blob: cba7d41ea815f094f9132588f04b18d917c308e0 [file] [edit]
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 6
# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -mattr=+wavefrontsize32 --run-pass=si-fix-sgpr-copies,livevars,phi-node-elimination -o - %s | FileCheck %s --check-prefixes=W32
# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -mattr=+wavefrontsize64 --run-pass=si-fix-sgpr-copies,livevars,phi-node-elimination -o - %s | FileCheck %s --check-prefixes=W64
# The waterfall loop computes the new EXEC value into a virtual register and
# writes EXEC from a terminator. PHI elimination therefore places the copy for
# the loop-carried lane mask before the EXEC write, while EXEC is still full.
---
name: waterfall_exec_update
tracksRegLiveness: true
liveins:
- { reg: '$vgpr0', virtual-reg: '%0' }
- { reg: '$vgpr1', virtual-reg: '%1' }
- { reg: '$vgpr2', virtual-reg: '%2' }
- { reg: '$vgpr3', virtual-reg: '%3' }
- { reg: '$vgpr4', virtual-reg: '%4' }
body: |
bb.0:
liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
; W32-LABEL: name: waterfall_exec_update
; W32: successors: %bb.1(0x80000000)
; W32-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
; W32-NEXT: {{ $}}
; W32-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY killed $vgpr4
; W32-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY killed $vgpr3
; W32-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY killed $vgpr2
; W32-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY killed $vgpr1
; W32-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY killed $vgpr0
; W32-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE killed [[COPY4]], %subreg.sub0, killed [[COPY3]], %subreg.sub1, killed [[COPY2]], %subreg.sub2, killed [[COPY1]], %subreg.sub3
; W32-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_MOV_B32 $exec_lo
; W32-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32_xm0_xexec = S_MOV_B32 $exec_lo
; W32-NEXT: [[COPY5:%[0-9]+]]:sreg_32_xm0_xexec = COPY killed [[S_MOV_B32_1]]
; W32-NEXT: {{ $}}
; W32-NEXT: .1:
; W32-NEXT: successors: %bb.2(0x80000000)
; W32-NEXT: {{ $}}
; W32-NEXT: [[COPY6:%[0-9]+]]:sreg_32_xm0_xexec = COPY killed [[COPY5]]
; W32-NEXT: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
; W32-NEXT: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
; W32-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
; W32-NEXT: [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
; W32-NEXT: [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
; W32-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
; W32-NEXT: [[REG_SEQUENCE3:%[0-9]+]]:sgpr_128 = REG_SEQUENCE killed [[V_READFIRSTLANE_B32_]], %subreg.sub0, killed [[V_READFIRSTLANE_B32_1]], %subreg.sub1, killed [[V_READFIRSTLANE_B32_2]], %subreg.sub2, killed [[V_READFIRSTLANE_B32_3]], %subreg.sub3
; W32-NEXT: V_CMPX_EQ_U64_nosdst_e32_term killed [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit-def $exec, implicit $exec
; W32-NEXT: V_CMPX_EQ_U64_nosdst_e32_term killed [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit-def $exec, implicit $exec
; W32-NEXT: {{ $}}
; W32-NEXT: .2:
; W32-NEXT: successors: %bb.1(0x40000000), %bb.3(0x40000000)
; W32-NEXT: {{ $}}
; W32-NEXT: [[BUFFER_LOAD_FORMAT_X_IDXEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_FORMAT_X_IDXEN [[COPY]], killed [[REG_SEQUENCE3]], 0, 0, 0, 0, implicit $exec
; W32-NEXT: [[S_ANDN2_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_ANDN2_B32 killed [[COPY6]], $exec_lo, implicit-def dead $scc
