maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

[AMDGPU] Mark s_barrier as having side effects but not accessing memory. 

Summary:
This fixes poor scheduling in a function containing a barrier and a few
load instructions.

Without this fix, ScheduleDAGInstrs::buildSchedGraph adds an artificial
edge in the dependency graph from the barrier instruction to the exit
node representing live-out latency, with a latency of about 500 cycles.
Because of this it thinks the critical path through the graph also has
a latency of about 500 cycles. And because of that it does not think
that any of the load instructions are on the critical path, so it
schedules them with no regard for their (80 cycle) latency, which gives
poor results.

Reviewers: arsenm, dstuttard, tpr, nhaehnle

Subscribers: kzhuravl, jvesely, wdng, yaxunl, t-tye, hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D67218

llvm-svn: 371192
This commit is contained in:
Jay Foad 2019-09-06 10:07:28 +00:00
parent 68df9dc098
commit 6c0204c794
7 changed files with 113 additions and 68 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
llvm/include/llvm/IR/IntrinsicsAMDGPU.td
View File

@ -206,7 +206,7 @@ def int_amdgcn_s_sendmsghalt : GCCBuiltin<"__builtin_amdgcn_s_sendmsghalt">,
[ImmArg<0>, IntrNoMem, IntrHasSideEffects]>;
def int_amdgcn_s_barrier : GCCBuiltin<"__builtin_amdgcn_s_barrier">,
Intrinsic<[], [], [IntrConvergent]>;
Intrinsic<[], [], [IntrNoMem, IntrHasSideEffects, IntrConvergent]>;
def int_amdgcn_wave_barrier : GCCBuiltin<"__builtin_amdgcn_wave_barrier">,
Intrinsic<[], [], [IntrConvergent]>;

2
llvm/lib/Target/AMDGPU/SOPInstructions.td
View File

@ -1076,8 +1076,6 @@ def S_BARRIER : SOPP <0x0000000a, (ins), "s_barrier",
[(int_amdgcn_s_barrier)]> {
let SchedRW = [WriteBarrier];
let simm16 = 0;
let mayLoad = 1;
let mayStore = 1;
let isConvergent = 1;
}

