
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_and_b32_e32 v1, 0xff800000, v1
|
|
; GFX11-TRUE16-NEXT: v_add_f32_e32 v1, v1, v6
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.h, v1
|
|
; GFX11-TRUE16-NEXT: v_div_fixup_f16 v0.h, v0.h, v7.l, v4.l
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_trunc_f16_e32 v0.h, v0.h
|
|
; GFX11-TRUE16-NEXT: v_fma_f16 v0.h, -v0.h, v7.l, v4.l
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_pack_b32_f16 v1, v0.l, v0.h
|
|
; GFX11-TRUE16-NEXT: global_store_b64 v5, v[1:2], s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_endpgm
|
|
;
|
|
; GFX11-FAKE16-LABEL: frem_v4f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_clause 0x1
|
|
; GFX11-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX11-FAKE16-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX11-FAKE16-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: s_clause 0x1
|
|
; GFX11-FAKE16-NEXT: global_load_b64 v[0:1], v4, s[2:3]
|
|
; GFX11-FAKE16-NEXT: global_load_b64 v[2:3], v4, s[4:5] offset:32
|
|
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(1)
|
|
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v5, v1
|
|
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v6, v3
|
|
; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v8, 16, v3
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_rcp_f32_e32 v6, v6
|
|
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v9, v8
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_rcp_f32_e32 v9, v9
|
|
; GFX11-FAKE16-NEXT: s_waitcnt_depctr 0xfff
|
|
; GFX11-FAKE16-NEXT: v_mul_f32_e32 v5, v5, v6
|
|
; GFX11-FAKE16-NEXT: v_fma_mix_f32 v7, -v3, v5, v1 op_sel_hi:[1,0,1]
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_fmac_f32_e32 v5, v7, v6
|
|
; GFX11-FAKE16-NEXT: v_fma_mix_f32 v7, -v3, v5, v1 op_sel_hi:[1,0,1]
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_mul_f32_e32 v6, v7, v6
|
|
; GFX11-FAKE16-NEXT: v_and_b32_e32 v6, 0xff800000, v6
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
|
|
; GFX11-FAKE16-NEXT: v_add_f32_e32 v5, v6, v5
|
|
; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v1
|
|
; GFX11-FAKE16-NEXT: v_cvt_f16_f32_e32 v5, v5
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
|
|
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v7, v6
|
|
; GFX11-FAKE16-NEXT: v_div_fixup_f16 v5, v5, v3, v1
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
|
|
; GFX11-FAKE16-NEXT: v_mul_f32_e32 v7, v7, v9
|
|
; GFX11-FAKE16-NEXT: v_trunc_f16_e32 v5, v5
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
|
|
; GFX11-FAKE16-NEXT: v_fma_mix_f32 v10, -v3, v7, v1 op_sel:[1,0,1] op_sel_hi:[1,0,1]
|
|
; GFX11-FAKE16-NEXT: v_fma_f16 v5, -v5, v3, v1
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_fmac_f32_e32 v7, v10, v9
|
|
; GFX11-FAKE16-NEXT: v_fma_mix_f32 v1, -v3, v7, v1 op_sel:[1,0,1] op_sel_hi:[1,0,1]
|
|
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v3, v0
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_mul_f32_e32 v1, v1, v9
|
|
; GFX11-FAKE16-NEXT: v_and_b32_e32 v1, 0xff800000, v1
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
|
|
; GFX11-FAKE16-NEXT: v_add_f32_e32 v1, v1, v7
|
|
; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v7, 16, v2
|
|
; GFX11-FAKE16-NEXT: v_cvt_f16_f32_e32 v1, v1
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_div_fixup_f16 v1, v1, v8, v6
|
|
; GFX11-FAKE16-NEXT: v_trunc_f16_e32 v1, v1
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
|
|
; GFX11-FAKE16-NEXT: v_fma_f16 v1, -v1, v8, v6
|
|
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v8, v7
|
|
; GFX11-FAKE16-NEXT: v_pack_b32_f16 v1, v5, v1
|
|
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v5, v2
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_rcp_f32_e32 v8, v8
|
|
; GFX11-FAKE16-NEXT: v_rcp_f32_e32 v5, v5
|
|
; GFX11-FAKE16-NEXT: s_waitcnt_depctr 0xfff
|
|
; GFX11-FAKE16-NEXT: v_mul_f32_e32 v3, v3, v5
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_fma_mix_f32 v6, -v2, v3, v0 op_sel_hi:[1,0,1]
|
|
; GFX11-FAKE16-NEXT: v_fmac_f32_e32 v3, v6, v5
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_fma_mix_f32 v6, -v2, v3, v0 op_sel_hi:[1,0,1]
|
|
; GFX11-FAKE16-NEXT: v_mul_f32_e32 v5, v6, v5
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_and_b32_e32 v5, 0xff800000, v5
|
|
; GFX11-FAKE16-NEXT: v_add_f32_e32 v3, v5, v3
|
|
; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v0
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
|
|
; GFX11-FAKE16-NEXT: v_cvt_f16_f32_e32 v3, v3
|
|
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v6, v5
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
|
|
; GFX11-FAKE16-NEXT: v_div_fixup_f16 v3, v3, v2, v0
|
|
; GFX11-FAKE16-NEXT: v_mul_f32_e32 v6, v6, v8
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
|
|
; GFX11-FAKE16-NEXT: v_trunc_f16_e32 v3, v3
|
|
; GFX11-FAKE16-NEXT: v_fma_mix_f32 v9, -v2, v6, v0 op_sel:[1,0,1] op_sel_hi:[1,0,1]
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
|
|
; GFX11-FAKE16-NEXT: v_fma_f16 v3, -v3, v2, v0
|
|
; GFX11-FAKE16-NEXT: v_fmac_f32_e32 v6, v9, v8
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_fma_mix_f32 v0, -v2, v6, v0 op_sel:[1,0,1] op_sel_hi:[1,0,1]
|
|
; GFX11-FAKE16-NEXT: v_mul_f32_e32 v0, v0, v8
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_and_b32_e32 v0, 0xff800000, v0
|
|
; GFX11-FAKE16-NEXT: v_add_f32_e32 v0, v0, v6
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; GFX11-FAKE16-NEXT: v_div_fixup_f16 v0, v0, v7, v5
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_trunc_f16_e32 v0, v0
|
|
; GFX11-FAKE16-NEXT: v_fma_f16 v0, -v0, v7, v5
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_pack_b32_f16 v0, v3, v0
|
|
; GFX11-FAKE16-NEXT: global_store_b64 v4, v[0:1], s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_endpgm
|
|
;
|
|
; GFX1150-LABEL: frem_v4f16:
|
|
; GFX1150: ; %bb.0:
|
|
; GFX1150-NEXT: s_clause 0x1
|
|
; GFX1150-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX1150-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX1150-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX1150-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX1150-NEXT: s_clause 0x1
|
|
; GFX1150-NEXT: global_load_b64 v[0:1], v4, s[2:3]
|
|
; GFX1150-NEXT: global_load_b64 v[2:3], v4, s[4:5] offset:32
|
|
; GFX1150-NEXT: s_waitcnt vmcnt(1)
|
|
; GFX1150-NEXT: v_lshrrev_b32_e32 v5, 16, v0
|
|
; GFX1150-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX1150-NEXT: v_lshrrev_b32_e32 v7, 16, v2
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
|
|
; GFX1150-NEXT: v_cvt_f32_f16_e32 v6, v5
|
|
; GFX1150-NEXT: v_cvt_f32_f16_e32 v8, v7
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(TRANS32_DEP_1)
|
|
; GFX1150-NEXT: v_rcp_f32_e32 v8, v8
|
|
; GFX1150-NEXT: v_mul_f32_e32 v6, v6, v8
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_mix_f32 v9, -v2, v6, v0 op_sel:[1,0,1] op_sel_hi:[1,0,1]
|
|
; GFX1150-NEXT: v_fmac_f32_e32 v6, v9, v8
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_mix_f32 v9, -v2, v6, v0 op_sel:[1,0,1] op_sel_hi:[1,0,1]
|
|
; GFX1150-NEXT: v_mul_f32_e32 v8, v9, v8
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_and_b32_e32 v8, 0xff800000, v8
|
|
; GFX1150-NEXT: v_add_f32_e32 v6, v8, v6
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_cvt_f16_f32_e32 v6, v6
|
|
; GFX1150-NEXT: v_div_fixup_f16 v6, v6, v7, v5
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_trunc_f16_e32 v6, v6
|
|
; GFX1150-NEXT: v_xor_b32_e32 v6, 0x8000, v6
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
|
|
; GFX1150-NEXT: v_fmac_f16_e32 v5, v6, v7
|
|
; GFX1150-NEXT: v_cvt_f32_f16_e32 v7, v2
|
|
; GFX1150-NEXT: v_cvt_f32_f16_e32 v6, v0
|
|
; GFX1150-NEXT: v_rcp_f32_e32 v7, v7
|
|
; GFX1150-NEXT: s_delay_alu instid0(TRANS32_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_mul_f32_e32 v6, v6, v7
|
|
; GFX1150-NEXT: v_fma_mix_f32 v8, -v2, v6, v0 op_sel_hi:[1,0,1]
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fmac_f32_e32 v6, v8, v7
|
|
; GFX1150-NEXT: v_fma_mix_f32 v8, -v2, v6, v0 op_sel_hi:[1,0,1]
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_mul_f32_e32 v7, v8, v7
|
|
; GFX1150-NEXT: v_and_b32_e32 v7, 0xff800000, v7
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_add_f32_e32 v6, v7, v6
|
|
; GFX1150-NEXT: v_cvt_f16_f32_e32 v6, v6
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_div_fixup_f16 v6, v6, v2, v0
|
|
; GFX1150-NEXT: v_trunc_f16_e32 v6, v6
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_xor_b32_e32 v6, 0x8000, v6
|
|
; GFX1150-NEXT: v_fma_f16 v0, v6, v2, v0
|
|
; GFX1150-NEXT: v_lshrrev_b32_e32 v6, 16, v3
|
|
; GFX1150-NEXT: v_lshrrev_b32_e32 v2, 16, v1
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
|
|
; GFX1150-NEXT: v_pack_b32_f16 v0, v0, v5
|
|
; GFX1150-NEXT: v_cvt_f32_f16_e32 v7, v6
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
|
|
; GFX1150-NEXT: v_cvt_f32_f16_e32 v5, v2
|
|
; GFX1150-NEXT: v_rcp_f32_e32 v7, v7
|
|
; GFX1150-NEXT: s_delay_alu instid0(TRANS32_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_mul_f32_e32 v5, v5, v7
