2019-07-19 23:43:56 +08:00
|
|
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
|
|
|
|
; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck %s -enable-var-scope -check-prefixes=GCN,SI
|
|
|
|
|
; RUN: llc -march=amdgcn -mcpu=fiji -verify-machineinstrs < %s | FileCheck %s -enable-var-scope -check-prefixes=GCN,GFX89,VI
|
|
|
|
|
; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck %s -enable-var-scope -check-prefixes=GCN,GFX89,GFX9
|
2017-02-22 08:27:34 +08:00
|
|
|
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @s_cvt_pkrtz_v2f16_f32(<2 x half> addrspace(1)* %out, float %x, float %y) #0 {
|
2019-07-19 23:43:56 +08:00
|
|
|
; SI-LABEL: s_cvt_pkrtz_v2f16_f32:
|
|
|
|
|
; SI: ; %bb.0:
|
|
|
|
|
; SI-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0xb
|
|
|
|
|
; SI-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x9
|
|
|
|
|
; SI-NEXT: s_mov_b32 s7, 0xf000
|
|
|
|
|
; SI-NEXT: s_mov_b32 s6, -1
|
|
|
|
|
; SI-NEXT: s_waitcnt lgkmcnt(0)
|
|
|
|
|
; SI-NEXT: v_mov_b32_e32 v0, s3
|
|
|
|
|
; SI-NEXT: v_cvt_pkrtz_f16_f32_e32 v0, s2, v0
|
|
|
|
|
; SI-NEXT: buffer_store_dword v0, off, s[4:7], 0
|
|
|
|
|
; SI-NEXT: s_endpgm
|
|
|
|
|
;
|
|
|
|
|
; VI-LABEL: s_cvt_pkrtz_v2f16_f32:
|
|
|
|
|
; VI: ; %bb.0:
|
|
|
|
|
; VI-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24
|
|
|
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x2c
|
|
|
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
[AMDGPU] Remove dubious logic in bidirectional list scheduler
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
|
|
|
; VI-NEXT: v_mov_b32_e32 v0, s1
|
|
|
|
|
; VI-NEXT: v_cvt_pkrtz_f16_f32 v2, s0, v0
|
2019-07-19 23:43:56 +08:00
|
|
|
; VI-NEXT: v_mov_b32_e32 v0, s2
|
|
|
|
|
; VI-NEXT: v_mov_b32_e32 v1, s3
|
|
|
|
|
; VI-NEXT: flat_store_dword v[0:1], v2
|
|
|
|
|
; VI-NEXT: s_endpgm
|
|
|
|
|
;
|
|
|
|
|
; GFX9-LABEL: s_cvt_pkrtz_v2f16_f32:
|
|
|
|
|
; GFX9: ; %bb.0:
|
|
|
|
|
; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24
|
|
|
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x2c
|
|
|
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
[AMDGPU] Remove dubious logic in bidirectional list scheduler
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
|
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, s1
|
|
|
|
|
; GFX9-NEXT: v_cvt_pkrtz_f16_f32 v2, s0, v0
|
2019-07-19 23:43:56 +08:00
|
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, s2
|
|
|
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, s3
|
|
|
|
|
; GFX9-NEXT: global_store_dword v[0:1], v2, off
|
|
|
|
|
; GFX9-NEXT: s_endpgm
|
2017-02-22 08:27:34 +08:00
|
|
|
%result = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float %x, float %y)
|
|
|
|
|
store <2 x half> %result, <2 x half> addrspace(1)* %out
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @s_cvt_pkrtz_samereg_v2f16_f32(<2 x half> addrspace(1)* %out, float %x) #0 {
|
2019-07-19 23:43:56 +08:00
|
|
|
; SI-LABEL: s_cvt_pkrtz_samereg_v2f16_f32:
|
|
|
|
|
; SI: ; %bb.0:
|
|
|
|
|
; SI-NEXT: s_load_dword s2, s[0:1], 0xb
|
|
|
|
|