; W32-NEXT: [[COPY5:%[0-9]+]]:sreg_32_xm0_xexec = COPY [[S_ANDN2_B32_]]
; W32-NEXT: $exec_lo = S_MOV_B32_term killed [[S_ANDN2_B32_]]
; W32-NEXT: SI_WATERFALL_LOOP %bb.1, implicit $exec
; W32-NEXT: {{ $}}
; W32-NEXT: .3:
; W32-NEXT: $exec_lo = S_MOV_B32 killed [[S_MOV_B32_]]
; W32-NEXT: $vgpr0 = COPY killed [[BUFFER_LOAD_FORMAT_X_IDXEN]]
; W32-NEXT: S_ENDPGM 0, implicit killed $vgpr0
;
; W64-LABEL: name: waterfall_exec_update
; W64: successors: %bb.1(0x80000000)
; W64-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
; W64-NEXT: {{ $}}
; W64-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY killed $vgpr4
; W64-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY killed $vgpr3
; W64-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY killed $vgpr2
; W64-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY killed $vgpr1
; W64-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY killed $vgpr0
; W64-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE killed [[COPY4]], %subreg.sub0, killed [[COPY3]], %subreg.sub1, killed [[COPY2]], %subreg.sub2, killed [[COPY1]], %subreg.sub3
; W64-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
; W64-NEXT: [[S_MOV_B64_1:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
; W64-NEXT: [[COPY5:%[0-9]+]]:sreg_64_xexec = COPY killed [[S_MOV_B64_1]]
; W64-NEXT: {{ $}}
; W64-NEXT: .1:
; W64-NEXT: successors: %bb.2(0x80000000)
; W64-NEXT: {{ $}}
; W64-NEXT: [[COPY6:%[0-9]+]]:sreg_64_xexec = COPY killed [[COPY5]]
; W64-NEXT: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
; W64-NEXT: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
; W64-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
; W64-NEXT: [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
; W64-NEXT: [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
; W64-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
; W64-NEXT: [[REG_SEQUENCE3:%[0-9]+]]:sgpr_128 = REG_SEQUENCE killed [[V_READFIRSTLANE_B32_]], %subreg.sub0, killed [[V_READFIRSTLANE_B32_1]], %subreg.sub1, killed [[V_READFIRSTLANE_B32_2]], %subreg.sub2, killed [[V_READFIRSTLANE_B32_3]], %subreg.sub3
; W64-NEXT: V_CMPX_EQ_U64_nosdst_e32_term killed [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit-def $exec, implicit $exec
; W64-NEXT: V_CMPX_EQ_U64_nosdst_e32_term killed [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit-def $exec, implicit $exec
; W64-NEXT: {{ $}}
; W64-NEXT: .2:
; W64-NEXT: successors: %bb.1(0x40000000), %bb.3(0x40000000)
; W64-NEXT: {{ $}}
; W64-NEXT: [[BUFFER_LOAD_FORMAT_X_IDXEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_FORMAT_X_IDXEN [[COPY]], killed [[REG_SEQUENCE3]], 0, 0, 0, 0, implicit $exec
; W64-NEXT: [[S_ANDN2_B64_:%[0-9]+]]:sreg_64_xexec = S_ANDN2_B64 killed [[COPY6]], $exec, implicit-def dead $scc
; W64-NEXT: [[COPY5:%[0-9]+]]:sreg_64_xexec = COPY [[S_ANDN2_B64_]]
; W64-NEXT: $exec = S_MOV_B64_term killed [[S_ANDN2_B64_]]
; W64-NEXT: SI_WATERFALL_LOOP %bb.1, implicit $exec
; W64-NEXT: {{ $}}
; W64-NEXT: .3:
; W64-NEXT: $exec = S_MOV_B64 killed [[S_MOV_B64_]]
; W64-NEXT: $vgpr0 = COPY killed [[BUFFER_LOAD_FORMAT_X_IDXEN]]
; W64-NEXT: S_ENDPGM 0, implicit killed $vgpr0
%4:vgpr_32 = COPY $vgpr4
%3:vgpr_32 = COPY $vgpr3
%2:vgpr_32 = COPY $vgpr2
%1:vgpr_32 = COPY $vgpr1
%0:vgpr_32 = COPY $vgpr0
%5:sgpr_128 = REG_SEQUENCE %0, %subreg.sub0, %1, %subreg.sub1, %2, %subreg.sub2, %3, %subreg.sub3
%6:vgpr_32 = BUFFER_LOAD_FORMAT_X_IDXEN %4, killed %5, 0, 0, 0, 0, implicit $exec
$vgpr0 = COPY %6
S_ENDPGM 0, implicit $vgpr0
...