|
|
; GFX1150-NEXT: v_fma_mix_f32 v8, -v3, v5, v1 op_sel:[1,0,1] op_sel_hi:[1,0,1]
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fmac_f32_e32 v5, v8, v7
|
|
; GFX1150-NEXT: v_fma_mix_f32 v8, -v3, v5, v1 op_sel:[1,0,1] op_sel_hi:[1,0,1]
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_mul_f32_e32 v7, v8, v7
|
|
; GFX1150-NEXT: v_and_b32_e32 v7, 0xff800000, v7
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_add_f32_e32 v5, v7, v5
|
|
; GFX1150-NEXT: v_cvt_f16_f32_e32 v5, v5
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_div_fixup_f16 v5, v5, v6, v2
|
|
; GFX1150-NEXT: v_trunc_f16_e32 v5, v5
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_xor_b32_e32 v5, 0x8000, v5
|
|
; GFX1150-NEXT: v_fmac_f16_e32 v2, v5, v6
|
|
; GFX1150-NEXT: v_cvt_f32_f16_e32 v6, v3
|
|
; GFX1150-NEXT: v_cvt_f32_f16_e32 v5, v1
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(TRANS32_DEP_1)
|
|
; GFX1150-NEXT: v_rcp_f32_e32 v6, v6
|
|
; GFX1150-NEXT: v_mul_f32_e32 v5, v5, v6
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_mix_f32 v7, -v3, v5, v1 op_sel_hi:[1,0,1]
|
|
; GFX1150-NEXT: v_fmac_f32_e32 v5, v7, v6
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_mix_f32 v7, -v3, v5, v1 op_sel_hi:[1,0,1]
|
|
; GFX1150-NEXT: v_mul_f32_e32 v6, v7, v6
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_and_b32_e32 v6, 0xff800000, v6
|
|
; GFX1150-NEXT: v_add_f32_e32 v5, v6, v5
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_cvt_f16_f32_e32 v5, v5
|
|
; GFX1150-NEXT: v_div_fixup_f16 v5, v5, v3, v1
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_trunc_f16_e32 v5, v5
|
|
; GFX1150-NEXT: v_xor_b32_e32 v5, 0x8000, v5
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fmac_f16_e32 v1, v5, v3
|
|
; GFX1150-NEXT: v_pack_b32_f16 v1, v1, v2
|
|
; GFX1150-NEXT: global_store_b64 v4, v[0:1], s[0:1]
|
|
; GFX1150-NEXT: s_endpgm
|
|
ptr addrspace(1) %in2) #0 {
|
|
%gep2 = getelementptr <4 x half>, ptr addrspace(1) %in2, i32 4
|
|
%r0 = load <4 x half>, ptr addrspace(1) %in1, align 16
|
|
%r1 = load <4 x half>, ptr addrspace(1) %gep2, align 16
|
|
%r2 = frem <4 x half> %r0, %r1
|
|
store <4 x half> %r2, ptr addrspace(1) %out, align 16
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @frem_v2f32(ptr addrspace(1) %out, ptr addrspace(1) %in1,
|
|
; SI-LABEL: frem_v2f32:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_load_dwordx4 s[8:11], s[4:5], 0x9
|
|
; SI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0xd
|
|
; SI-NEXT: s_mov_b32 s3, 0xf000
|
|
; SI-NEXT: s_mov_b32 s2, -1
|
|
; SI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SI-NEXT: s_mov_b32 s0, s8
|
|
; SI-NEXT: s_mov_b32 s1, s9
|
|
; SI-NEXT: s_mov_b32 s8, s10
|
|
; SI-NEXT: s_mov_b32 s9, s11
|
|
; SI-NEXT: s_mov_b32 s10, s2
|
|
; SI-NEXT: s_mov_b32 s11, s3
|
|
; SI-NEXT: s_mov_b32 s6, s2
|
|
; SI-NEXT: s_mov_b32 s7, s3
|
|
; SI-NEXT: buffer_load_dwordx2 v[0:1], off, s[8:11], 0
|
|
; SI-NEXT: buffer_load_dwordx2 v[2:3], off, s[4:7], 0 offset:32
|
|
; SI-NEXT: s_waitcnt vmcnt(0)
|
|
; SI-NEXT: v_div_scale_f32 v4, vcc, v1, v3, v1
|
|
; SI-NEXT: v_div_scale_f32 v5, s[4:5], v3, v3, v1
|
|
; SI-NEXT: v_rcp_f32_e32 v6, v5
|
|
; SI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; SI-NEXT: v_fma_f32 v7, -v5, v6, 1.0
|
|
; SI-NEXT: v_fma_f32 v6, v7, v6, v6
|
|
; SI-NEXT: v_mul_f32_e32 v7, v4, v6
|
|
; SI-NEXT: v_fma_f32 v8, -v5, v7, v4
|
|
; SI-NEXT: v_fma_f32 v7, v8, v6, v7
|
|
; SI-NEXT: v_fma_f32 v4, -v5, v7, v4
|
|
; SI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; SI-NEXT: v_div_fmas_f32 v4, v4, v6, v7
|
|
; SI-NEXT: v_div_fixup_f32 v4, v4, v3, v1
|
|
; SI-NEXT: v_trunc_f32_e32 v4, v4
|
|
; SI-NEXT: v_fma_f32 v1, -v4, v3, v1
|
|
; SI-NEXT: v_div_scale_f32 v3, vcc, v0, v2, v0
|
|
; SI-NEXT: v_div_scale_f32 v4, s[4:5], v2, v2, v0
|
|
; SI-NEXT: v_rcp_f32_e32 v5, v4
|
|
; SI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; SI-NEXT: v_fma_f32 v6, -v4, v5, 1.0
|
|
; SI-NEXT: v_fma_f32 v5, v6, v5, v5
|
|
; SI-NEXT: v_mul_f32_e32 v6, v3, v5
|
|
; SI-NEXT: v_fma_f32 v7, -v4, v6, v3
|
|
; SI-NEXT: v_fma_f32 v6, v7, v5, v6
|
|
; SI-NEXT: v_fma_f32 v3, -v4, v6, v3
|
|
; SI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; SI-NEXT: v_div_fmas_f32 v3, v3, v5, v6
|
|
; SI-NEXT: v_div_fixup_f32 v3, v3, v2, v0
|
|
; SI-NEXT: v_trunc_f32_e32 v3, v3
|
|
; SI-NEXT: v_fma_f32 v0, -v3, v2, v0
|
|
; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0
|
|
; SI-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: frem_v2f32:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx4 s[8:11], s[4:5], 0x9
|
|
; CI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0xd
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: s_mov_b32 s6, s2
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_mov_b32 s0, s8
|
|
; CI-NEXT: s_mov_b32 s1, s9
|
|
; CI-NEXT: s_mov_b32 s8, s10
|
|
; CI-NEXT: s_mov_b32 s9, s11
|
|
; CI-NEXT: s_mov_b32 s10, s2
|
|
; CI-NEXT: s_mov_b32 s11, s3
|
|
; CI-NEXT: s_mov_b32 s7, s3
|
|
; CI-NEXT: buffer_load_dwordx2 v[0:1], off, s[8:11], 0
|
|
; CI-NEXT: buffer_load_dwordx2 v[2:3], off, s[4:7], 0 offset:32
|
|
; CI-NEXT: s_waitcnt vmcnt(0)
|
|
; CI-NEXT: v_div_scale_f32 v5, s[4:5], v3, v3, v1
|
|
; CI-NEXT: v_div_scale_f32 v4, vcc, v1, v3, v1
|
|
; CI-NEXT: v_rcp_f32_e32 v6, v5
|
|
; CI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; CI-NEXT: v_fma_f32 v7, -v5, v6, 1.0
|
|
; CI-NEXT: v_fma_f32 v6, v7, v6, v6
|
|
; CI-NEXT: v_mul_f32_e32 v7, v4, v6
|
|
; CI-NEXT: v_fma_f32 v8, -v5, v7, v4
|
|
; CI-NEXT: v_fma_f32 v7, v8, v6, v7
|
|
; CI-NEXT: v_fma_f32 v4, -v5, v7, v4
|
|
; CI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; CI-NEXT: v_div_fmas_f32 v4, v4, v6, v7
|
|
; CI-NEXT: v_div_fixup_f32 v4, v4, v3, v1
|
|
; CI-NEXT: v_trunc_f32_e32 v4, v4
|
|
; CI-NEXT: v_fma_f32 v1, -v4, v3, v1
|
|
; CI-NEXT: v_div_scale_f32 v4, s[4:5], v2, v2, v0
|
|
; CI-NEXT: v_div_scale_f32 v3, vcc, v0, v2, v0
|
|
; CI-NEXT: v_rcp_f32_e32 v5, v4
|
|
; CI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; CI-NEXT: v_fma_f32 v6, -v4, v5, 1.0
|
|
; CI-NEXT: v_fma_f32 v5, v6, v5, v5
|
|
; CI-NEXT: v_mul_f32_e32 v6, v3, v5
|
|
; CI-NEXT: v_fma_f32 v7, -v4, v6, v3
|
|
; CI-NEXT: v_fma_f32 v6, v7, v5, v6
|
|
; CI-NEXT: v_fma_f32 v3, -v4, v6, v3
|
|
; CI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; CI-NEXT: v_div_fmas_f32 v3, v3, v5, v6
|
|
; CI-NEXT: v_div_fixup_f32 v3, v3, v2, v0
|
|
; CI-NEXT: v_trunc_f32_e32 v3, v3
|
|
; CI-NEXT: v_fma_f32 v0, -v3, v2, v0
|
|
; CI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; VI-LABEL: frem_v2f32:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; VI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x34
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: s_add_u32 s0, s4, 32
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: s_addc_u32 s1, s5, 0
|
|
; VI-NEXT: v_mov_b32_e32 v5, s1
|
|
; VI-NEXT: v_mov_b32_e32 v2, s2
|
|
; VI-NEXT: v_mov_b32_e32 v3, s3
|
|
; VI-NEXT: v_mov_b32_e32 v4, s0
|
|
; VI-NEXT: flat_load_dwordx2 v[2:3], v[2:3]
|
|
; VI-NEXT: flat_load_dwordx2 v[4:5], v[4:5]
|
|
; VI-NEXT: s_waitcnt vmcnt(0)
|
|
; VI-NEXT: v_div_scale_f32 v7, s[0:1], v5, v5, v3
|
|
; VI-NEXT: v_div_scale_f32 v6, vcc, v3, v5, v3
|
|
; VI-NEXT: v_rcp_f32_e32 v8, v7
|
|
; VI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; VI-NEXT: v_fma_f32 v9, -v7, v8, 1.0
|
|
; VI-NEXT: v_fma_f32 v8, v9, v8, v8
|
|
; VI-NEXT: v_mul_f32_e32 v9, v6, v8
|
|
; VI-NEXT: v_fma_f32 v10, -v7, v9, v6
|
|
; VI-NEXT: v_fma_f32 v9, v10, v8, v9
|
|
; VI-NEXT: v_fma_f32 v6, -v7, v9, v6
|
|
; VI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; VI-NEXT: v_div_fmas_f32 v6, v6, v8, v9
|
|
; VI-NEXT: v_div_fixup_f32 v6, v6, v5, v3
|
|
; VI-NEXT: v_trunc_f32_e32 v6, v6
|
|
; VI-NEXT: v_fma_f32 v3, -v6, v5, v3
|
|
; VI-NEXT: v_div_scale_f32 v6, s[0:1], v4, v4, v2
|
|
; VI-NEXT: v_div_scale_f32 v5, vcc, v2, v4, v2
|
|
; VI-NEXT: v_rcp_f32_e32 v7, v6
|
|
; VI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; VI-NEXT: v_fma_f32 v8, -v6, v7, 1.0
|
|
; VI-NEXT: v_fma_f32 v7, v8, v7, v7
|
|
; VI-NEXT: v_mul_f32_e32 v8, v5, v7
|
|
; VI-NEXT: v_fma_f32 v9, -v6, v8, v5
|
|
; VI-NEXT: v_fma_f32 v8, v9, v7, v8
|
|
; VI-NEXT: v_fma_f32 v5, -v6, v8, v5
|
|
; VI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; VI-NEXT: v_div_fmas_f32 v5, v5, v7, v8
|
|
; VI-NEXT: v_div_fixup_f32 v5, v5, v4, v2
|
|
; VI-NEXT: v_trunc_f32_e32 v5, v5
|
|
; VI-NEXT: v_fma_f32 v2, -v5, v4, v2
|
|
; VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3]
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: frem_v2f32:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; GFX9-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
|
|
; GFX9-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_load_dwordx2 v[0:1], v4, s[2:3]
|
|
; GFX9-NEXT: global_load_dwordx2 v[2:3], v4, s[6:7] offset:32
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX9-NEXT: v_div_scale_f32 v6, s[2:3], v3, v3, v1
|
|
; GFX9-NEXT: v_div_scale_f32 v5, vcc, v1, v3, v1