; SI-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x9
|
|
|
|
|
; SI-NEXT: s_mov_b32 s7, 0xf000
|
|
|
|
|
; SI-NEXT: s_mov_b32 s6, -1
|
|
|
|
|
; SI-NEXT: s_waitcnt lgkmcnt(0)
|
|
|
|
|
; SI-NEXT: v_cvt_pkrtz_f16_f32_e64 v0, s2, s2
|
|
|
|
|
; SI-NEXT: buffer_store_dword v0, off, s[4:7], 0
|
|
|
|
|
; SI-NEXT: s_endpgm
|
|
|
|
|
;
|
|
|
|
|
; VI-LABEL: s_cvt_pkrtz_samereg_v2f16_f32:
|
|
|
|
|
; VI: ; %bb.0:
|
|
|
|
|
; VI-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24
|
|
|
|
|
; VI-NEXT: s_load_dword s0, s[0:1], 0x2c
|
|
|
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
|
|
|
; VI-NEXT: v_mov_b32_e32 v0, s2
|
|
|
|
|
; VI-NEXT: v_cvt_pkrtz_f16_f32 v2, s0, s0
|
[AMDGPU] Remove dubious logic in bidirectional list scheduler
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
|
|
|
; VI-NEXT: v_mov_b32_e32 v1, s3
|
2019-07-19 23:43:56 +08:00
|
|
|
; VI-NEXT: flat_store_dword v[0:1], v2
|
|
|
|
|
; VI-NEXT: s_endpgm
|
|
|
|
|
;
|
|
|
|
|
; GFX9-LABEL: s_cvt_pkrtz_samereg_v2f16_f32:
|
|
|
|
|
; GFX9: ; %bb.0:
|
|
|
|
|
; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24
|
|
|
|
|
; GFX9-NEXT: s_load_dword s0, s[0:1], 0x2c
|
|
|
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
|
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, s2
|
|
|
|
|
; GFX9-NEXT: v_cvt_pkrtz_f16_f32 v2, s0, s0
|
[AMDGPU] Remove dubious logic in bidirectional list scheduler
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
|
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, s3
|
2019-07-19 23:43:56 +08:00
|
|
|
; GFX9-NEXT: global_store_dword v[0:1], v2, off
|
|
|
|
|
; GFX9-NEXT: s_endpgm
|
2017-02-22 08:27:34 +08:00
|
|
|
%result = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float %x, float %x)
|
|
|
|
|
store <2 x half> %result, <2 x half> addrspace(1)* %out
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @s_cvt_pkrtz_undef_undef(<2 x half> addrspace(1)* %out) #0 {
|
2019-07-19 23:43:56 +08:00
|
|
|
; GCN-LABEL: s_cvt_pkrtz_undef_undef:
|
|
|
|
|
; GCN: ; %bb.0:
|
|
|
|
|
; GCN-NEXT: s_endpgm
|
2017-02-22 08:27:34 +08:00
|
|
|
%result = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float undef, float undef)
|
|
|
|
|
store <2 x half> %result, <2 x half> addrspace(1)* %out
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @v_cvt_pkrtz_v2f16_f32(<2 x half> addrspace(1)* %out, float addrspace(1)* %a.ptr, float addrspace(1)* %b.ptr) #0 {
|
2019-07-19 23:43:56 +08:00
|
|
|
; SI-LABEL: v_cvt_pkrtz_v2f16_f32:
|
|
|
|
|
; SI: ; %bb.0:
|
|
|
|
|
; SI-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x9
|
|
|
|
|
; SI-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
|
|
|
|
|
; SI-NEXT: s_mov_b32 s3, 0xf000
|
|
|
|
|
; SI-NEXT: s_mov_b32 s2, 0
|
|
|
|
|
; SI-NEXT: v_lshlrev_b32_e32 v0, 2, v0
|
|
|
|
|
; SI-NEXT: v_mov_b32_e32 v1, 0
|
|
|
|
|
; SI-NEXT: s_mov_b64 s[6:7], s[2:3]
|
|
|
|
|
; SI-NEXT: s_waitcnt lgkmcnt(0)
|
|
|
|
|
; SI-NEXT: s_mov_b64 s[0:1], s[10:11]
|
|
|
|
|
; SI-NEXT: buffer_load_dword v2, v[0:1], s[0:3], 0 addr64
|
|
|
|
|
; SI-NEXT: buffer_load_dword v3, v[0:1], s[4:7], 0 addr64
|
|
|
|