56
llvm/test/CodeGen/AMDGPU/llvm.amdgcn.s.barrier.ll
View File

@ -16,9 +16,9 @@ define amdgpu_kernel void @test_barrier(i32 addrspace(1)* %out, i32 %size) #0 {
; VARIANT0-NEXT: v_mov_b32_e32 v2, 0
; VARIANT0-NEXT: s_waitcnt lgkmcnt(0)
; VARIANT0-NEXT: buffer_store_dword v0, v[1:2], s[4:7], 0 addr64
; VARIANT0-NEXT: v_add_i32_e32 v3, vcc, s2, v3
; VARIANT0-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; VARIANT0-NEXT: s_barrier
; VARIANT0-NEXT: v_add_i32_e32 v3, vcc, s2, v3
; VARIANT0-NEXT: v_ashrrev_i32_e32 v4, 31, v3
; VARIANT0-NEXT: v_lshl_b64 v[3:4], v[3:4], 2
; VARIANT0-NEXT: buffer_load_dword v0, v[3:4], s[4:7], 0 addr64
@ -37,8 +37,8 @@ define amdgpu_kernel void @test_barrier(i32 addrspace(1)* %out, i32 %size) #0 {
; VARIANT1-NEXT: v_mov_b32_e32 v2, 0
; VARIANT1-NEXT: s_waitcnt lgkmcnt(0)
; VARIANT1-NEXT: buffer_store_dword v0, v[1:2], s[4:7], 0 addr64
; VARIANT1-NEXT: v_add_i32_e32 v3, vcc, s2, v3
; VARIANT1-NEXT: s_barrier
; VARIANT1-NEXT: v_add_i32_e32 v3, vcc, s2, v3
; VARIANT1-NEXT: v_ashrrev_i32_e32 v4, 31, v3
; VARIANT1-NEXT: v_lshl_b64 v[3:4], v[3:4], 2
; VARIANT1-NEXT: s_waitcnt expcnt(0)
@ -50,46 +50,46 @@ define amdgpu_kernel void @test_barrier(i32 addrspace(1)* %out, i32 %size) #0 {
; VARIANT2-LABEL: test_barrier:
; VARIANT2: ; %bb.0: ; %entry
; VARIANT2-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24
; VARIANT2-NEXT: v_lshlrev_b32_e32 v1, 2, v0
; VARIANT2-NEXT: s_load_dword s0, s[0:1], 0x2c
; VARIANT2-NEXT: v_lshlrev_b32_e32 v3, 2, v0
; VARIANT2-NEXT: s_waitcnt lgkmcnt(0)
; VARIANT2-NEXT: v_mov_b32_e32 v2, s3
; VARIANT2-NEXT: v_add_co_u32_e32 v1, vcc, s2, v1
; VARIANT2-NEXT: v_addc_co_u32_e32 v2, vcc, 0, v2, vcc
; VARIANT2-NEXT: global_store_dword v[1:2], v0, off
; VARIANT2-NEXT: v_mov_b32_e32 v4, s3
; VARIANT2-NEXT: v_xad_u32 v1, v0, -1, s0
; VARIANT2-NEXT: v_ashrrev_i32_e32 v2, 31, v1
; VARIANT2-NEXT: v_add_co_u32_e32 v3, vcc, s2, v3
; VARIANT2-NEXT: v_lshlrev_b64 v[1:2], 2, v[1:2]
; VARIANT2-NEXT: v_addc_co_u32_e32 v4, vcc, 0, v4, vcc
; VARIANT2-NEXT: global_store_dword v[3:4], v0, off
; VARIANT2-NEXT: v_mov_b32_e32 v5, s3
; VARIANT2-NEXT: v_add_co_u32_e32 v0, vcc, s2, v1
; VARIANT2-NEXT: v_addc_co_u32_e32 v1, vcc, v5, v2, vcc
; VARIANT2-NEXT: s_waitcnt vmcnt(0)
; VARIANT2-NEXT: s_barrier
; VARIANT2-NEXT: v_xad_u32 v3, v0, -1, s0
; VARIANT2-NEXT: v_ashrrev_i32_e32 v4, 31, v3
; VARIANT2-NEXT: v_lshlrev_b64 v[3:4], 2, v[3:4]
; VARIANT2-NEXT: v_mov_b32_e32 v0, s3
; VARIANT2-NEXT: v_add_co_u32_e32 v3, vcc, s2, v3
; VARIANT2-NEXT: v_addc_co_u32_e32 v4, vcc, v0, v4, vcc
; VARIANT2-NEXT: global_load_dword v0, v[3:4], off
; VARIANT2-NEXT: global_load_dword v0, v[0:1], off
; VARIANT2-NEXT: s_waitcnt vmcnt(0)
; VARIANT2-NEXT: global_store_dword v[1:2], v0, off
; VARIANT2-NEXT: global_store_dword v[3:4], v0, off
; VARIANT2-NEXT: s_endpgm
;
; VARIANT3-LABEL: test_barrier:
; VARIANT3: ; %bb.0: ; %entry
; VARIANT3-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24
; VARIANT3-NEXT: v_lshlrev_b32_e32 v1, 2, v0
; VARIANT3-NEXT: s_load_dword s0, s[0:1], 0x2c
; VARIANT3-NEXT: v_lshlrev_b32_e32 v3, 2, v0
; VARIANT3-NEXT: s_waitcnt lgkmcnt(0)
; VARIANT3-NEXT: v_mov_b32_e32 v2, s3
; VARIANT3-NEXT: v_add_co_u32_e32 v1, vcc, s2, v1
; VARIANT3-NEXT: v_addc_co_u32_e32 v2, vcc, 0, v2, vcc
; VARIANT3-NEXT: global_store_dword v[1:2], v0, off
; VARIANT3-NEXT: s_barrier
; VARIANT3-NEXT: v_xad_u32 v3, v0, -1, s0
; VARIANT3-NEXT: v_ashrrev_i32_e32 v4, 31, v3
; VARIANT3-NEXT: v_lshlrev_b64 v[3:4], 2, v[3:4]
; VARIANT3-NEXT: v_mov_b32_e32 v0, s3
; VARIANT3-NEXT: v_mov_b32_e32 v4, s3
; VARIANT3-NEXT: v_xad_u32 v1, v0, -1, s0
; VARIANT3-NEXT: v_ashrrev_i32_e32 v2, 31, v1
; VARIANT3-NEXT: v_add_co_u32_e32 v3, vcc, s2, v3
; VARIANT3-NEXT: v_addc_co_u32_e32 v4, vcc, v0, v4, vcc
; VARIANT3-NEXT: global_load_dword v0, v[3:4], off
; VARIANT3-NEXT: v_lshlrev_b64 v[1:2], 2, v[1:2]
; VARIANT3-NEXT: v_addc_co_u32_e32 v4, vcc, 0, v4, vcc
; VARIANT3-NEXT: global_store_dword v[3:4], v0, off
; VARIANT3-NEXT: v_mov_b32_e32 v5, s3
; VARIANT3-NEXT: v_add_co_u32_e32 v0, vcc, s2, v1
; VARIANT3-NEXT: v_addc_co_u32_e32 v1, vcc, v5, v2, vcc
; VARIANT3-NEXT: s_barrier
; VARIANT3-NEXT: global_load_dword v0, v[0:1], off
; VARIANT3-NEXT: s_waitcnt vmcnt(0)
; VARIANT3-NEXT: global_store_dword v[1:2], v0, off
; VARIANT3-NEXT: global_store_dword v[3:4], v0, off
; VARIANT3-NEXT: s_endpgm
entry:
%tmp = call i32 @llvm.amdgcn.workitem.id.x()