|
|
; GFX9-NEXT: v_rcp_f32_e32 v7, v6
|
|
; GFX9-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; GFX9-NEXT: v_fma_f32 v8, -v6, v7, 1.0
|
|
; GFX9-NEXT: v_fma_f32 v7, v8, v7, v7
|
|
; GFX9-NEXT: v_mul_f32_e32 v8, v5, v7
|
|
; GFX9-NEXT: v_fma_f32 v9, -v6, v8, v5
|
|
; GFX9-NEXT: v_fma_f32 v8, v9, v7, v8
|
|
; GFX9-NEXT: v_fma_f32 v5, -v6, v8, v5
|
|
; GFX9-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; GFX9-NEXT: v_div_fmas_f32 v5, v5, v7, v8
|
|
; GFX9-NEXT: v_div_fixup_f32 v5, v5, v3, v1
|
|
; GFX9-NEXT: v_trunc_f32_e32 v5, v5
|
|
; GFX9-NEXT: v_fma_f32 v1, -v5, v3, v1
|
|
; GFX9-NEXT: v_div_scale_f32 v5, s[2:3], v2, v2, v0
|
|
; GFX9-NEXT: v_div_scale_f32 v3, vcc, v0, v2, v0
|
|
; GFX9-NEXT: v_rcp_f32_e32 v6, v5
|
|
; GFX9-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; GFX9-NEXT: v_fma_f32 v7, -v5, v6, 1.0
|
|
; GFX9-NEXT: v_fma_f32 v6, v7, v6, v6
|
|
; GFX9-NEXT: v_mul_f32_e32 v7, v3, v6
|
|
; GFX9-NEXT: v_fma_f32 v8, -v5, v7, v3
|
|
; GFX9-NEXT: v_fma_f32 v7, v8, v6, v7
|
|
; GFX9-NEXT: v_fma_f32 v3, -v5, v7, v3
|
|
; GFX9-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; GFX9-NEXT: v_div_fmas_f32 v3, v3, v6, v7
|
|
; GFX9-NEXT: v_div_fixup_f32 v3, v3, v2, v0
|
|
; GFX9-NEXT: v_trunc_f32_e32 v3, v3
|
|
; GFX9-NEXT: v_fma_f32 v0, -v3, v2, v0
|
|
; GFX9-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; GFX10-LABEL: frem_v2f32:
|
|
; GFX10: ; %bb.0:
|
|
; GFX10-NEXT: s_clause 0x1
|
|
; GFX10-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; GFX10-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
|
|
; GFX10-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX10-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-NEXT: s_clause 0x1
|
|
; GFX10-NEXT: global_load_dwordx2 v[0:1], v4, s[2:3]
|
|
; GFX10-NEXT: global_load_dwordx2 v[2:3], v4, s[6:7] offset:32
|
|
; GFX10-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX10-NEXT: v_div_scale_f32 v6, s2, v3, v3, v1
|
|
; GFX10-NEXT: v_div_scale_f32 v5, vcc_lo, v1, v3, v1
|
|
; GFX10-NEXT: v_rcp_f32_e32 v7, v6
|
|
; GFX10-NEXT: s_denorm_mode 15
|
|
; GFX10-NEXT: v_fma_f32 v8, -v6, v7, 1.0
|
|
; GFX10-NEXT: v_fmac_f32_e32 v7, v8, v7
|
|
; GFX10-NEXT: v_mul_f32_e32 v8, v5, v7
|
|
; GFX10-NEXT: v_fma_f32 v9, -v6, v8, v5
|
|
; GFX10-NEXT: v_fmac_f32_e32 v8, v9, v7
|
|
; GFX10-NEXT: v_fma_f32 v5, -v6, v8, v5
|
|
; GFX10-NEXT: s_denorm_mode 12
|
|
; GFX10-NEXT: v_div_fmas_f32 v5, v5, v7, v8
|
|
; GFX10-NEXT: v_div_fixup_f32 v5, v5, v3, v1
|
|
; GFX10-NEXT: v_trunc_f32_e32 v5, v5
|
|
; GFX10-NEXT: v_fma_f32 v1, -v5, v3, v1
|
|
; GFX10-NEXT: v_div_scale_f32 v5, s2, v2, v2, v0
|
|
; GFX10-NEXT: v_div_scale_f32 v3, vcc_lo, v0, v2, v0
|
|
; GFX10-NEXT: v_rcp_f32_e32 v6, v5
|
|
; GFX10-NEXT: s_denorm_mode 15
|
|
; GFX10-NEXT: v_fma_f32 v7, -v5, v6, 1.0
|
|
; GFX10-NEXT: v_fmac_f32_e32 v6, v7, v6
|
|
; GFX10-NEXT: v_mul_f32_e32 v7, v3, v6
|
|
; GFX10-NEXT: v_fma_f32 v8, -v5, v7, v3
|
|
; GFX10-NEXT: v_fmac_f32_e32 v7, v8, v6
|
|
; GFX10-NEXT: v_fma_f32 v3, -v5, v7, v3
|
|
; GFX10-NEXT: s_denorm_mode 12
|
|
; GFX10-NEXT: v_div_fmas_f32 v3, v3, v6, v7
|
|
; GFX10-NEXT: v_div_fixup_f32 v3, v3, v2, v0
|
|
; GFX10-NEXT: v_trunc_f32_e32 v3, v3
|
|
; GFX10-NEXT: v_fma_f32 v0, -v3, v2, v0
|
|
; GFX10-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX10-NEXT: s_endpgm
|
|
;
|
|
; GFX11-LABEL: frem_v2f32:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_clause 0x1
|
|
; GFX11-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX11-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX11-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: s_clause 0x1
|
|
; GFX11-NEXT: global_load_b64 v[0:1], v4, s[2:3]
|
|
; GFX11-NEXT: global_load_b64 v[2:3], v4, s[4:5] offset:32
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-NEXT: v_div_scale_f32 v6, null, v3, v3, v1
|
|
; GFX11-NEXT: v_div_scale_f32 v5, vcc_lo, v1, v3, v1
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_3) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_rcp_f32_e32 v7, v6
|
|
; GFX11-NEXT: s_denorm_mode 15
|
|
; GFX11-NEXT: s_waitcnt_depctr 0xfff
|
|
; GFX11-NEXT: v_fma_f32 v8, -v6, v7, 1.0
|
|
; GFX11-NEXT: v_fmac_f32_e32 v7, v8, v7
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_mul_f32_e32 v8, v5, v7
|
|
; GFX11-NEXT: v_fma_f32 v9, -v6, v8, v5
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_fmac_f32_e32 v8, v9, v7
|
|
; GFX11-NEXT: v_fma_f32 v5, -v6, v8, v5
|
|
; GFX11-NEXT: s_denorm_mode 12
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_div_fmas_f32 v5, v5, v7, v8
|
|
; GFX11-NEXT: v_div_fixup_f32 v5, v5, v3, v1
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_trunc_f32_e32 v5, v5
|
|
; GFX11-NEXT: v_fma_f32 v1, -v5, v3, v1
|
|
; GFX11-NEXT: v_div_scale_f32 v5, null, v2, v2, v0
|
|
; GFX11-NEXT: v_div_scale_f32 v3, vcc_lo, v0, v2, v0
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_3) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_rcp_f32_e32 v6, v5
|
|
; GFX11-NEXT: s_denorm_mode 15
|
|
; GFX11-NEXT: s_waitcnt_depctr 0xfff
|
|
; GFX11-NEXT: v_fma_f32 v7, -v5, v6, 1.0
|
|
; GFX11-NEXT: v_fmac_f32_e32 v6, v7, v6
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_mul_f32_e32 v7, v3, v6
|
|
; GFX11-NEXT: v_fma_f32 v8, -v5, v7, v3
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_fmac_f32_e32 v7, v8, v6
|
|
; GFX11-NEXT: v_fma_f32 v3, -v5, v7, v3
|
|
; GFX11-NEXT: s_denorm_mode 12
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_div_fmas_f32 v3, v3, v6, v7
|
|
; GFX11-NEXT: v_div_fixup_f32 v3, v3, v2, v0
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_trunc_f32_e32 v3, v3
|
|
; GFX11-NEXT: v_fma_f32 v0, -v3, v2, v0
|
|
; GFX11-NEXT: global_store_b64 v4, v[0:1], s[0:1]
|
|
; GFX11-NEXT: s_endpgm
|
|
;
|
|
; GFX1150-LABEL: frem_v2f32:
|
|
; GFX1150: ; %bb.0:
|
|
; GFX1150-NEXT: s_clause 0x1
|
|
; GFX1150-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX1150-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX1150-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX1150-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX1150-NEXT: s_clause 0x1
|
|
; GFX1150-NEXT: global_load_b64 v[0:1], v4, s[2:3]
|
|
; GFX1150-NEXT: global_load_b64 v[2:3], v4, s[4:5] offset:32
|
|
; GFX1150-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX1150-NEXT: v_div_scale_f32 v6, null, v3, v3, v1
|
|
; GFX1150-NEXT: v_div_scale_f32 v5, vcc_lo, v1, v3, v1
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(TRANS32_DEP_1)
|
|
; GFX1150-NEXT: v_rcp_f32_e32 v7, v6
|
|
; GFX1150-NEXT: s_denorm_mode 15
|
|
; GFX1150-NEXT: v_fma_f32 v8, -v6, v7, 1.0
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fmac_f32_e32 v7, v8, v7
|
|
; GFX1150-NEXT: v_mul_f32_e32 v8, v5, v7
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_f32 v9, -v6, v8, v5
|
|
; GFX1150-NEXT: v_fmac_f32_e32 v8, v9, v7
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_f32 v5, -v6, v8, v5
|
|
; GFX1150-NEXT: s_denorm_mode 12
|
|
; GFX1150-NEXT: v_div_fmas_f32 v5, v5, v7, v8
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_div_fixup_f32 v5, v5, v3, v1
|
|
; GFX1150-NEXT: v_trunc_f32_e32 v5, v5
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_xor_b32_e32 v5, 0x80000000, v5
|
|
; GFX1150-NEXT: v_fma_f32 v1, v5, v3, v1
|
|
; GFX1150-NEXT: v_div_scale_f32 v5, null, v2, v2, v0
|
|
; GFX1150-NEXT: v_div_scale_f32 v3, vcc_lo, v0, v2, v0
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(TRANS32_DEP_1)
|
|
; GFX1150-NEXT: v_rcp_f32_e32 v6, v5
|
|
; GFX1150-NEXT: s_denorm_mode 15
|
|
; GFX1150-NEXT: v_fma_f32 v7, -v5, v6, 1.0
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fmac_f32_e32 v6, v7, v6
|
|
; GFX1150-NEXT: v_mul_f32_e32 v7, v3, v6
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_f32 v8, -v5, v7, v3
|
|
; GFX1150-NEXT: v_fmac_f32_e32 v7, v8, v6
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_f32 v3, -v5, v7, v3
|
|
; GFX1150-NEXT: s_denorm_mode 12
|
|
; GFX1150-NEXT: v_div_fmas_f32 v3, v3, v6, v7
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_div_fixup_f32 v3, v3, v2, v0
|
|
; GFX1150-NEXT: v_trunc_f32_e32 v3, v3
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_xor_b32_e32 v3, 0x80000000, v3
|
|
; GFX1150-NEXT: v_fmac_f32_e32 v0, v3, v2
|
|
; GFX1150-NEXT: global_store_b64 v4, v[0:1], s[0:1]
|
|
; GFX1150-NEXT: s_endpgm
|
|
ptr addrspace(1) %in2) #0 {
|
|
%gep2 = getelementptr <2 x float>, ptr addrspace(1) %in2, i32 4
|
|
%r0 = load <2 x float>, ptr addrspace(1) %in1, align 8
|
|
%r1 = load <2 x float>, ptr addrspace(1) %gep2, align 8
|
|
%r2 = frem <2 x float> %r0, %r1
|
|
store <2 x float> %r2, ptr addrspace(1) %out, align 8
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @frem_v4f32(ptr addrspace(1) %out, ptr addrspace(1) %in1,
|
|
; SI-LABEL: frem_v4f32:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_load_dwordx4 s[8:11], s[4:5], 0x9
|
|
; SI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0xd
|
|
; SI-NEXT: s_mov_b32 s3, 0xf000
|
|
; SI-NEXT: s_mov_b32 s2, -1
|
|
; SI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SI-NEXT: s_mov_b32 s0, s8
