29f88b93fd
Artifact combiner is not able to access individual elements after using LCMTy style merge/unmerge, extract and insert to change vector number of elements (pad with undef or split to sub-vector instructions). Use unmerge to individual elements instead and then merge elements into requested types. Change argument lowering for vectors and moreElementsVector to use buildPadVectorWithUndefElements and buildDeleteTrailingVectorElements. FewerElementsVector had a few helpers that had different behavior, introduce new helper for most of the opcodes. FewerElementsVector helper is more flexible since it can create leftover instruction smaller then requested type (useful in case target wants to avoid pad with undef and use fewer registers). If target does not want leftover of different type it should call more elements first. Some helpers were performing more elements first to have split without leftover. Opcodes that used this helper use clampMaxNumElementsStrict (does more elements first) in LegalizerInfo to avoid test changes. Fixes failures caused by failing to combine artifacts created during more/fewer elements vector. Differential Revision: https://reviews.llvm.org/D114198
113 lines
6.3 KiB
LLVM
113 lines
6.3 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
|
; RUN: llc -global-isel -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
|
|
; RUN: llc -global-isel -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck -check-prefix=GFX10 %s
|
|
|
|
define amdgpu_kernel void @v_insert_v64i32_37(<64 x i32> addrspace(1)* %ptr.in, <64 x i32> addrspace(1)* %ptr.out) #0 {
|
|
; GCN-LABEL: v_insert_v64i32_37:
|
|
; GCN: ; %bb.0:
|
|
; GCN-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x0
|
|
; GCN-NEXT: v_lshlrev_b32_e32 v64, 8, v0
|
|
; GCN-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GCN-NEXT: global_load_dwordx4 v[32:35], v64, s[0:1]
|
|
; GCN-NEXT: global_load_dwordx4 v[36:39], v64, s[0:1] offset:16
|
|
; GCN-NEXT: global_load_dwordx4 v[40:43], v64, s[0:1] offset:32
|
|
; GCN-NEXT: global_load_dwordx4 v[44:47], v64, s[0:1] offset:48
|
|
; GCN-NEXT: global_load_dwordx4 v[48:51], v64, s[0:1] offset:64
|
|
; GCN-NEXT: global_load_dwordx4 v[52:55], v64, s[0:1] offset:80
|
|
; GCN-NEXT: global_load_dwordx4 v[0:3], v64, s[0:1] offset:128
|
|
; GCN-NEXT: global_load_dwordx4 v[56:59], v64, s[0:1] offset:96
|
|
; GCN-NEXT: global_load_dwordx4 v[60:63], v64, s[0:1] offset:112
|
|
; GCN-NEXT: global_load_dwordx4 v[4:7], v64, s[0:1] offset:144
|
|
; GCN-NEXT: global_load_dwordx4 v[8:11], v64, s[0:1] offset:160
|
|
; GCN-NEXT: global_load_dwordx4 v[12:15], v64, s[0:1] offset:176
|
|
; GCN-NEXT: global_load_dwordx4 v[16:19], v64, s[0:1] offset:192
|
|
; GCN-NEXT: global_load_dwordx4 v[20:23], v64, s[0:1] offset:208
|
|
; GCN-NEXT: global_load_dwordx4 v[24:27], v64, s[0:1] offset:224
|
|
; GCN-NEXT: global_load_dwordx4 v[28:31], v64, s[0:1] offset:240
|
|
; GCN-NEXT: s_waitcnt vmcnt(6)
|
|
; GCN-NEXT: v_mov_b32_e32 v5, 0x3e7
|
|
; GCN-NEXT: global_store_dwordx4 v64, v[0:3], s[2:3] offset:128
|
|
; GCN-NEXT: global_store_dwordx4 v64, v[4:7], s[2:3] offset:144
|
|
; GCN-NEXT: s_waitcnt vmcnt(7)
|
|
; GCN-NEXT: global_store_dwordx4 v64, v[8:11], s[2:3] offset:160
|
|
; GCN-NEXT: s_waitcnt vmcnt(7)
|
|
; GCN-NEXT: global_store_dwordx4 v64, v[12:15], s[2:3] offset:176
|
|
; GCN-NEXT: s_waitcnt vmcnt(7)
|
|
; GCN-NEXT: global_store_dwordx4 v64, v[16:19], s[2:3] offset:192
|
|
; GCN-NEXT: s_waitcnt vmcnt(7)
|
|
; GCN-NEXT: global_store_dwordx4 v64, v[20:23], s[2:3] offset:208
|
|
; GCN-NEXT: s_waitcnt vmcnt(7)
|
|
; GCN-NEXT: global_store_dwordx4 v64, v[24:27], s[2:3] offset:224
|
|
; GCN-NEXT: global_store_dwordx4 v64, v[32:35], s[2:3]
|
|
; GCN-NEXT: global_store_dwordx4 v64, v[36:39], s[2:3] offset:16
|
|
; GCN-NEXT: global_store_dwordx4 v64, v[40:43], s[2:3] offset:32
|
|
; GCN-NEXT: global_store_dwordx4 v64, v[44:47], s[2:3] offset:48