8
llvm/test/CodeGen/AMDGPU/local-memory.amdgcn.ll
View File

@ -38,18 +38,16 @@ entry:
; GCN-LABEL: {{^}}local_memory_two_objects:
; GCN: v_lshlrev_b32_e32 [[ADDRW:v[0-9]+]], 2, v0
; CI-DAG: v_sub_i32_e32 [[SUB:v[0-9]+]], vcc, 0, [[ADDRW]]
; CI-DAG: ds_write2_b32 [[ADDRW]], {{v[0-9]+}}, {{v[0-9]+}} offset1:4
; SI-DAG: ds_write2_b32 [[ADDRW]], {{v[0-9]+}}, {{v[0-9]+}} offset1:4
; GCN: s_barrier
; SI-DAG: v_sub_i32_e32 [[SUB0:v[0-9]+]], vcc, 28, [[ADDRW]]
; SI-DAG: v_sub_i32_e32 [[SUB1:v[0-9]+]], vcc, 12, [[ADDRW]]
; GCN: s_barrier
; SI-DAG: ds_read_b32 v{{[0-9]+}}, [[SUB0]]
; SI-DAG: ds_read_b32 v{{[0-9]+}}, [[SUB1]]
; CI: v_sub_i32_e32 [[SUB:v[0-9]+]], vcc, 0, [[ADDRW]]
; CI: ds_read2_b32 {{v\[[0-9]+:[0-9]+\]}}, [[SUB]] offset0:3 offset1:7
define amdgpu_kernel void @local_memory_two_objects(i32 addrspace(1)* %out) #0 {

46
llvm/test/CodeGen/AMDGPU/schedule-barrier.mir Normal file
View File

@ -0,0 +1,46 @@
# RUN: llc -march=amdgcn -mcpu=gfx900 -run-pass=machine-scheduler -o - %s | FileCheck %s
---
# Check that the high latency loads are both scheduled first, before the
# multiplies, despite the presence of a barrier in the function.
# CHECK: BUFFER_LOAD_DWORD_OFFSET
# CHECK: BUFFER_LOAD_DWORD_OFFSET
# CHECK: V_MUL_LO_U32
# CHECK: V_MUL_LO_U32
name: test
tracksRegLiveness: true
body: |
bb.0:
liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9
undef %43.sub3:vreg_128 = COPY $vgpr9
undef %42.sub2:vreg_128 = COPY $vgpr8
undef %41.sub1:vreg_128 = COPY $vgpr7
undef %26.sub0:vreg_128 = COPY $vgpr6
undef %46.sub3:vreg_128 = COPY $vgpr5
undef %45.sub2:vreg_128 = COPY $vgpr4
undef %44.sub1:vreg_128 = COPY $vgpr3
undef %32.sub0:vreg_128 = COPY $vgpr2
undef %38.sub1:vreg_64 = COPY $vgpr1
%38.sub0:vreg_64 = COPY $vgpr0
S_BARRIER
undef %33.sub0:sgpr_128 = V_READFIRSTLANE_B32 %32.sub0, implicit $exec
%33.sub1:sgpr_128 = V_READFIRSTLANE_B32 %44.sub1, implicit $exec
%33.sub2:sgpr_128 = V_READFIRSTLANE_B32 %45.sub2, implicit $exec
%33.sub3:sgpr_128 = V_READFIRSTLANE_B32 %46.sub3, implicit $exec
%15:vgpr_32 = BUFFER_LOAD_DWORD_OFFSET %33, 0, 0, 0, 0, 0, 0, implicit $exec
%39:vgpr_32 = V_MUL_LO_U32 %15, %15, implicit $exec
undef %27.sub0:sgpr_128 = V_READFIRSTLANE_B32 %26.sub0, implicit $exec
%27.sub1:sgpr_128 = V_READFIRSTLANE_B32 %41.sub1, implicit $exec
%27.sub2:sgpr_128 = V_READFIRSTLANE_B32 %42.sub2, implicit $exec
%27.sub3:sgpr_128 = V_READFIRSTLANE_B32 %43.sub3, implicit $exec
%19:vgpr_32 = BUFFER_LOAD_DWORD_OFFSET %27, 0, 0, 0, 0, 0, 0, implicit $exec
%40:vgpr_32 = V_MUL_LO_U32 %19, %19, implicit $exec
%23:vgpr_32 = V_ADD_U32_e32 %39, %40, implicit $exec
GLOBAL_STORE_DWORD %38, %23, 0, 0, 0, 0, implicit $exec
S_ENDPGM 0
...