|
|
; SI-NEXT: s_mov_b32 s1, s9
|
|
; SI-NEXT: s_mov_b32 s8, s10
|
|
; SI-NEXT: s_mov_b32 s9, s11
|
|
; SI-NEXT: s_mov_b32 s10, s2
|
|
; SI-NEXT: s_mov_b32 s11, s3
|
|
; SI-NEXT: s_mov_b32 s6, s2
|
|
; SI-NEXT: s_mov_b32 s7, s3
|
|
; SI-NEXT: buffer_load_dwordx4 v[0:3], off, s[8:11], 0
|
|
; SI-NEXT: buffer_load_dwordx4 v[4:7], off, s[4:7], 0 offset:64
|
|
; SI-NEXT: s_waitcnt vmcnt(0)
|
|
; SI-NEXT: v_div_scale_f32 v8, vcc, v3, v7, v3
|
|
; SI-NEXT: v_div_scale_f32 v9, s[4:5], v7, v7, v3
|
|
; SI-NEXT: v_rcp_f32_e32 v10, v9
|
|
; SI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; SI-NEXT: v_fma_f32 v11, -v9, v10, 1.0
|
|
; SI-NEXT: v_fma_f32 v10, v11, v10, v10
|
|
; SI-NEXT: v_mul_f32_e32 v11, v8, v10
|
|
; SI-NEXT: v_fma_f32 v12, -v9, v11, v8
|
|
; SI-NEXT: v_fma_f32 v11, v12, v10, v11
|
|
; SI-NEXT: v_fma_f32 v8, -v9, v11, v8
|
|
; SI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; SI-NEXT: v_div_fmas_f32 v8, v8, v10, v11
|
|
; SI-NEXT: v_div_fixup_f32 v8, v8, v7, v3
|
|
; SI-NEXT: v_trunc_f32_e32 v8, v8
|
|
; SI-NEXT: v_fma_f32 v3, -v8, v7, v3
|
|
; SI-NEXT: v_div_scale_f32 v7, vcc, v2, v6, v2
|
|
; SI-NEXT: v_div_scale_f32 v8, s[4:5], v6, v6, v2
|
|
; SI-NEXT: v_rcp_f32_e32 v9, v8
|
|
; SI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; SI-NEXT: v_fma_f32 v10, -v8, v9, 1.0
|
|
; SI-NEXT: v_fma_f32 v9, v10, v9, v9
|
|
; SI-NEXT: v_mul_f32_e32 v10, v7, v9
|
|
; SI-NEXT: v_fma_f32 v11, -v8, v10, v7
|
|
; SI-NEXT: v_fma_f32 v10, v11, v9, v10
|
|
; SI-NEXT: v_fma_f32 v7, -v8, v10, v7
|
|
; SI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; SI-NEXT: v_div_fmas_f32 v7, v7, v9, v10
|
|
; SI-NEXT: v_div_fixup_f32 v7, v7, v6, v2
|
|
; SI-NEXT: v_trunc_f32_e32 v7, v7
|
|
; SI-NEXT: v_fma_f32 v2, -v7, v6, v2
|
|
; SI-NEXT: v_div_scale_f32 v6, vcc, v1, v5, v1
|
|
; SI-NEXT: v_div_scale_f32 v7, s[4:5], v5, v5, v1
|
|
; SI-NEXT: v_rcp_f32_e32 v8, v7
|
|
; SI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; SI-NEXT: v_fma_f32 v9, -v7, v8, 1.0
|
|
; SI-NEXT: v_fma_f32 v8, v9, v8, v8
|
|
; SI-NEXT: v_mul_f32_e32 v9, v6, v8
|
|
; SI-NEXT: v_fma_f32 v10, -v7, v9, v6
|
|
; SI-NEXT: v_fma_f32 v9, v10, v8, v9
|
|
; SI-NEXT: v_fma_f32 v6, -v7, v9, v6
|
|
; SI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; SI-NEXT: v_div_fmas_f32 v6, v6, v8, v9
|
|
; SI-NEXT: v_div_fixup_f32 v6, v6, v5, v1
|
|
; SI-NEXT: v_trunc_f32_e32 v6, v6
|
|
; SI-NEXT: v_fma_f32 v1, -v6, v5, v1
|
|
; SI-NEXT: v_div_scale_f32 v5, vcc, v0, v4, v0
|
|
; SI-NEXT: v_div_scale_f32 v6, s[4:5], v4, v4, v0
|
|
; SI-NEXT: v_rcp_f32_e32 v7, v6
|
|
; SI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; SI-NEXT: v_fma_f32 v8, -v6, v7, 1.0
|
|
; SI-NEXT: v_fma_f32 v7, v8, v7, v7
|
|
; SI-NEXT: v_mul_f32_e32 v8, v5, v7
|
|
; SI-NEXT: v_fma_f32 v9, -v6, v8, v5
|
|
; SI-NEXT: v_fma_f32 v8, v9, v7, v8
|
|
; SI-NEXT: v_fma_f32 v5, -v6, v8, v5
|
|
; SI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; SI-NEXT: v_div_fmas_f32 v5, v5, v7, v8
|
|
; SI-NEXT: v_div_fixup_f32 v5, v5, v4, v0
|
|
; SI-NEXT: v_trunc_f32_e32 v5, v5
|
|
; SI-NEXT: v_fma_f32 v0, -v5, v4, v0
|
|
; SI-NEXT: buffer_store_dwordx4 v[0:3], off, s[0:3], 0
|
|
; SI-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: frem_v4f32:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx4 s[8:11], s[4:5], 0x9
|
|
; CI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0xd
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: s_mov_b32 s6, s2
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_mov_b32 s0, s8
|
|
; CI-NEXT: s_mov_b32 s1, s9
|
|
; CI-NEXT: s_mov_b32 s8, s10
|
|
; CI-NEXT: s_mov_b32 s9, s11
|
|
; CI-NEXT: s_mov_b32 s10, s2
|
|
; CI-NEXT: s_mov_b32 s11, s3
|
|
; CI-NEXT: s_mov_b32 s7, s3
|
|
; CI-NEXT: buffer_load_dwordx4 v[0:3], off, s[8:11], 0
|
|
; CI-NEXT: buffer_load_dwordx4 v[4:7], off, s[4:7], 0 offset:64
|
|
; CI-NEXT: s_waitcnt vmcnt(0)
|
|
; CI-NEXT: v_div_scale_f32 v9, s[4:5], v7, v7, v3
|
|
; CI-NEXT: v_div_scale_f32 v8, vcc, v3, v7, v3
|
|
; CI-NEXT: v_rcp_f32_e32 v10, v9
|
|
; CI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; CI-NEXT: v_fma_f32 v11, -v9, v10, 1.0
|
|
; CI-NEXT: v_fma_f32 v10, v11, v10, v10
|
|
; CI-NEXT: v_mul_f32_e32 v11, v8, v10
|
|
; CI-NEXT: v_fma_f32 v12, -v9, v11, v8
|
|
; CI-NEXT: v_fma_f32 v11, v12, v10, v11
|
|
; CI-NEXT: v_fma_f32 v8, -v9, v11, v8
|
|
; CI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; CI-NEXT: v_div_fmas_f32 v8, v8, v10, v11
|
|
; CI-NEXT: v_div_fixup_f32 v8, v8, v7, v3
|
|
; CI-NEXT: v_trunc_f32_e32 v8, v8
|
|
; CI-NEXT: v_fma_f32 v3, -v8, v7, v3
|
|
; CI-NEXT: v_div_scale_f32 v8, s[4:5], v6, v6, v2
|
|
; CI-NEXT: v_div_scale_f32 v7, vcc, v2, v6, v2
|
|
; CI-NEXT: v_rcp_f32_e32 v9, v8
|
|
; CI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; CI-NEXT: v_fma_f32 v10, -v8, v9, 1.0
|
|
; CI-NEXT: v_fma_f32 v9, v10, v9, v9
|
|
; CI-NEXT: v_mul_f32_e32 v10, v7, v9
|
|
; CI-NEXT: v_fma_f32 v11, -v8, v10, v7
|
|
; CI-NEXT: v_fma_f32 v10, v11, v9, v10
|
|
; CI-NEXT: v_fma_f32 v7, -v8, v10, v7
|
|
; CI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; CI-NEXT: v_div_fmas_f32 v7, v7, v9, v10
|
|
; CI-NEXT: v_div_fixup_f32 v7, v7, v6, v2
|
|
; CI-NEXT: v_trunc_f32_e32 v7, v7
|
|
; CI-NEXT: v_fma_f32 v2, -v7, v6, v2
|
|
; CI-NEXT: v_div_scale_f32 v7, s[4:5], v5, v5, v1
|
|
; CI-NEXT: v_div_scale_f32 v6, vcc, v1, v5, v1
|
|
; CI-NEXT: v_rcp_f32_e32 v8, v7
|
|
; CI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; CI-NEXT: v_fma_f32 v9, -v7, v8, 1.0
|
|
; CI-NEXT: v_fma_f32 v8, v9, v8, v8
|
|
; CI-NEXT: v_mul_f32_e32 v9, v6, v8
|
|
; CI-NEXT: v_fma_f32 v10, -v7, v9, v6
|
|
; CI-NEXT: v_fma_f32 v9, v10, v8, v9
|
|
; CI-NEXT: v_fma_f32 v6, -v7, v9, v6
|
|
; CI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; CI-NEXT: v_div_fmas_f32 v6, v6, v8, v9
|
|
; CI-NEXT: v_div_fixup_f32 v6, v6, v5, v1
|
|
; CI-NEXT: v_trunc_f32_e32 v6, v6
|
|
; CI-NEXT: v_fma_f32 v1, -v6, v5, v1
|
|
; CI-NEXT: v_div_scale_f32 v6, s[4:5], v4, v4, v0
|
|
; CI-NEXT: v_div_scale_f32 v5, vcc, v0, v4, v0
|
|
; CI-NEXT: v_rcp_f32_e32 v7, v6
|
|
; CI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; CI-NEXT: v_fma_f32 v8, -v6, v7, 1.0
|
|
; CI-NEXT: v_fma_f32 v7, v8, v7, v7
|
|
; CI-NEXT: v_mul_f32_e32 v8, v5, v7
|
|
; CI-NEXT: v_fma_f32 v9, -v6, v8, v5
|
|
; CI-NEXT: v_fma_f32 v8, v9, v7, v8
|
|
; CI-NEXT: v_fma_f32 v5, -v6, v8, v5
|
|
; CI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; CI-NEXT: v_div_fmas_f32 v5, v5, v7, v8
|
|
; CI-NEXT: v_div_fixup_f32 v5, v5, v4, v0
|
|
; CI-NEXT: v_trunc_f32_e32 v5, v5
|
|
; CI-NEXT: v_fma_f32 v0, -v5, v4, v0
|
|
; CI-NEXT: buffer_store_dwordx4 v[0:3], off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; VI-LABEL: frem_v4f32:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; VI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x34
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v8, s0
|
|
; VI-NEXT: s_add_u32 s0, s4, 64
|
|
; VI-NEXT: v_mov_b32_e32 v9, s1
|
|
; VI-NEXT: s_addc_u32 s1, s5, 0
|
|
; VI-NEXT: v_mov_b32_e32 v5, s1
|
|
; VI-NEXT: v_mov_b32_e32 v0, s2
|
|
; VI-NEXT: v_mov_b32_e32 v1, s3
|
|
; VI-NEXT: v_mov_b32_e32 v4, s0
|
|
; VI-NEXT: flat_load_dwordx4 v[0:3], v[0:1]
|
|
; VI-NEXT: flat_load_dwordx4 v[4:7], v[4:5]
|
|
; VI-NEXT: s_waitcnt vmcnt(0)
|
|
; VI-NEXT: v_div_scale_f32 v11, s[0:1], v7, v7, v3
|
|
; VI-NEXT: v_div_scale_f32 v10, vcc, v3, v7, v3
|
|
; VI-NEXT: v_rcp_f32_e32 v12, v11
|
|
; VI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; VI-NEXT: v_fma_f32 v13, -v11, v12, 1.0
|
|
; VI-NEXT: v_fma_f32 v12, v13, v12, v12
|
|
; VI-NEXT: v_mul_f32_e32 v13, v10, v12
|
|
; VI-NEXT: v_fma_f32 v14, -v11, v13, v10
|
|
; VI-NEXT: v_fma_f32 v13, v14, v12, v13
|
|
; VI-NEXT: v_fma_f32 v10, -v11, v13, v10
|
|
; VI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; VI-NEXT: v_div_fmas_f32 v10, v10, v12, v13
|
|
; VI-NEXT: v_div_fixup_f32 v10, v10, v7, v3
|
|
; VI-NEXT: v_trunc_f32_e32 v10, v10
|
|
; VI-NEXT: v_fma_f32 v3, -v10, v7, v3
|
|
; VI-NEXT: v_div_scale_f32 v10, s[0:1], v6, v6, v2
|
|
; VI-NEXT: v_div_scale_f32 v7, vcc, v2, v6, v2
|
|
; VI-NEXT: v_rcp_f32_e32 v11, v10
|
|
; VI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; VI-NEXT: v_fma_f32 v12, -v10, v11, 1.0
|
|
; VI-NEXT: v_fma_f32 v11, v12, v11, v11
|
|
; VI-NEXT: v_mul_f32_e32 v12, v7, v11
|
|
; VI-NEXT: v_fma_f32 v13, -v10, v12, v7
|
|
; VI-NEXT: v_fma_f32 v12, v13, v11, v12
|
|
; VI-NEXT: v_fma_f32 v7, -v10, v12, v7
|
|
; VI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; VI-NEXT: v_div_fmas_f32 v7, v7, v11, v12
|
|
; VI-NEXT: v_div_fixup_f32 v7, v7, v6, v2
|
|
; VI-NEXT: v_trunc_f32_e32 v7, v7
|
|
; VI-NEXT: v_fma_f32 v2, -v7, v6, v2
|
|
; VI-NEXT: v_div_scale_f32 v7, s[0:1], v5, v5, v1
|
|
; VI-NEXT: v_div_scale_f32 v6, vcc, v1, v5, v1
|
|
; VI-NEXT: v_rcp_f32_e32 v10, v7
|
|
; VI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; VI-NEXT: v_fma_f32 v11, -v7, v10, 1.0
|
|
; VI-NEXT: v_fma_f32 v10, v11, v10, v10
|
|
; VI-NEXT: v_mul_f32_e32 v11, v6, v10
|
|