|
|
; GCN-NEXT: global_store_dwordx4 v64, v[48:51], s[2:3] offset:64
|
|
; GCN-NEXT: global_store_dwordx4 v64, v[52:55], s[2:3] offset:80
|
|
; GCN-NEXT: global_store_dwordx4 v64, v[56:59], s[2:3] offset:96
|
|
; GCN-NEXT: global_store_dwordx4 v64, v[60:63], s[2:3] offset:112
|
|
; GCN-NEXT: s_waitcnt vmcnt(15)
|
|
; GCN-NEXT: global_store_dwordx4 v64, v[28:31], s[2:3] offset:240
|
|
; GCN-NEXT: s_endpgm
|
|
;
|
|
; GFX10-LABEL: v_insert_v64i32_37:
|
|
; GFX10: ; %bb.0:
|
|
; GFX10-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x0
|
|
; GFX10-NEXT: v_lshlrev_b32_e32 v64, 8, v0
|
|
; GFX10-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-NEXT: s_clause 0xf
|
|
; GFX10-NEXT: global_load_dwordx4 v[32:35], v64, s[0:1]
|
|
; GFX10-NEXT: global_load_dwordx4 v[36:39], v64, s[0:1] offset:16
|
|
; GFX10-NEXT: global_load_dwordx4 v[0:3], v64, s[0:1] offset:128
|
|
; GFX10-NEXT: global_load_dwordx4 v[40:43], v64, s[0:1] offset:32
|
|
; GFX10-NEXT: global_load_dwordx4 v[44:47], v64, s[0:1] offset:48
|
|
; GFX10-NEXT: global_load_dwordx4 v[48:51], v64, s[0:1] offset:64
|
|
; GFX10-NEXT: global_load_dwordx4 v[52:55], v64, s[0:1] offset:80
|
|
; GFX10-NEXT: global_load_dwordx4 v[56:59], v64, s[0:1] offset:96
|
|
; GFX10-NEXT: global_load_dwordx4 v[60:63], v64, s[0:1] offset:112
|
|
; GFX10-NEXT: global_load_dwordx4 v[4:7], v64, s[0:1] offset:144
|
|
; GFX10-NEXT: global_load_dwordx4 v[8:11], v64, s[0:1] offset:160
|
|
; GFX10-NEXT: global_load_dwordx4 v[12:15], v64, s[0:1] offset:176
|
|
; GFX10-NEXT: global_load_dwordx4 v[16:19], v64, s[0:1] offset:192
|
|
; GFX10-NEXT: global_load_dwordx4 v[20:23], v64, s[0:1] offset:208
|
|
; GFX10-NEXT: global_load_dwordx4 v[24:27], v64, s[0:1] offset:224
|
|
; GFX10-NEXT: global_load_dwordx4 v[28:31], v64, s[0:1] offset:240
|
|
; GFX10-NEXT: s_waitcnt vmcnt(6)
|
|
; GFX10-NEXT: v_mov_b32_e32 v5, 0x3e7
|
|
; GFX10-NEXT: global_store_dwordx4 v64, v[0:3], s[2:3] offset:128
|
|
; GFX10-NEXT: global_store_dwordx4 v64, v[4:7], s[2:3] offset:144
|
|
; GFX10-NEXT: s_waitcnt vmcnt(5)
|
|
; GFX10-NEXT: global_store_dwordx4 v64, v[8:11], s[2:3] offset:160
|
|
; GFX10-NEXT: s_waitcnt vmcnt(4)
|
|
; GFX10-NEXT: global_store_dwordx4 v64, v[12:15], s[2:3] offset:176
|
|
; GFX10-NEXT: s_waitcnt vmcnt(3)
|
|
; GFX10-NEXT: global_store_dwordx4 v64, v[16:19], s[2:3] offset:192
|
|
; GFX10-NEXT: s_waitcnt vmcnt(2)
|
|
; GFX10-NEXT: global_store_dwordx4 v64, v[20:23], s[2:3] offset:208
|
|
; GFX10-NEXT: global_store_dwordx4 v64, v[32:35], s[2:3]
|
|
; GFX10-NEXT: global_store_dwordx4 v64, v[36:39], s[2:3] offset:16
|
|
; GFX10-NEXT: global_store_dwordx4 v64, v[40:43], s[2:3] offset:32
|
|
; GFX10-NEXT: global_store_dwordx4 v64, v[44:47], s[2:3] offset:48
|
|
; GFX10-NEXT: global_store_dwordx4 v64, v[48:51], s[2:3] offset:64
|
|
; GFX10-NEXT: global_store_dwordx4 v64, v[52:55], s[2:3] offset:80
|
|
; GFX10-NEXT: global_store_dwordx4 v64, v[56:59], s[2:3] offset:96
|
|
; GFX10-NEXT: global_store_dwordx4 v64, v[60:63], s[2:3] offset:112
|
|
; GFX10-NEXT: s_waitcnt vmcnt(1)
|
|
; GFX10-NEXT: global_store_dwordx4 v64, v[24:27], s[2:3] offset:224
|
|
; GFX10-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX10-NEXT: global_store_dwordx4 v64, v[28:31], s[2:3] offset:240
|
|
; GFX10-NEXT: s_endpgm
|
|
%id = call i32 @llvm.amdgcn.workitem.id.x()
|
|
%gep.in = getelementptr <64 x i32>, <64 x i32> addrspace(1)* %ptr.in, i32 %id
|
|
%vec = load <64 x i32>, <64 x i32> addrspace(1)* %gep.in
|
|
%insert = insertelement <64 x i32> %vec, i32 999, i32 37
|
|
%gep.out = getelementptr <64 x i32>, <64 x i32> addrspace(1)* %ptr.out, i32 %id
|
|
store <64 x i32> %insert, <64 x i32> addrspace(1)* %gep.out
|
|
ret void
|
|
}
|
|
|
|
declare i32 @llvm.amdgcn.workitem.id.x() #1
|
|
|
|
attributes #0 = { "amdgpu-flat-work-group-size"="1,256" "amdgpu-waves-per-eu"="1,10" }
|
|
attributes #1 = { nounwind readnone speculatable willreturn }
|