17
llvm/test/CodeGen/AMDGPU/v_mac_f16.ll
View File

@ -304,14 +304,17 @@ entry:
; GCN: {{buffer|flat}}_load_dword v[[C_V2_F16:[0-9]+]]
; SI: v_cvt_f32_f16_e32 v[[A_F32_0:[0-9]+]], v[[A_V2_F16]]
; SI: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
; SI: v_cvt_f32_f16_e32 v[[A_F32_1:[0-9]+]], v[[A_F16_1]]
; SI: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
; SI: v_cvt_f32_f16_e32 v[[B_F32_1:[0-9]+]], v[[B_F16_1]]
; SI: v_cvt_f32_f16_e32 v[[B_F32_0:[0-9]+]], v[[B_V2_F16]]
; SI: v_lshrrev_b32_e32 v[[C_F16_1:[0-9]+]], 16, v[[C_V2_F16]]
; SI: v_cvt_f32_f16_e32 v[[C_F32_1:[0-9]+]], v[[C_F16_1]]
; SI-DAG: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
; SI-DAG: v_cvt_f32_f16_e32 v[[A_F32_1:[0-9]+]], v[[A_F16_1]]
; SI-DAG: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
; SI-DAG: v_cvt_f32_f16_e32 v[[B_F32_1:[0-9]+]], v[[B_F16_1]]
; SI-DAG: v_cvt_f32_f16_e32 v[[B_F32_0:[0-9]+]], v[[B_V2_F16]]
; SI-DAG: v_lshrrev_b32_e32 v[[C_F16_1:[0-9]+]], 16, v[[C_V2_F16]]
; SI-DAG: v_cvt_f32_f16_e32 v[[C_F32_1:[0-9]+]], v[[C_F16_1]]
; SI-DAG: v_cvt_f32_f16_e32 v[[C_F32_0:[0-9]+]], v[[C_V2_F16]]
; SI-DAG: v_mac_f32_e32 v[[C_F32_0]], v[[A_F32_0]], v[[B_F32_0]]
; SI-DAG: v_cvt_f16_f32_e32 v[[R_F16_LO:[0-9]+]], v[[C_F32_0]]
; SI-DAG: v_mac_f32_e32 v[[C_F32_1]], v[[A_F32_1]], v[[B_F32_1]]