; VI-NEXT: v_fma_f32 v12, -v7, v11, v6
|
|
; VI-NEXT: v_fma_f32 v11, v12, v10, v11
|
|
; VI-NEXT: v_fma_f32 v6, -v7, v11, v6
|
|
; VI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; VI-NEXT: v_div_fmas_f32 v6, v6, v10, v11
|
|
; VI-NEXT: v_div_fixup_f32 v6, v6, v5, v1
|
|
; VI-NEXT: v_trunc_f32_e32 v6, v6
|
|
; VI-NEXT: v_fma_f32 v1, -v6, v5, v1
|
|
; VI-NEXT: v_div_scale_f32 v6, s[0:1], v4, v4, v0
|
|
; VI-NEXT: v_div_scale_f32 v5, vcc, v0, v4, v0
|
|
; VI-NEXT: v_rcp_f32_e32 v7, v6
|
|
; VI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; VI-NEXT: v_fma_f32 v10, -v6, v7, 1.0
|
|
; VI-NEXT: v_fma_f32 v7, v10, v7, v7
|
|
; VI-NEXT: v_mul_f32_e32 v10, v5, v7
|
|
; VI-NEXT: v_fma_f32 v11, -v6, v10, v5
|
|
; VI-NEXT: v_fma_f32 v10, v11, v7, v10
|
|
; VI-NEXT: v_fma_f32 v5, -v6, v10, v5
|
|
; VI-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; VI-NEXT: v_div_fmas_f32 v5, v5, v7, v10
|
|
; VI-NEXT: v_div_fixup_f32 v5, v5, v4, v0
|
|
; VI-NEXT: v_trunc_f32_e32 v5, v5
|
|
; VI-NEXT: v_fma_f32 v0, -v5, v4, v0
|
|
; VI-NEXT: flat_store_dwordx4 v[8:9], v[0:3]
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: frem_v4f32:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; GFX9-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
|
|
; GFX9-NEXT: v_mov_b32_e32 v8, 0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_load_dwordx4 v[0:3], v8, s[2:3]
|
|
; GFX9-NEXT: global_load_dwordx4 v[4:7], v8, s[6:7] offset:64
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX9-NEXT: v_div_scale_f32 v10, s[2:3], v7, v7, v3
|
|
; GFX9-NEXT: v_div_scale_f32 v9, vcc, v3, v7, v3
|
|
; GFX9-NEXT: v_rcp_f32_e32 v11, v10
|
|
; GFX9-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; GFX9-NEXT: v_fma_f32 v12, -v10, v11, 1.0
|
|
; GFX9-NEXT: v_fma_f32 v11, v12, v11, v11
|
|
; GFX9-NEXT: v_mul_f32_e32 v12, v9, v11
|
|
; GFX9-NEXT: v_fma_f32 v13, -v10, v12, v9
|
|
; GFX9-NEXT: v_fma_f32 v12, v13, v11, v12
|
|
; GFX9-NEXT: v_fma_f32 v9, -v10, v12, v9
|
|
; GFX9-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; GFX9-NEXT: v_div_fmas_f32 v9, v9, v11, v12
|
|
; GFX9-NEXT: v_div_fixup_f32 v9, v9, v7, v3
|
|
; GFX9-NEXT: v_trunc_f32_e32 v9, v9
|
|
; GFX9-NEXT: v_fma_f32 v3, -v9, v7, v3
|
|
; GFX9-NEXT: v_div_scale_f32 v9, s[2:3], v6, v6, v2
|
|
; GFX9-NEXT: v_div_scale_f32 v7, vcc, v2, v6, v2
|
|
; GFX9-NEXT: v_rcp_f32_e32 v10, v9
|
|
; GFX9-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; GFX9-NEXT: v_fma_f32 v11, -v9, v10, 1.0
|
|
; GFX9-NEXT: v_fma_f32 v10, v11, v10, v10
|
|
; GFX9-NEXT: v_mul_f32_e32 v11, v7, v10
|
|
; GFX9-NEXT: v_fma_f32 v12, -v9, v11, v7
|
|
; GFX9-NEXT: v_fma_f32 v11, v12, v10, v11
|
|
; GFX9-NEXT: v_fma_f32 v7, -v9, v11, v7
|
|
; GFX9-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; GFX9-NEXT: v_div_fmas_f32 v7, v7, v10, v11
|
|
; GFX9-NEXT: v_div_fixup_f32 v7, v7, v6, v2
|
|
; GFX9-NEXT: v_trunc_f32_e32 v7, v7
|
|
; GFX9-NEXT: v_fma_f32 v2, -v7, v6, v2
|
|
; GFX9-NEXT: v_div_scale_f32 v7, s[2:3], v5, v5, v1
|
|
; GFX9-NEXT: v_div_scale_f32 v6, vcc, v1, v5, v1
|
|
; GFX9-NEXT: v_rcp_f32_e32 v9, v7
|
|
; GFX9-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; GFX9-NEXT: v_fma_f32 v10, -v7, v9, 1.0
|
|
; GFX9-NEXT: v_fma_f32 v9, v10, v9, v9
|
|
; GFX9-NEXT: v_mul_f32_e32 v10, v6, v9
|
|
; GFX9-NEXT: v_fma_f32 v11, -v7, v10, v6
|
|
; GFX9-NEXT: v_fma_f32 v10, v11, v9, v10
|
|
; GFX9-NEXT: v_fma_f32 v6, -v7, v10, v6
|
|
; GFX9-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; GFX9-NEXT: v_div_fmas_f32 v6, v6, v9, v10
|
|
; GFX9-NEXT: v_div_fixup_f32 v6, v6, v5, v1
|
|
; GFX9-NEXT: v_trunc_f32_e32 v6, v6
|
|
; GFX9-NEXT: v_fma_f32 v1, -v6, v5, v1
|
|
; GFX9-NEXT: v_div_scale_f32 v6, s[2:3], v4, v4, v0
|
|
; GFX9-NEXT: v_div_scale_f32 v5, vcc, v0, v4, v0
|
|
; GFX9-NEXT: v_rcp_f32_e32 v7, v6
|
|
; GFX9-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
|
|
; GFX9-NEXT: v_fma_f32 v9, -v6, v7, 1.0
|
|
; GFX9-NEXT: v_fma_f32 v7, v9, v7, v7
|
|
; GFX9-NEXT: v_mul_f32_e32 v9, v5, v7
|
|
; GFX9-NEXT: v_fma_f32 v10, -v6, v9, v5
|
|
; GFX9-NEXT: v_fma_f32 v9, v10, v7, v9
|
|
; GFX9-NEXT: v_fma_f32 v5, -v6, v9, v5
|
|
; GFX9-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0
|
|
; GFX9-NEXT: v_div_fmas_f32 v5, v5, v7, v9
|
|
; GFX9-NEXT: v_div_fixup_f32 v5, v5, v4, v0
|
|
; GFX9-NEXT: v_trunc_f32_e32 v5, v5
|
|
; GFX9-NEXT: v_fma_f32 v0, -v5, v4, v0
|
|
; GFX9-NEXT: global_store_dwordx4 v8, v[0:3], s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; GFX10-LABEL: frem_v4f32:
|
|
; GFX10: ; %bb.0:
|
|
; GFX10-NEXT: s_clause 0x1
|
|
; GFX10-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; GFX10-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
|
|
; GFX10-NEXT: v_mov_b32_e32 v8, 0
|
|
; GFX10-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-NEXT: s_clause 0x1
|
|
; GFX10-NEXT: global_load_dwordx4 v[0:3], v8, s[2:3]
|
|
; GFX10-NEXT: global_load_dwordx4 v[4:7], v8, s[6:7] offset:64
|
|
; GFX10-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX10-NEXT: v_div_scale_f32 v10, s2, v7, v7, v3
|
|
; GFX10-NEXT: v_div_scale_f32 v9, vcc_lo, v3, v7, v3
|
|
; GFX10-NEXT: v_rcp_f32_e32 v11, v10
|
|
; GFX10-NEXT: s_denorm_mode 15
|
|
; GFX10-NEXT: v_fma_f32 v12, -v10, v11, 1.0
|
|
; GFX10-NEXT: v_fmac_f32_e32 v11, v12, v11
|
|
; GFX10-NEXT: v_mul_f32_e32 v12, v9, v11
|
|
; GFX10-NEXT: v_fma_f32 v13, -v10, v12, v9
|
|
; GFX10-NEXT: v_fmac_f32_e32 v12, v13, v11
|
|
; GFX10-NEXT: v_fma_f32 v9, -v10, v12, v9
|
|
; GFX10-NEXT: s_denorm_mode 12
|
|
; GFX10-NEXT: v_div_fmas_f32 v9, v9, v11, v12
|
|
; GFX10-NEXT: v_div_fixup_f32 v9, v9, v7, v3
|
|
; GFX10-NEXT: v_trunc_f32_e32 v9, v9
|
|
; GFX10-NEXT: v_fma_f32 v3, -v9, v7, v3
|
|
; GFX10-NEXT: v_div_scale_f32 v9, s2, v6, v6, v2
|
|
; GFX10-NEXT: v_div_scale_f32 v7, vcc_lo, v2, v6, v2
|
|
; GFX10-NEXT: v_rcp_f32_e32 v10, v9
|
|
; GFX10-NEXT: s_denorm_mode 15
|
|
; GFX10-NEXT: v_fma_f32 v11, -v9, v10, 1.0
|
|
; GFX10-NEXT: v_fmac_f32_e32 v10, v11, v10
|
|
; GFX10-NEXT: v_mul_f32_e32 v11, v7, v10
|
|
; GFX10-NEXT: v_fma_f32 v12, -v9, v11, v7
|
|
; GFX10-NEXT: v_fmac_f32_e32 v11, v12, v10
|
|
; GFX10-NEXT: v_fma_f32 v7, -v9, v11, v7
|
|
; GFX10-NEXT: s_denorm_mode 12
|
|
; GFX10-NEXT: v_div_fmas_f32 v7, v7, v10, v11
|
|
; GFX10-NEXT: v_div_fixup_f32 v7, v7, v6, v2
|
|
; GFX10-NEXT: v_trunc_f32_e32 v7, v7
|
|
; GFX10-NEXT: v_fma_f32 v2, -v7, v6, v2
|
|
; GFX10-NEXT: v_div_scale_f32 v7, s2, v5, v5, v1
|
|
; GFX10-NEXT: v_div_scale_f32 v6, vcc_lo, v1, v5, v1
|
|
; GFX10-NEXT: v_rcp_f32_e32 v9, v7
|
|
; GFX10-NEXT: s_denorm_mode 15
|
|
; GFX10-NEXT: v_fma_f32 v10, -v7, v9, 1.0
|
|
; GFX10-NEXT: v_fmac_f32_e32 v9, v10, v9
|
|
; GFX10-NEXT: v_mul_f32_e32 v10, v6, v9
|
|
; GFX10-NEXT: v_fma_f32 v11, -v7, v10, v6
|
|
; GFX10-NEXT: v_fmac_f32_e32 v10, v11, v9
|
|
; GFX10-NEXT: v_fma_f32 v6, -v7, v10, v6
|
|
; GFX10-NEXT: s_denorm_mode 12
|
|
; GFX10-NEXT: v_div_fmas_f32 v6, v6, v9, v10
|
|
; GFX10-NEXT: v_div_fixup_f32 v6, v6, v5, v1
|
|
; GFX10-NEXT: v_trunc_f32_e32 v6, v6
|
|
; GFX10-NEXT: v_fma_f32 v1, -v6, v5, v1
|
|
; GFX10-NEXT: v_div_scale_f32 v6, s2, v4, v4, v0
|
|
; GFX10-NEXT: v_div_scale_f32 v5, vcc_lo, v0, v4, v0
|
|
; GFX10-NEXT: v_rcp_f32_e32 v7, v6
|
|
; GFX10-NEXT: s_denorm_mode 15
|
|
; GFX10-NEXT: v_fma_f32 v9, -v6, v7, 1.0
|
|
; GFX10-NEXT: v_fmac_f32_e32 v7, v9, v7
|
|
; GFX10-NEXT: v_mul_f32_e32 v9, v5, v7
|
|
; GFX10-NEXT: v_fma_f32 v10, -v6, v9, v5
|
|
; GFX10-NEXT: v_fmac_f32_e32 v9, v10, v7
|
|
; GFX10-NEXT: v_fma_f32 v5, -v6, v9, v5
|
|
; GFX10-NEXT: s_denorm_mode 12
|
|
; GFX10-NEXT: v_div_fmas_f32 v5, v5, v7, v9
|
|
; GFX10-NEXT: v_div_fixup_f32 v5, v5, v4, v0
|
|
; GFX10-NEXT: v_trunc_f32_e32 v5, v5
|
|
; GFX10-NEXT: v_fma_f32 v0, -v5, v4, v0
|
|
; GFX10-NEXT: global_store_dwordx4 v8, v[0:3], s[0:1]
|
|
; GFX10-NEXT: s_endpgm
|
|
;
|
|
; GFX11-LABEL: frem_v4f32:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_clause 0x1
|
|
; GFX11-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX11-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX11-NEXT: v_mov_b32_e32 v8, 0
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: s_clause 0x1
|
|
; GFX11-NEXT: global_load_b128 v[0:3], v8, s[2:3]
|
|
; GFX11-NEXT: global_load_b128 v[4:7], v8, s[4:5] offset:64
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-NEXT: v_div_scale_f32 v10, null, v7, v7, v3
|
|
; GFX11-NEXT: v_div_scale_f32 v9, vcc_lo, v3, v7, v3
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_3) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_rcp_f32_e32 v11, v10
|
|
; GFX11-NEXT: s_denorm_mode 15
|
|
; GFX11-NEXT: s_waitcnt_depctr 0xfff
|
|
; GFX11-NEXT: v_fma_f32 v12, -v10, v11, 1.0
|
|
; GFX11-NEXT: v_fmac_f32_e32 v11, v12, v11
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_mul_f32_e32 v12, v9, v11
|
|
; GFX11-NEXT: v_fma_f32 v13, -v10, v12, v9
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_fmac_f32_e32 v12, v13, v11
|
|
; GFX11-NEXT: v_fma_f32 v9, -v10, v12, v9
|
|
; GFX11-NEXT: s_denorm_mode 12
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_div_fmas_f32 v9, v9, v11, v12
