5231736329
M0 can only be written to by the SALU, so `v_readfirstlane_b32 m0` is effectively useless. Represent this by restricting the dest RC of that instruction to `SReg_32_XM0` which excludes M0. There is a lot of test changes due to the register class changing, but most changes are trivial. In some cases, an extra register and `s_mov_b32` is needed. Fixes SWDEV-513269
57 lines
3.9 KiB
LLVM
57 lines
3.9 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5
|
|
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -stop-after=si-fix-sgpr-copies < %s | FileCheck %s
|
|
|
|
; FIXME: Merge with tail-call-inreg-arguments.ll
|
|
|
|
; Currently all functions with convergence tokens will fail the
|
|
; machine verifier after register allocation.
|
|
declare void @void_func_i64_inreg(i64 inreg)
|
|
|
|
define void @tail_call_i64_inreg_uniform_in_vgpr_convergence_tokens() #0 {
|
|
; CHECK-LABEL: name: tail_call_i64_inreg_uniform_in_vgpr_convergence_tokens
|
|
; CHECK: bb.0 (%ir-block.0):
|
|
; CHECK-NEXT: liveins: $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9, $sgpr10_sgpr11, $sgpr12, $sgpr13, $sgpr14, $sgpr15, $vgpr31
|
|
; CHECK-NEXT: {{ $}}
|
|
; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr31
|
|
; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr15
|
|
; CHECK-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr14
|
|
; CHECK-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr13
|
|
; CHECK-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr12
|
|
; CHECK-NEXT: [[COPY5:%[0-9]+]]:sgpr_64 = COPY $sgpr10_sgpr11
|
|
; CHECK-NEXT: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr8_sgpr9
|
|
; CHECK-NEXT: [[COPY7:%[0-9]+]]:sgpr_64 = COPY $sgpr6_sgpr7
|
|
; CHECK-NEXT: [[COPY8:%[0-9]+]]:sgpr_64 = COPY $sgpr4_sgpr5
|
|
; CHECK-NEXT: [[CONVERGENCECTRL_ENTRY:%[0-9]+]]:sreg_64 = CONVERGENCECTRL_ENTRY
|
|
; CHECK-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
|
|
; CHECK-NEXT: [[DS_READ_B64_gfx9_:%[0-9]+]]:vreg_64 = DS_READ_B64_gfx9 killed [[V_MOV_B32_e32_]], 0, 0, implicit $exec :: (load (s64) from `ptr addrspace(3) null`, addrspace 3)
|
|
; CHECK-NEXT: [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[DS_READ_B64_gfx9_]].sub1
|
|
; CHECK-NEXT: [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[DS_READ_B64_gfx9_]].sub0
|
|
; CHECK-NEXT: CONVERGENCECTRL_GLUE [[CONVERGENCECTRL_ENTRY]]
|
|
; CHECK-NEXT: [[COPY11:%[0-9]+]]:vgpr_32 = COPY [[COPY10]]
|
|
; CHECK-NEXT: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 killed [[COPY11]], implicit $exec, implicit [[CONVERGENCECTRL_ENTRY]]
|
|
; CHECK-NEXT: CONVERGENCECTRL_GLUE [[CONVERGENCECTRL_ENTRY]]
|
|
; CHECK-NEXT: [[COPY12:%[0-9]+]]:vgpr_32 = COPY [[COPY9]]
|
|
; CHECK-NEXT: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 killed [[COPY12]], implicit $exec, implicit [[CONVERGENCECTRL_ENTRY]]
|
|
; CHECK-NEXT: [[SI_PC_ADD_REL_OFFSET:%[0-9]+]]:sreg_64 = SI_PC_ADD_REL_OFFSET target-flags(amdgpu-gotprel32-lo) @void_func_i64_inreg, target-flags(amdgpu-gotprel32-hi) @void_func_i64_inreg, implicit-def dead $scc
|
|
; CHECK-NEXT: [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:ccr_sgpr_64 = S_LOAD_DWORDX2_IMM killed [[SI_PC_ADD_REL_OFFSET]], 0, 0 :: (dereferenceable invariant load (s64) from got, addrspace 4)
|
|
; CHECK-NEXT: $sgpr4_sgpr5 = COPY [[COPY8]]
|
|
; CHECK-NEXT: $sgpr6_sgpr7 = COPY [[COPY7]]
|
|
; CHECK-NEXT: $sgpr8_sgpr9 = COPY [[COPY6]]
|
|
; CHECK-NEXT: $sgpr10_sgpr11 = COPY [[COPY5]]
|
|
; CHECK-NEXT: $sgpr12 = COPY [[COPY4]]
|
|
; CHECK-NEXT: $sgpr13 = COPY [[COPY3]]
|
|
; CHECK-NEXT: $sgpr14 = COPY [[COPY2]]
|
|
; CHECK-NEXT: $sgpr15 = COPY [[COPY1]]
|
|
; CHECK-NEXT: $vgpr31 = COPY [[COPY]]
|
|
; CHECK-NEXT: $sgpr0 = COPY [[V_READFIRSTLANE_B32_]]
|
|
; CHECK-NEXT: $sgpr1 = COPY [[V_READFIRSTLANE_B32_1]]
|
|
; CHECK-NEXT: CONVERGENCECTRL_GLUE [[CONVERGENCECTRL_ENTRY]]
|
|
; CHECK-NEXT: SI_TCRETURN killed [[S_LOAD_DWORDX2_IMM]], @void_func_i64_inreg, 0, csr_amdgpu, implicit $sgpr4_sgpr5, implicit $sgpr6_sgpr7, implicit $sgpr8_sgpr9, implicit $sgpr10_sgpr11, implicit $sgpr12, implicit $sgpr13, implicit $sgpr14, implicit $sgpr15, implicit $vgpr31, implicit $sgpr0, implicit $sgpr1, implicit [[CONVERGENCECTRL_ENTRY]]
|
|
%t = call token @llvm.experimental.convergence.entry()
|
|
%uniform.vgpr = load i64, ptr addrspace(3) null, align 8
|
|
tail call void @void_func_i64_inreg(i64 inreg %uniform.vgpr) #0 [ "convergencectrl"(token %t) ]
|
|
ret void
|
|
}
|
|
|
|
attributes #0 = { convergent }
|