RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-22 18:54:02 +01:00
Code Issues Projects Releases Wiki Activity
88aa158bd7
llvm-mirror/test/CodeGen/AMDGPU/GlobalISel/inst-select-fminnum-ieee.mir
Stanislav Mekhanoshin 196e7f3138 [AMDGPU] Use single cache policy operand 
Replace individual operands GLC, SLC, and DLC with a single cache_policy
bitmask operand. This will reduce the number of operands in MIR and I hope
the amount of code. These operands are mostly 0 anyway.

Additional advantage that parser will accept these flags in any order unlike
now.

Differential Revision: https://reviews.llvm.org/D96469
2021-03-15 13:00:59 -07:00

137 lines
5.4 KiB
YAML
Raw Blame History

# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=hawaii -run-pass=instruction-select -verify-machineinstrs -o - %s | FileCheck -check-prefix=GFX7 %s
---
name: fminnum_ieee_f32_f64_ieee_mode_on
legalized: true
regBankSelected: true
machineFunctionInfo:
mode:
ieee: true
body: |
bb.0:
liveins: $sgpr0, $vgpr0, $vgpr1, $vgpr3_vgpr4, $sgpr10_sgpr11, $vgpr10_vgpr11, $vgpr12_vgpr13
; GFX7-LABEL: name: fminnum_ieee_f32_f64_ieee_mode_on
; GFX7: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
; GFX7: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX7: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX7: [[COPY3:%[0-9]+]]:vreg_64 = COPY $vgpr3_vgpr4
; GFX7: [[COPY4:%[0-9]+]]:sreg_64 = COPY $sgpr10_sgpr11
; GFX7: [[COPY5:%[0-9]+]]:vreg_64 = COPY $vgpr10_vgpr11
; GFX7: [[COPY6:%[0-9]+]]:vreg_64 = COPY $vgpr12_vgpr13
; GFX7: %7:vgpr_32 = nofpexcept V_MIN_F32_e64 0, [[COPY1]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
; GFX7: %8:vgpr_32 = nofpexcept V_MIN_F32_e64 0, [[COPY]], 0, [[COPY1]], 0, 0, implicit $mode, implicit $exec
; GFX7: %9:vgpr_32 = nofpexcept V_MIN_F32_e64 0, [[COPY1]], 0, [[COPY2]], 0, 0, implicit $mode, implicit $exec
; GFX7: FLAT_STORE_DWORD [[COPY3]], %7, 0, 0, implicit $exec, implicit $flat_scr :: (store 4, addrspace 1)
; GFX7: FLAT_STORE_DWORD [[COPY3]], %8, 0, 0, implicit $exec, implicit $flat_scr :: (store 4, addrspace 1)
; GFX7: FLAT_STORE_DWORD [[COPY3]], %9, 0, 0, implicit $exec, implicit $flat_scr :: (store 4, addrspace 1)
; GFX7: %10:vreg_64 = nofpexcept V_MIN_F64_e64 0, [[COPY4]], 0, [[COPY5]], 0, 0, implicit $mode, implicit $exec
; GFX7: %11:vreg_64 = nofpexcept V_MIN_F64_e64 0, [[COPY5]], 0, [[COPY4]], 0, 0, implicit $mode, implicit $exec
; GFX7: %12:vreg_64 = nofpexcept V_MIN_F64_e64 0, [[COPY5]], 0, [[COPY6]], 0, 0, implicit $mode, implicit $exec
; GFX7: S_ENDPGM 0, implicit %10, implicit %11, implicit %12
%0:sgpr(s32) = COPY $sgpr0
%1:vgpr(s32) = COPY $vgpr0
%2:vgpr(s32) = COPY $vgpr1
%3:vgpr(p1) = COPY $vgpr3_vgpr4
%10:sgpr(s64) = COPY $sgpr10_sgpr11
%11:vgpr(s64) = COPY $vgpr10_vgpr11
%12:vgpr(s64) = COPY $vgpr12_vgpr13
; minnum_ieee vs
%4:vgpr(s32) = G_FMINNUM_IEEE %1, %0
; minnum_ieee sv
%5:vgpr(s32) = G_FMINNUM_IEEE %0, %1
; minnum_ieee vv
%6:vgpr(s32) = G_FMINNUM_IEEE %1, %2