|
|
; GFX11-NEXT: v_div_fixup_f32 v9, v9, v7, v3
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_trunc_f32_e32 v9, v9
|
|
; GFX11-NEXT: v_fma_f32 v3, -v9, v7, v3
|
|
; GFX11-NEXT: v_div_scale_f32 v9, null, v6, v6, v2
|
|
; GFX11-NEXT: v_div_scale_f32 v7, vcc_lo, v2, v6, v2
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_3) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_rcp_f32_e32 v10, v9
|
|
; GFX11-NEXT: s_denorm_mode 15
|
|
; GFX11-NEXT: s_waitcnt_depctr 0xfff
|
|
; GFX11-NEXT: v_fma_f32 v11, -v9, v10, 1.0
|
|
; GFX11-NEXT: v_fmac_f32_e32 v10, v11, v10
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_mul_f32_e32 v11, v7, v10
|
|
; GFX11-NEXT: v_fma_f32 v12, -v9, v11, v7
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_fmac_f32_e32 v11, v12, v10
|
|
; GFX11-NEXT: v_fma_f32 v7, -v9, v11, v7
|
|
; GFX11-NEXT: s_denorm_mode 12
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_div_fmas_f32 v7, v7, v10, v11
|
|
; GFX11-NEXT: v_div_fixup_f32 v7, v7, v6, v2
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_trunc_f32_e32 v7, v7
|
|
; GFX11-NEXT: v_fma_f32 v2, -v7, v6, v2
|
|
; GFX11-NEXT: v_div_scale_f32 v7, null, v5, v5, v1
|
|
; GFX11-NEXT: v_div_scale_f32 v6, vcc_lo, v1, v5, v1
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_3) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_rcp_f32_e32 v9, v7
|
|
; GFX11-NEXT: s_denorm_mode 15
|
|
; GFX11-NEXT: s_waitcnt_depctr 0xfff
|
|
; GFX11-NEXT: v_fma_f32 v10, -v7, v9, 1.0
|
|
; GFX11-NEXT: v_fmac_f32_e32 v9, v10, v9
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_mul_f32_e32 v10, v6, v9
|
|
; GFX11-NEXT: v_fma_f32 v11, -v7, v10, v6
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_fmac_f32_e32 v10, v11, v9
|
|
; GFX11-NEXT: v_fma_f32 v6, -v7, v10, v6
|
|
; GFX11-NEXT: s_denorm_mode 12
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_div_fmas_f32 v6, v6, v9, v10
|
|
; GFX11-NEXT: v_div_fixup_f32 v6, v6, v5, v1
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_trunc_f32_e32 v6, v6
|
|
; GFX11-NEXT: v_fma_f32 v1, -v6, v5, v1
|
|
; GFX11-NEXT: v_div_scale_f32 v6, null, v4, v4, v0
|
|
; GFX11-NEXT: v_div_scale_f32 v5, vcc_lo, v0, v4, v0
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_3) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_rcp_f32_e32 v7, v6
|
|
; GFX11-NEXT: s_denorm_mode 15
|
|
; GFX11-NEXT: s_waitcnt_depctr 0xfff
|
|
; GFX11-NEXT: v_fma_f32 v9, -v6, v7, 1.0
|
|
; GFX11-NEXT: v_fmac_f32_e32 v7, v9, v7
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_mul_f32_e32 v9, v5, v7
|
|
; GFX11-NEXT: v_fma_f32 v10, -v6, v9, v5
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_fmac_f32_e32 v9, v10, v7
|
|
; GFX11-NEXT: v_fma_f32 v5, -v6, v9, v5
|
|
; GFX11-NEXT: s_denorm_mode 12
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_div_fmas_f32 v5, v5, v7, v9
|
|
; GFX11-NEXT: v_div_fixup_f32 v5, v5, v4, v0
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_trunc_f32_e32 v5, v5
|
|
; GFX11-NEXT: v_fma_f32 v0, -v5, v4, v0
|
|
; GFX11-NEXT: global_store_b128 v8, v[0:3], s[0:1]
|
|
; GFX11-NEXT: s_endpgm
|
|
;
|
|
; GFX1150-LABEL: frem_v4f32:
|
|
; GFX1150: ; %bb.0:
|
|
; GFX1150-NEXT: s_clause 0x1
|
|
; GFX1150-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX1150-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX1150-NEXT: v_mov_b32_e32 v8, 0
|
|
; GFX1150-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX1150-NEXT: s_clause 0x1
|
|
; GFX1150-NEXT: global_load_b128 v[0:3], v8, s[2:3]
|
|
; GFX1150-NEXT: global_load_b128 v[4:7], v8, s[4:5] offset:64
|
|
; GFX1150-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX1150-NEXT: v_div_scale_f32 v10, null, v7, v7, v3
|
|
; GFX1150-NEXT: v_div_scale_f32 v9, vcc_lo, v3, v7, v3
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(TRANS32_DEP_1)
|
|
; GFX1150-NEXT: v_rcp_f32_e32 v11, v10
|
|
; GFX1150-NEXT: s_denorm_mode 15
|
|
; GFX1150-NEXT: v_fma_f32 v12, -v10, v11, 1.0
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fmac_f32_e32 v11, v12, v11
|
|
; GFX1150-NEXT: v_mul_f32_e32 v12, v9, v11
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_f32 v13, -v10, v12, v9
|
|
; GFX1150-NEXT: v_fmac_f32_e32 v12, v13, v11
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_f32 v9, -v10, v12, v9
|
|
; GFX1150-NEXT: s_denorm_mode 12
|
|
; GFX1150-NEXT: v_div_fmas_f32 v9, v9, v11, v12
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_div_fixup_f32 v9, v9, v7, v3
|
|
; GFX1150-NEXT: v_trunc_f32_e32 v9, v9
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_xor_b32_e32 v9, 0x80000000, v9
|
|
; GFX1150-NEXT: v_fma_f32 v3, v9, v7, v3
|
|
; GFX1150-NEXT: v_div_scale_f32 v9, null, v6, v6, v2
|
|
; GFX1150-NEXT: v_div_scale_f32 v7, vcc_lo, v2, v6, v2
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(TRANS32_DEP_1)
|
|
; GFX1150-NEXT: v_rcp_f32_e32 v10, v9
|
|
; GFX1150-NEXT: s_denorm_mode 15
|
|
; GFX1150-NEXT: v_fma_f32 v11, -v9, v10, 1.0
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fmac_f32_e32 v10, v11, v10
|
|
; GFX1150-NEXT: v_mul_f32_e32 v11, v7, v10
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_f32 v12, -v9, v11, v7
|
|
; GFX1150-NEXT: v_fmac_f32_e32 v11, v12, v10
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_f32 v7, -v9, v11, v7
|
|
; GFX1150-NEXT: s_denorm_mode 12
|
|
; GFX1150-NEXT: v_div_fmas_f32 v7, v7, v10, v11
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_div_fixup_f32 v7, v7, v6, v2
|
|
; GFX1150-NEXT: v_trunc_f32_e32 v7, v7
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_xor_b32_e32 v7, 0x80000000, v7
|
|
; GFX1150-NEXT: v_fma_f32 v2, v7, v6, v2
|
|
; GFX1150-NEXT: v_div_scale_f32 v7, null, v5, v5, v1
|
|
; GFX1150-NEXT: v_div_scale_f32 v6, vcc_lo, v1, v5, v1
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(TRANS32_DEP_1)
|
|
; GFX1150-NEXT: v_rcp_f32_e32 v9, v7
|
|
; GFX1150-NEXT: s_denorm_mode 15
|
|
; GFX1150-NEXT: v_fma_f32 v10, -v7, v9, 1.0
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fmac_f32_e32 v9, v10, v9
|
|
; GFX1150-NEXT: v_mul_f32_e32 v10, v6, v9
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_f32 v11, -v7, v10, v6
|
|
; GFX1150-NEXT: v_fmac_f32_e32 v10, v11, v9
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_f32 v6, -v7, v10, v6
|
|
; GFX1150-NEXT: s_denorm_mode 12
|
|
; GFX1150-NEXT: v_div_fmas_f32 v6, v6, v9, v10
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_div_fixup_f32 v6, v6, v5, v1
|
|
; GFX1150-NEXT: v_trunc_f32_e32 v6, v6
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_xor_b32_e32 v6, 0x80000000, v6
|
|
; GFX1150-NEXT: v_fma_f32 v1, v6, v5, v1
|
|
; GFX1150-NEXT: v_div_scale_f32 v6, null, v4, v4, v0
|
|
; GFX1150-NEXT: v_div_scale_f32 v5, vcc_lo, v0, v4, v0
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(TRANS32_DEP_1)
|
|
; GFX1150-NEXT: v_rcp_f32_e32 v7, v6
|
|
; GFX1150-NEXT: s_denorm_mode 15
|
|
; GFX1150-NEXT: v_fma_f32 v9, -v6, v7, 1.0
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fmac_f32_e32 v7, v9, v7
|
|
; GFX1150-NEXT: v_mul_f32_e32 v9, v5, v7
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_f32 v10, -v6, v9, v5
|
|
; GFX1150-NEXT: v_fmac_f32_e32 v9, v10, v7
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_f32 v5, -v6, v9, v5
|
|
; GFX1150-NEXT: s_denorm_mode 12
|
|
; GFX1150-NEXT: v_div_fmas_f32 v5, v5, v7, v9
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_div_fixup_f32 v5, v5, v4, v0
|
|
; GFX1150-NEXT: v_trunc_f32_e32 v5, v5
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_xor_b32_e32 v5, 0x80000000, v5
|
|
; GFX1150-NEXT: v_fmac_f32_e32 v0, v5, v4
|
|
; GFX1150-NEXT: global_store_b128 v8, v[0:3], s[0:1]
|
|
; GFX1150-NEXT: s_endpgm
|
|
ptr addrspace(1) %in2) #0 {
|
|
%gep2 = getelementptr <4 x float>, ptr addrspace(1) %in2, i32 4
|
|
%r0 = load <4 x float>, ptr addrspace(1) %in1, align 16
|
|
%r1 = load <4 x float>, ptr addrspace(1) %gep2, align 16
|
|
%r2 = frem <4 x float> %r0, %r1
|
|
store <4 x float> %r2, ptr addrspace(1) %out, align 16
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @frem_v2f64(ptr addrspace(1) %out, ptr addrspace(1) %in1,
|
|
; SI-LABEL: frem_v2f64:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9
|
|
; SI-NEXT: s_load_dwordx2 s[8:9], s[4:5], 0xd
|
|
; SI-NEXT: s_mov_b32 s7, 0xf000
|
|
; SI-NEXT: s_mov_b32 s6, -1
|
|
; SI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; SI-NEXT: s_mov_b32 s4, s0
|
|
; SI-NEXT: s_mov_b32 s5, s1
|
|
; SI-NEXT: s_mov_b32 s0, s2
|
|
; SI-NEXT: s_mov_b32 s1, s3
|
|
; SI-NEXT: s_mov_b32 s2, s6
|
|
; SI-NEXT: s_mov_b32 s3, s7
|
|
; SI-NEXT: s_mov_b32 s10, s6
|
|
; SI-NEXT: s_mov_b32 s11, s7
|
|
; SI-NEXT: buffer_load_dwordx4 v[0:3], off, s[0:3], 0
|
|
; SI-NEXT: buffer_load_dwordx4 v[4:7], off, s[8:11], 0 offset:64
|
|
; SI-NEXT: s_waitcnt vmcnt(0)
|
|
; SI-NEXT: v_div_scale_f64 v[8:9], s[0:1], v[6:7], v[6:7], v[2:3]
|
|
; SI-NEXT: v_rcp_f64_e32 v[10:11], v[8:9]
|
|
; SI-NEXT: v_fma_f64 v[12:13], -v[8:9], v[10:11], 1.0
|
|
; SI-NEXT: v_fma_f64 v[10:11], v[10:11], v[12:13], v[10:11]
|
|
; SI-NEXT: v_fma_f64 v[12:13], -v[8:9], v[10:11], 1.0