50
llvm/test/CodeGen/AMDGPU/waitcnt-vscnt.ll
View File

@ -5,8 +5,8 @@
; GCN-LABEL: barrier_vmcnt_global:
; GFX8: flat_load_dword
; GFX9_10: global_load_dword
; GFX8-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
; GFX9_10-NEXT: s_waitcnt vmcnt(0){{$}}
; GFX8: s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
; GFX9_10: s_waitcnt vmcnt(0){{$}}
; GCN-NEXT: s_barrier
define amdgpu_kernel void @barrier_vmcnt_global(i32 addrspace(1)* %arg) {
bb:
@ -28,9 +28,9 @@ bb:
; GCN-LABEL: barrier_vscnt_global:
; GFX8: flat_store_dword
; GFX9_10: global_store_dword
; GFX8-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
; GFX9-NEXT: s_waitcnt vmcnt(0){{$}}
; GFX10-NEXT: s_waitcnt_vscnt null, 0x0
; GFX8: s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
; GFX9: s_waitcnt vmcnt(0){{$}}
; GFX10: s_waitcnt_vscnt null, 0x0
; GCN-NEXT: s_barrier
define amdgpu_kernel void @barrier_vscnt_global(i32 addrspace(1)* %arg) {
bb:
@ -54,9 +54,9 @@ bb:
; GCN-LABEL: barrier_vmcnt_vscnt_global:
; GFX8: flat_load_dword
; GFX9_10: global_load_dword
; GFX8-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
; GFX9_10-NEXT: s_waitcnt vmcnt(0){{$}}
; GFX10-NEXT: s_waitcnt_vscnt null, 0x0
; GFX8: s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
; GFX9_10: s_waitcnt vmcnt(0){{$}}
; GFX10: s_waitcnt_vscnt null, 0x0
; GCN-NEXT: s_barrier
define amdgpu_kernel void @barrier_vmcnt_vscnt_global(i32 addrspace(1)* %arg) {
bb:
@ -81,7 +81,7 @@ bb:
; GCN-LABEL: barrier_vmcnt_flat:
; GCN: flat_load_dword
; GCN-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
; GCN: s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
; GCN-NEXT: s_barrier
define amdgpu_kernel void @barrier_vmcnt_flat(i32* %arg) {
bb:
@ -102,9 +102,9 @@ bb:
; GCN-LABEL: barrier_vscnt_flat:
; GCN: flat_store_dword
; GFX8_9-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
; GFX10-NEXT: s_waitcnt lgkmcnt(0){{$}}
; GFX10-NEXT: s_waitcnt_vscnt null, 0x0
; GFX8_9: s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
; GFX10: s_waitcnt lgkmcnt(0){{$}}
; GFX10: s_waitcnt_vscnt null, 0x0
; GCN-NEXT: s_barrier
define amdgpu_kernel void @barrier_vscnt_flat(i32* %arg) {
bb:
@ -127,8 +127,8 @@ bb:
; GCN-LABEL: barrier_vmcnt_vscnt_flat:
; GCN: flat_load_dword
; GCN-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
; GFX10-NEXT: s_waitcnt_vscnt null, 0x0
; GCN: s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
; GFX10: s_waitcnt_vscnt null, 0x0
; GCN-NEXT: s_barrier
define amdgpu_kernel void @barrier_vmcnt_vscnt_flat(i32* %arg) {
bb:
@ -153,8 +153,8 @@ bb:
; GCN-LABEL: barrier_vmcnt_vscnt_flat_workgroup:
; GCN: flat_load_dword
; GCN-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
; GFX10-NEXT: s_waitcnt_vscnt null, 0x0
; GCN: s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
; GFX10: s_waitcnt_vscnt null, 0x0
; GCN-NEXT: s_barrier
define amdgpu_kernel void @barrier_vmcnt_vscnt_flat_workgroup(i32* %arg) {
bb:
@ -218,8 +218,8 @@ bb:
; GCN-LABEL: store_vscnt_private:
; GCN: buffer_store_dword
; GFX8_9-NEXT: s_waitcnt vmcnt(0)
; GFX10-NEXT: s_waitcnt_vscnt null, 0x0
; GFX8_9: s_waitcnt vmcnt(0)
; GFX10: s_waitcnt_vscnt null, 0x0
; GCN-NEXT: s_setpc_b64
define void @store_vscnt_private(i32 addrspace(5)* %p) {
store i32 0, i32 addrspace(5)* %p
@ -229,8 +229,8 @@ define void @store_vscnt_private(i32 addrspace(5)* %p) {
; GCN-LABEL: store_vscnt_global:
; GFX8: flat_store_dword
; GFX9_10: global_store_dword
; GFX8_9-NEXT: s_waitcnt vmcnt(0)
; GFX10-NEXT: s_waitcnt_vscnt null, 0x0
; GFX8_9: s_waitcnt vmcnt(0)
; GFX10: s_waitcnt_vscnt null, 0x0
; GCN-NEXT: s_setpc_b64
define void @store_vscnt_global(i32 addrspace(1)* %p) {
store i32 0, i32 addrspace(1)* %p
@ -239,9 +239,9 @@ define void @store_vscnt_global(i32 addrspace(1)* %p) {
; GCN-LABEL: store_vscnt_flat:
; GCN: flat_store_dword
; GFX8_9-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
; GFX10-NEXT: s_waitcnt lgkmcnt(0){{$}}
; GFX10-NEXT: s_waitcnt_vscnt null, 0x0
; GFX8_9: s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
; GFX10: s_waitcnt lgkmcnt(0){{$}}
; GFX10: s_waitcnt_vscnt null, 0x0
; GCN-NEXT: s_setpc_b64
define void @store_vscnt_flat(i32* %p) {
store i32 0, i32* %p
@ -249,8 +249,8 @@ define void @store_vscnt_flat(i32* %p) {
}
; GCN-LABEL: function_prologue:
; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0){{$}}
; GFX10-NEXT: s_waitcnt_vscnt null, 0x0
; GCN: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0){{$}}
; GFX10: s_waitcnt_vscnt null, 0x0
; GCN-NEXT: s_setpc_b64
define void @function_prologue() {
ret void