G_STORE %4, %3 :: (store 4, addrspace 1)
G_STORE %5, %3 :: (store 4, addrspace 1)
G_STORE %6, %3 :: (store 4, addrspace 1)
; 64-bit
; minnum_ieee vs
%14:vgpr(s64) = G_FMINNUM_IEEE %10, %11
; minnum_ieee sv
%15:vgpr(s64) = G_FMINNUM_IEEE %11, %10
; minnum_ieee vv
%16:vgpr(s64) = G_FMINNUM_IEEE %11, %12
S_ENDPGM 0, implicit %14, implicit %15, implicit %16
...
# FIXME: Ideally this would fail to select with ieee mode disabled
---
name: fminnum_ieee_f32_f64_ieee_mode_off
legalized: true
regBankSelected: true
machineFunctionInfo:
mode:
ieee: false
body: |
bb.0:
liveins: $sgpr0, $vgpr0, $vgpr1, $vgpr3_vgpr4, $sgpr10_sgpr11, $vgpr10_vgpr11, $vgpr12_vgpr13
; GFX7-LABEL: name: fminnum_ieee_f32_f64_ieee_mode_off
; GFX7: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
; GFX7: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX7: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX7: [[COPY3:%[0-9]+]]:vreg_64 = COPY $vgpr3_vgpr4
; GFX7: [[COPY4:%[0-9]+]]:sreg_64 = COPY $sgpr10_sgpr11
; GFX7: [[COPY5:%[0-9]+]]:vreg_64 = COPY $vgpr10_vgpr11
; GFX7: [[COPY6:%[0-9]+]]:vreg_64 = COPY $vgpr12_vgpr13
; GFX7: %7:vgpr_32 = nofpexcept V_MIN_F32_e64 0, [[COPY1]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
; GFX7: %8:vgpr_32 = nofpexcept V_MIN_F32_e64 0, [[COPY]], 0, [[COPY1]], 0, 0, implicit $mode, implicit $exec
; GFX7: %9:vgpr_32 = nofpexcept V_MIN_F32_e64 0, [[COPY1]], 0, [[COPY2]], 0, 0, implicit $mode, implicit $exec
; GFX7: FLAT_STORE_DWORD [[COPY3]], %7, 0, 0, implicit $exec, implicit $flat_scr :: (store 4, addrspace 1)
; GFX7: FLAT_STORE_DWORD [[COPY3]], %8, 0, 0, implicit $exec, implicit $flat_scr :: (store 4, addrspace 1)
; GFX7: FLAT_STORE_DWORD [[COPY3]], %9, 0, 0, implicit $exec, implicit $flat_scr :: (store 4, addrspace 1)
; GFX7: %10:vreg_64 = nofpexcept V_MIN_F64_e64 0, [[COPY4]], 0, [[COPY5]], 0, 0, implicit $mode, implicit $exec
; GFX7: %11:vreg_64 = nofpexcept V_MIN_F64_e64 0, [[COPY5]], 0, [[COPY4]], 0, 0, implicit $mode, implicit $exec
; GFX7: %12:vreg_64 = nofpexcept V_MIN_F64_e64 0, [[COPY5]], 0, [[COPY6]], 0, 0, implicit $mode, implicit $exec
; GFX7: S_ENDPGM 0, implicit %10, implicit %11, implicit %12
%0:sgpr(s32) = COPY $sgpr0
%1:vgpr(s32) = COPY $vgpr0
%2:vgpr(s32) = COPY $vgpr1
%3:vgpr(p1) = COPY $vgpr3_vgpr4
%10:sgpr(s64) = COPY $sgpr10_sgpr11
%11:vgpr(s64) = COPY $vgpr10_vgpr11
%12:vgpr(s64) = COPY $vgpr12_vgpr13
; minnum_ieee vs
%4:vgpr(s32) = G_FMINNUM_IEEE %1, %0
; minnum_ieee sv
%5:vgpr(s32) = G_FMINNUM_IEEE %0, %1
; minnum_ieee vv
%6:vgpr(s32) = G_FMINNUM_IEEE %1, %2
G_STORE %4, %3 :: (store 4, addrspace 1)
G_STORE %5, %3 :: (store 4, addrspace 1)
G_STORE %6, %3 :: (store 4, addrspace 1)
; 64-bit
; minnum_ieee vs
%14:vgpr(s64) = G_FMINNUM_IEEE %10, %11
; minnum_ieee sv
%15:vgpr(s64) = G_FMINNUM_IEEE %11, %10
; minnum_ieee vv
%16:vgpr(s64) = G_FMINNUM_IEEE %11, %12
S_ENDPGM 0, implicit %14, implicit %15, implicit %16
...
Reference in New Issue View Git Blame Copy Permalink