|
|
; SI-NEXT: v_fma_f64 v[10:11], v[10:11], v[12:13], v[10:11]
|
|
; SI-NEXT: v_div_scale_f64 v[12:13], s[0:1], v[2:3], v[6:7], v[2:3]
|
|
; SI-NEXT: v_mul_f64 v[14:15], v[12:13], v[10:11]
|
|
; SI-NEXT: v_fma_f64 v[16:17], -v[8:9], v[14:15], v[12:13]
|
|
; SI-NEXT: v_cmp_eq_u32_e32 vcc, v7, v9
|
|
; SI-NEXT: v_cmp_eq_u32_e64 s[0:1], v3, v13
|
|
; SI-NEXT: s_xor_b64 vcc, s[0:1], vcc
|
|
; SI-NEXT: s_nop 1
|
|
; SI-NEXT: v_div_fmas_f64 v[8:9], v[16:17], v[10:11], v[14:15]
|
|
; SI-NEXT: v_div_fixup_f64 v[8:9], v[8:9], v[6:7], v[2:3]
|
|
; SI-NEXT: v_readfirstlane_b32 s8, v9
|
|
; SI-NEXT: s_bfe_u32 s0, s8, 0xb0014
|
|
; SI-NEXT: s_add_i32 s9, s0, 0xfffffc01
|
|
; SI-NEXT: s_mov_b32 s3, 0xfffff
|
|
; SI-NEXT: s_lshr_b64 s[0:1], s[2:3], s9
|
|
; SI-NEXT: v_not_b32_e32 v10, s0
|
|
; SI-NEXT: v_and_b32_e32 v10, v8, v10
|
|
; SI-NEXT: v_not_b32_e32 v11, s1
|
|
; SI-NEXT: v_and_b32_e32 v9, v9, v11
|
|
; SI-NEXT: s_and_b32 s0, s8, 0x80000000
|
|
; SI-NEXT: s_cmp_lt_i32 s9, 0
|
|
; SI-NEXT: s_cselect_b64 vcc, -1, 0
|
|
; SI-NEXT: v_cndmask_b32_e64 v10, v10, 0, vcc
|
|
; SI-NEXT: v_mov_b32_e32 v11, s0
|
|
; SI-NEXT: v_cndmask_b32_e32 v9, v9, v11, vcc
|
|
; SI-NEXT: s_cmp_gt_i32 s9, 51
|
|
; SI-NEXT: s_cselect_b64 vcc, -1, 0
|
|
; SI-NEXT: v_mov_b32_e32 v11, s8
|
|
; SI-NEXT: v_cndmask_b32_e32 v9, v9, v11, vcc
|
|
; SI-NEXT: v_cndmask_b32_e32 v8, v10, v8, vcc
|
|
; SI-NEXT: v_fma_f64 v[2:3], -v[8:9], v[6:7], v[2:3]
|
|
; SI-NEXT: v_div_scale_f64 v[6:7], s[0:1], v[4:5], v[4:5], v[0:1]
|
|
; SI-NEXT: v_rcp_f64_e32 v[8:9], v[6:7]
|
|
; SI-NEXT: v_fma_f64 v[10:11], -v[6:7], v[8:9], 1.0
|
|
; SI-NEXT: v_fma_f64 v[8:9], v[8:9], v[10:11], v[8:9]
|
|
; SI-NEXT: v_fma_f64 v[10:11], -v[6:7], v[8:9], 1.0
|
|
; SI-NEXT: v_fma_f64 v[8:9], v[8:9], v[10:11], v[8:9]
|
|
; SI-NEXT: v_div_scale_f64 v[10:11], s[0:1], v[0:1], v[4:5], v[0:1]
|
|
; SI-NEXT: v_mul_f64 v[12:13], v[10:11], v[8:9]
|
|
; SI-NEXT: v_fma_f64 v[14:15], -v[6:7], v[12:13], v[10:11]
|
|
; SI-NEXT: v_cmp_eq_u32_e32 vcc, v5, v7
|
|
; SI-NEXT: v_cmp_eq_u32_e64 s[0:1], v1, v11
|
|
; SI-NEXT: s_xor_b64 vcc, s[0:1], vcc
|
|
; SI-NEXT: s_nop 1
|
|
; SI-NEXT: v_div_fmas_f64 v[6:7], v[14:15], v[8:9], v[12:13]
|
|
; SI-NEXT: v_div_fixup_f64 v[6:7], v[6:7], v[4:5], v[0:1]
|
|
; SI-NEXT: v_readfirstlane_b32 s8, v7
|
|
; SI-NEXT: s_bfe_u32 s0, s8, 0xb0014
|
|
; SI-NEXT: s_add_i32 s9, s0, 0xfffffc01
|
|
; SI-NEXT: s_lshr_b64 s[0:1], s[2:3], s9
|
|
; SI-NEXT: v_not_b32_e32 v8, s0
|
|
; SI-NEXT: v_and_b32_e32 v8, v6, v8
|
|
; SI-NEXT: v_not_b32_e32 v9, s1
|
|
; SI-NEXT: v_and_b32_e32 v7, v7, v9
|
|
; SI-NEXT: s_and_b32 s0, s8, 0x80000000
|
|
; SI-NEXT: s_cmp_lt_i32 s9, 0
|
|
; SI-NEXT: s_cselect_b64 vcc, -1, 0
|
|
; SI-NEXT: v_cndmask_b32_e64 v8, v8, 0, vcc
|
|
; SI-NEXT: v_mov_b32_e32 v9, s0
|
|
; SI-NEXT: v_cndmask_b32_e32 v7, v7, v9, vcc
|
|
; SI-NEXT: s_cmp_gt_i32 s9, 51
|
|
; SI-NEXT: s_cselect_b64 vcc, -1, 0
|
|
; SI-NEXT: v_mov_b32_e32 v9, s8
|
|
; SI-NEXT: v_cndmask_b32_e32 v7, v7, v9, vcc
|
|
; SI-NEXT: v_cndmask_b32_e32 v6, v8, v6, vcc
|
|
; SI-NEXT: v_fma_f64 v[0:1], -v[6:7], v[4:5], v[0:1]
|
|
; SI-NEXT: buffer_store_dwordx4 v[0:3], off, s[4:7], 0
|
|
; SI-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: frem_v2f64:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx4 s[8:11], s[4:5], 0x9
|
|
; CI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0xd
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: s_mov_b32 s6, s2
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_mov_b32 s0, s8
|
|
; CI-NEXT: s_mov_b32 s1, s9
|
|
; CI-NEXT: s_mov_b32 s8, s10
|
|
; CI-NEXT: s_mov_b32 s9, s11
|
|
; CI-NEXT: s_mov_b32 s10, s2
|
|
; CI-NEXT: s_mov_b32 s11, s3
|
|
; CI-NEXT: s_mov_b32 s7, s3
|
|
; CI-NEXT: buffer_load_dwordx4 v[0:3], off, s[8:11], 0
|
|
; CI-NEXT: buffer_load_dwordx4 v[4:7], off, s[4:7], 0 offset:64
|
|
; CI-NEXT: s_waitcnt vmcnt(0)
|
|
; CI-NEXT: v_div_scale_f64 v[8:9], s[4:5], v[6:7], v[6:7], v[2:3]
|
|
; CI-NEXT: v_rcp_f64_e32 v[10:11], v[8:9]
|
|
; CI-NEXT: v_fma_f64 v[12:13], -v[8:9], v[10:11], 1.0
|
|
; CI-NEXT: v_fma_f64 v[10:11], v[10:11], v[12:13], v[10:11]
|
|
; CI-NEXT: v_fma_f64 v[12:13], -v[8:9], v[10:11], 1.0
|
|
; CI-NEXT: v_fma_f64 v[10:11], v[10:11], v[12:13], v[10:11]
|
|
; CI-NEXT: v_div_scale_f64 v[12:13], vcc, v[2:3], v[6:7], v[2:3]
|
|
; CI-NEXT: v_mul_f64 v[14:15], v[12:13], v[10:11]
|
|
; CI-NEXT: v_fma_f64 v[8:9], -v[8:9], v[14:15], v[12:13]
|
|
; CI-NEXT: s_nop 1
|
|
; CI-NEXT: v_div_fmas_f64 v[8:9], v[8:9], v[10:11], v[14:15]
|
|
; CI-NEXT: v_div_fixup_f64 v[8:9], v[8:9], v[6:7], v[2:3]
|
|
; CI-NEXT: v_trunc_f64_e32 v[8:9], v[8:9]
|
|
; CI-NEXT: v_fma_f64 v[2:3], -v[8:9], v[6:7], v[2:3]
|
|
; CI-NEXT: v_div_scale_f64 v[6:7], s[4:5], v[4:5], v[4:5], v[0:1]
|
|
; CI-NEXT: v_rcp_f64_e32 v[8:9], v[6:7]
|
|
; CI-NEXT: v_fma_f64 v[10:11], -v[6:7], v[8:9], 1.0
|
|
; CI-NEXT: v_fma_f64 v[8:9], v[8:9], v[10:11], v[8:9]
|
|
; CI-NEXT: v_fma_f64 v[10:11], -v[6:7], v[8:9], 1.0
|
|
; CI-NEXT: v_fma_f64 v[8:9], v[8:9], v[10:11], v[8:9]
|
|
; CI-NEXT: v_div_scale_f64 v[10:11], vcc, v[0:1], v[4:5], v[0:1]
|
|
; CI-NEXT: v_mul_f64 v[12:13], v[10:11], v[8:9]
|
|
; CI-NEXT: v_fma_f64 v[6:7], -v[6:7], v[12:13], v[10:11]
|
|
; CI-NEXT: s_nop 1
|
|
; CI-NEXT: v_div_fmas_f64 v[6:7], v[6:7], v[8:9], v[12:13]
|
|
; CI-NEXT: v_div_fixup_f64 v[6:7], v[6:7], v[4:5], v[0:1]
|
|
; CI-NEXT: v_trunc_f64_e32 v[6:7], v[6:7]
|
|
; CI-NEXT: v_fma_f64 v[0:1], -v[6:7], v[4:5], v[0:1]
|
|
; CI-NEXT: buffer_store_dwordx4 v[0:3], off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; VI-LABEL: frem_v2f64:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; VI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x34
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v8, s0
|
|
; VI-NEXT: s_add_u32 s0, s4, 64
|
|
; VI-NEXT: v_mov_b32_e32 v9, s1
|
|
; VI-NEXT: s_addc_u32 s1, s5, 0
|
|
; VI-NEXT: v_mov_b32_e32 v5, s1
|
|
; VI-NEXT: v_mov_b32_e32 v0, s2
|
|
; VI-NEXT: v_mov_b32_e32 v1, s3
|
|
; VI-NEXT: v_mov_b32_e32 v4, s0
|
|
; VI-NEXT: flat_load_dwordx4 v[0:3], v[0:1]
|
|
; VI-NEXT: flat_load_dwordx4 v[4:7], v[4:5]
|
|
; VI-NEXT: s_waitcnt vmcnt(0)
|
|
; VI-NEXT: v_div_scale_f64 v[10:11], s[0:1], v[6:7], v[6:7], v[2:3]
|
|
; VI-NEXT: v_rcp_f64_e32 v[12:13], v[10:11]
|
|
; VI-NEXT: v_fma_f64 v[14:15], -v[10:11], v[12:13], 1.0
|
|
; VI-NEXT: v_fma_f64 v[12:13], v[12:13], v[14:15], v[12:13]
|
|
; VI-NEXT: v_fma_f64 v[14:15], -v[10:11], v[12:13], 1.0
|
|
; VI-NEXT: v_fma_f64 v[12:13], v[12:13], v[14:15], v[12:13]
|
|
; VI-NEXT: v_div_scale_f64 v[14:15], vcc, v[2:3], v[6:7], v[2:3]
|
|
; VI-NEXT: v_mul_f64 v[16:17], v[14:15], v[12:13]
|
|
; VI-NEXT: v_fma_f64 v[10:11], -v[10:11], v[16:17], v[14:15]
|
|
; VI-NEXT: s_nop 1
|
|
; VI-NEXT: v_div_fmas_f64 v[10:11], v[10:11], v[12:13], v[16:17]
|
|
; VI-NEXT: v_div_fixup_f64 v[10:11], v[10:11], v[6:7], v[2:3]
|
|
; VI-NEXT: v_trunc_f64_e32 v[10:11], v[10:11]
|
|
; VI-NEXT: v_fma_f64 v[2:3], -v[10:11], v[6:7], v[2:3]
|
|
; VI-NEXT: v_div_scale_f64 v[6:7], s[0:1], v[4:5], v[4:5], v[0:1]
|
|
; VI-NEXT: v_rcp_f64_e32 v[10:11], v[6:7]
|
|
; VI-NEXT: v_fma_f64 v[12:13], -v[6:7], v[10:11], 1.0
|
|
; VI-NEXT: v_fma_f64 v[10:11], v[10:11], v[12:13], v[10:11]
|
|
; VI-NEXT: v_fma_f64 v[12:13], -v[6:7], v[10:11], 1.0
|
|
; VI-NEXT: v_fma_f64 v[10:11], v[10:11], v[12:13], v[10:11]
|
|
; VI-NEXT: v_div_scale_f64 v[12:13], vcc, v[0:1], v[4:5], v[0:1]
|
|
; VI-NEXT: v_mul_f64 v[14:15], v[12:13], v[10:11]
|
|
; VI-NEXT: v_fma_f64 v[6:7], -v[6:7], v[14:15], v[12:13]
|
|
; VI-NEXT: s_nop 1
|
|
; VI-NEXT: v_div_fmas_f64 v[6:7], v[6:7], v[10:11], v[14:15]
|
|
; VI-NEXT: v_div_fixup_f64 v[6:7], v[6:7], v[4:5], v[0:1]
|
|
; VI-NEXT: v_trunc_f64_e32 v[6:7], v[6:7]
|
|
; VI-NEXT: v_fma_f64 v[0:1], -v[6:7], v[4:5], v[0:1]
|
|
; VI-NEXT: flat_store_dwordx4 v[8:9], v[0:3]
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: frem_v2f64:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; GFX9-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
|
|
; GFX9-NEXT: v_mov_b32_e32 v16, 0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_load_dwordx4 v[0:3], v16, s[2:3]
|
|
; GFX9-NEXT: global_load_dwordx4 v[4:7], v16, s[6:7] offset:64
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX9-NEXT: v_div_scale_f64 v[8:9], s[2:3], v[6:7], v[6:7], v[2:3]
|
|
; GFX9-NEXT: v_rcp_f64_e32 v[10:11], v[8:9]
|
|
; GFX9-NEXT: v_fma_f64 v[12:13], -v[8:9], v[10:11], 1.0
|
|
; GFX9-NEXT: v_fma_f64 v[10:11], v[10:11], v[12:13], v[10:11]
|
|
; GFX9-NEXT: v_fma_f64 v[12:13], -v[8:9], v[10:11], 1.0
|
|
; GFX9-NEXT: v_fma_f64 v[10:11], v[10:11], v[12:13], v[10:11]
|
|
; GFX9-NEXT: v_div_scale_f64 v[12:13], vcc, v[2:3], v[6:7], v[2:3]
|
|
; GFX9-NEXT: v_mul_f64 v[14:15], v[12:13], v[10:11]
|
|
; GFX9-NEXT: v_fma_f64 v[8:9], -v[8:9], v[14:15], v[12:13]
|
|
; GFX9-NEXT: s_nop 1
|
|
; GFX9-NEXT: v_div_fmas_f64 v[8:9], v[8:9], v[10:11], v[14:15]
|
|
; GFX9-NEXT: v_div_fixup_f64 v[8:9], v[8:9], v[6:7], v[2:3]
|
|
; GFX9-NEXT: v_trunc_f64_e32 v[8:9], v[8:9]
|
|
; GFX9-NEXT: v_fma_f64 v[2:3], -v[8:9], v[6:7], v[2:3]
|
|
; GFX9-NEXT: v_div_scale_f64 v[6:7], s[2:3], v[4:5], v[4:5], v[0:1]
|
|
; GFX9-NEXT: v_rcp_f64_e32 v[8:9], v[6:7]
|
|
; GFX9-NEXT: v_fma_f64 v[10:11], -v[6:7], v[8:9], 1.0
|
|
; GFX9-NEXT: v_fma_f64 v[8:9], v[8:9], v[10:11], v[8:9]
|
|
; GFX9-NEXT: v_fma_f64 v[10:11], -v[6:7], v[8:9], 1.0
|
|
; GFX9-NEXT: v_fma_f64 v[8:9], v[8:9], v[10:11], v[8:9]
|
|
; GFX9-NEXT: v_div_scale_f64 v[10:11], vcc, v[0:1], v[4:5], v[0:1]
|
|
; GFX9-NEXT: v_mul_f64 v[12:13], v[10:11], v[8:9]
|
|
; GFX9-NEXT: v_fma_f64 v[6:7], -v[6:7], v[12:13], v[10:11]
|
|
; GFX9-NEXT: s_nop 1
|
|
; GFX9-NEXT: v_div_fmas_f64 v[6:7], v[6:7], v[8:9], v[12:13]
|
|
; GFX9-NEXT: v_div_fixup_f64 v[6:7], v[6:7], v[4:5], v[0:1]
|
|
; GFX9-NEXT: v_trunc_f64_e32 v[6:7], v[6:7]
|
|
; GFX9-NEXT: v_fma_f64 v[0:1], -v[6:7], v[4:5], v[0:1]
|
|
; GFX9-NEXT: global_store_dwordx4 v16, v[0:3], s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; GFX10-LABEL: frem_v2f64:
|
|
; GFX10: ; %bb.0:
|
|
; GFX10-NEXT: s_clause 0x1
|
|
; GFX10-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
|
|
; GFX10-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
|
|
; GFX10-NEXT: v_mov_b32_e32 v16, 0
|
|
; GFX10-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-NEXT: s_clause 0x1
|
|
; GFX10-NEXT: global_load_dwordx4 v[0:3], v16, s[2:3]
|
|
; GFX10-NEXT: global_load_dwordx4 v[4:7], v16, s[6:7] offset:64
|
|
; GFX10-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX10-NEXT: v_div_scale_f64 v[8:9], s2, v[6:7], v[6:7], v[2:3]
|
|
; GFX10-NEXT: v_rcp_f64_e32 v[10:11], v[8:9]
|
|
; GFX10-NEXT: v_fma_f64 v[12:13], -v[8:9], v[10:11], 1.0
|
|
; GFX10-NEXT: v_fma_f64 v[10:11], v[10:11], v[12:13], v[10:11]
|
|
; GFX10-NEXT: v_fma_f64 v[12:13], -v[8:9], v[10:11], 1.0
|
|
; GFX10-NEXT: v_fma_f64 v[10:11], v[10:11], v[12:13], v[10:11]
|
|
; GFX10-NEXT: v_div_scale_f64 v[12:13], vcc_lo, v[2:3], v[6:7], v[2:3]
|
|
; GFX10-NEXT: v_mul_f64 v[14:15], v[12:13], v[10:11]
|
|
; GFX10-NEXT: v_fma_f64 v[8:9], -v[8:9], v[14:15], v[12:13]
|
|
; GFX10-NEXT: v_div_fmas_f64 v[8:9], v[8:9], v[10:11], v[14:15]
|
|
; GFX10-NEXT: v_div_fixup_f64 v[8:9], v[8:9], v[6:7], v[2:3]
|
|
; GFX10-NEXT: v_trunc_f64_e32 v[8:9], v[8:9]
|
|
; GFX10-NEXT: v_fma_f64 v[2:3], -v[8:9], v[6:7], v[2:3]
|
|
; GFX10-NEXT: v_div_scale_f64 v[6:7], s2, v[4:5], v[4:5], v[0:1]
|
|
; GFX10-NEXT: v_rcp_f64_e32 v[8:9], v[6:7]
|
|
; GFX10-NEXT: v_fma_f64 v[10:11], -v[6:7], v[8:9], 1.0
|
|
; GFX10-NEXT: v_fma_f64 v[8:9], v[8:9], v[10:11], v[8:9]
|
|
; GFX10-NEXT: v_fma_f64 v[10:11], -v[6:7], v[8:9], 1.0
|
|
; GFX10-NEXT: v_fma_f64 v[8:9], v[8:9], v[10:11], v[8:9]
|
|
; GFX10-NEXT: v_div_scale_f64 v[10:11], vcc_lo, v[0:1], v[4:5], v[0:1]
|
|
; GFX10-NEXT: v_mul_f64 v[12:13], v[10:11], v[8:9]
|
|
; GFX10-NEXT: v_fma_f64 v[6:7], -v[6:7], v[12:13], v[10:11]
|
|
; GFX10-NEXT: v_div_fmas_f64 v[6:7], v[6:7], v[8:9], v[12:13]
|
|
; GFX10-NEXT: v_div_fixup_f64 v[6:7], v[6:7], v[4:5], v[0:1]
|
|
; GFX10-NEXT: v_trunc_f64_e32 v[6:7], v[6:7]
|
|
; GFX10-NEXT: v_fma_f64 v[0:1], -v[6:7], v[4:5], v[0:1]
|
|
; GFX10-NEXT: global_store_dwordx4 v16, v[0:3], s[0:1]
|
|
; GFX10-NEXT: s_endpgm
|
|
;
|
|
; GFX11-LABEL: frem_v2f64:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_clause 0x1
|
|
; GFX11-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX11-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX11-NEXT: v_mov_b32_e32 v16, 0
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: s_clause 0x1
|
|
; GFX11-NEXT: global_load_b128 v[0:3], v16, s[2:3]
|
|
; GFX11-NEXT: global_load_b128 v[4:7], v16, s[4:5] offset:64
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-NEXT: v_div_scale_f64 v[8:9], null, v[6:7], v[6:7], v[2:3]
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_rcp_f64_e32 v[10:11], v[8:9]
|
|
; GFX11-NEXT: s_waitcnt_depctr 0xfff
|
|
; GFX11-NEXT: v_fma_f64 v[12:13], -v[8:9], v[10:11], 1.0
|
|
; GFX11-NEXT: v_fma_f64 v[10:11], v[10:11], v[12:13], v[10:11]
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_fma_f64 v[12:13], -v[8:9], v[10:11], 1.0
|
|
; GFX11-NEXT: v_fma_f64 v[10:11], v[10:11], v[12:13], v[10:11]
|
|
; GFX11-NEXT: v_div_scale_f64 v[12:13], vcc_lo, v[2:3], v[6:7], v[2:3]
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_mul_f64 v[14:15], v[12:13], v[10:11]
|
|
; GFX11-NEXT: v_fma_f64 v[8:9], -v[8:9], v[14:15], v[12:13]
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_div_fmas_f64 v[8:9], v[8:9], v[10:11], v[14:15]
|
|
; GFX11-NEXT: v_div_fixup_f64 v[8:9], v[8:9], v[6:7], v[2:3]
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_trunc_f64_e32 v[8:9], v[8:9]
|
|
; GFX11-NEXT: v_fma_f64 v[2:3], -v[8:9], v[6:7], v[2:3]
|
|
; GFX11-NEXT: v_div_scale_f64 v[6:7], null, v[4:5], v[4:5], v[0:1]
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_rcp_f64_e32 v[8:9], v[6:7]
|
|
; GFX11-NEXT: s_waitcnt_depctr 0xfff
|
|
; GFX11-NEXT: v_fma_f64 v[10:11], -v[6:7], v[8:9], 1.0
|
|
; GFX11-NEXT: v_fma_f64 v[8:9], v[8:9], v[10:11], v[8:9]
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_fma_f64 v[10:11], -v[6:7], v[8:9], 1.0
|
|
; GFX11-NEXT: v_fma_f64 v[8:9], v[8:9], v[10:11], v[8:9]
|
|
; GFX11-NEXT: v_div_scale_f64 v[10:11], vcc_lo, v[0:1], v[4:5], v[0:1]
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_mul_f64 v[12:13], v[10:11], v[8:9]
|
|
; GFX11-NEXT: v_fma_f64 v[6:7], -v[6:7], v[12:13], v[10:11]
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_div_fmas_f64 v[6:7], v[6:7], v[8:9], v[12:13]
|
|
; GFX11-NEXT: v_div_fixup_f64 v[6:7], v[6:7], v[4:5], v[0:1]
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_trunc_f64_e32 v[6:7], v[6:7]
|
|
; GFX11-NEXT: v_fma_f64 v[0:1], -v[6:7], v[4:5], v[0:1]
|
|
; GFX11-NEXT: global_store_b128 v16, v[0:3], s[0:1]
|
|
; GFX11-NEXT: s_endpgm
|
|
;
|
|
; GFX1150-LABEL: frem_v2f64:
|
|
; GFX1150: ; %bb.0:
|
|
; GFX1150-NEXT: s_clause 0x1
|
|
; GFX1150-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
|
|
; GFX1150-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
|
|
; GFX1150-NEXT: v_mov_b32_e32 v16, 0
|
|
; GFX1150-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX1150-NEXT: s_clause 0x1
|
|
; GFX1150-NEXT: global_load_b128 v[0:3], v16, s[2:3]
|
|
; GFX1150-NEXT: global_load_b128 v[4:7], v16, s[4:5] offset:64
|
|
; GFX1150-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX1150-NEXT: v_div_scale_f64 v[8:9], null, v[6:7], v[6:7], v[2:3]
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(TRANS32_DEP_1)
|
|
; GFX1150-NEXT: v_rcp_f64_e32 v[10:11], v[8:9]
|
|
; GFX1150-NEXT: v_fma_f64 v[12:13], -v[8:9], v[10:11], 1.0
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_f64 v[10:11], v[10:11], v[12:13], v[10:11]
|
|
; GFX1150-NEXT: v_fma_f64 v[12:13], -v[8:9], v[10:11], 1.0
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_f64 v[10:11], v[10:11], v[12:13], v[10:11]
|
|
; GFX1150-NEXT: v_div_scale_f64 v[12:13], vcc_lo, v[2:3], v[6:7], v[2:3]
|
|
; GFX1150-NEXT: v_mul_f64 v[14:15], v[12:13], v[10:11]
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_f64 v[8:9], -v[8:9], v[14:15], v[12:13]
|
|
; GFX1150-NEXT: v_div_fmas_f64 v[8:9], v[8:9], v[10:11], v[14:15]
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_div_fixup_f64 v[8:9], v[8:9], v[6:7], v[2:3]
|
|
; GFX1150-NEXT: v_trunc_f64_e32 v[8:9], v[8:9]
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_f64 v[2:3], -v[8:9], v[6:7], v[2:3]
|
|
; GFX1150-NEXT: v_div_scale_f64 v[6:7], null, v[4:5], v[4:5], v[0:1]
|
|
; GFX1150-NEXT: v_rcp_f64_e32 v[8:9], v[6:7]
|
|
; GFX1150-NEXT: s_delay_alu instid0(TRANS32_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_f64 v[10:11], -v[6:7], v[8:9], 1.0
|
|
; GFX1150-NEXT: v_fma_f64 v[8:9], v[8:9], v[10:11], v[8:9]
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_fma_f64 v[10:11], -v[6:7], v[8:9], 1.0
|
|
; GFX1150-NEXT: v_fma_f64 v[8:9], v[8:9], v[10:11], v[8:9]
|
|
; GFX1150-NEXT: v_div_scale_f64 v[10:11], vcc_lo, v[0:1], v[4:5], v[0:1]
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_mul_f64 v[12:13], v[10:11], v[8:9]
|
|
; GFX1150-NEXT: v_fma_f64 v[6:7], -v[6:7], v[12:13], v[10:11]
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_div_fmas_f64 v[6:7], v[6:7], v[8:9], v[12:13]
|
|
; GFX1150-NEXT: v_div_fixup_f64 v[6:7], v[6:7], v[4:5], v[0:1]
|
|
; GFX1150-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
|
|
; GFX1150-NEXT: v_trunc_f64_e32 v[6:7], v[6:7]
|
|
; GFX1150-NEXT: v_fma_f64 v[0:1], -v[6:7], v[4:5], v[0:1]
|
|
; GFX1150-NEXT: global_store_b128 v16, v[0:3], s[0:1]
|
|
; GFX1150-NEXT: s_endpgm
|
|
ptr addrspace(1) %in2) #0 {
|
|
%gep2 = getelementptr <2 x double>, ptr addrspace(1) %in2, i32 4
|
|
%r0 = load <2 x double>, ptr addrspace(1) %in1, align 16
|
|
%r1 = load <2 x double>, ptr addrspace(1) %gep2, align 16
|
|
%r2 = frem <2 x double> %r0, %r1
|
|
store <2 x double> %r2, ptr addrspace(1) %out, align 16
|
|
ret void
|
|
}
|
|
|
|
attributes #0 = { nounwind "unsafe-fp-math"="false" "denormal-fp-math-f32"="preserve-sign,preserve-sign" }
|
|
attributes #1 = { nounwind "unsafe-fp-math"="true" "denormal-fp-math-f32"="preserve-sign,preserve-sign" }
|