|
; SI-NEXT: s_mov_b64 s[10:11], s[2:3]
|
|
|
|
|
; SI-NEXT: s_waitcnt vmcnt(0)
|
|
|
|
|
; SI-NEXT: v_cvt_pkrtz_f16_f32_e32 v2, v2, v3
|
|
|
|
|
; SI-NEXT: buffer_store_dword v2, v[0:1], s[8:11], 0 addr64
|
|
|
|
|
; SI-NEXT: s_endpgm
|
|
|
|
|
;
|
|
|
|
|
; VI-LABEL: v_cvt_pkrtz_v2f16_f32:
|
|
|
|
|
; VI: ; %bb.0:
|
|
|
|
|
; VI-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
|
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
|
|
|
|
|
; VI-NEXT: v_lshlrev_b32_e32 v4, 2, v0
|
|
|
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
|
|
|
; VI-NEXT: v_mov_b32_e32 v1, s7
|
|
|
|
|
; VI-NEXT: v_add_u32_e32 v0, vcc, s6, v4
|
|
|
|
|
; VI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
|
|
|
|
|
; VI-NEXT: v_mov_b32_e32 v3, s1
|
|
|
|
|
; VI-NEXT: v_add_u32_e32 v2, vcc, s0, v4
|
|
|
|
|
; VI-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc
|
[AMDGPU] Remove dubious logic in bidirectional list scheduler
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
|
|
|
; VI-NEXT: flat_load_dword v0, v[0:1]
|
|
|
|
|
; VI-NEXT: flat_load_dword v1, v[2:3]
|
|
|
|
|
; VI-NEXT: v_mov_b32_e32 v5, s5
|
|
|
|
|
; VI-NEXT: v_add_u32_e32 v4, vcc, s4, v4
|
|
|
|
|
; VI-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc
|
2019-07-19 23:43:56 +08:00
|
|
|
; VI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
[AMDGPU] Remove dubious logic in bidirectional list scheduler
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
|
|
|
; VI-NEXT: v_cvt_pkrtz_f16_f32 v0, v0, v1
|
|
|
|
|
; VI-NEXT: flat_store_dword v[4:5], v0
|
2019-07-19 23:43:56 +08:00
|
|
|
; VI-NEXT: s_endpgm
|
|
|
|
|
;
|
|
|
|
|
; GFX9-LABEL: v_cvt_pkrtz_v2f16_f32:
|
|
|
|
|
; GFX9: ; %bb.0:
|
|
|
|
|
; GFX9-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
|
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
|
|
|
|
|
; GFX9-NEXT: v_lshlrev_b32_e32 v4, 2, v0
|
|
|
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
|
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, s7
|
|
|
|
|
; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, s6, v4
|
|
|
|
|
; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
|
|
|
|
|
; GFX9-NEXT: v_mov_b32_e32 v3, s1
|
|
|
|
|
; GFX9-NEXT: v_add_co_u32_e32 v2, vcc, s0, v4
|
|
|
|
|
; GFX9-NEXT: v_addc_co_u32_e32 v3, vcc, 0, v3, vcc
|
[AMDGPU] Remove dubious logic in bidirectional list scheduler
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
|
|
|
; GFX9-NEXT: global_load_dword v0, v[0:1], off
|
|
|
|
|
; GFX9-NEXT: global_load_dword v1, v[2:3], off
|
|
|
|
|
; GFX9-NEXT: v_mov_b32_e32 v5, s5
|
|
|
|
|
; GFX9-NEXT: v_add_co_u32_e32 v4, vcc, s4, v4
|
|
|
|
|
; GFX9-NEXT: v_addc_co_u32_e32 v5, vcc, 0, v5, vcc
|
2019-07-19 23:43:56 +08:00
|
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0)
|
[AMDGPU] Remove dubious logic in bidirectional list scheduler
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
|
|
|
; GFX9-NEXT: v_cvt_pkrtz_f16_f32 v0, v0, v1
|
|
|
|
|
; GFX9-NEXT: global_store_dword v[4:5], v0, off
|
2019-07-19 23:43:56 +08:00
|
|
|
; GFX9-NEXT: s_endpgm
|
2017-02-22 08:27:34 +08:00
|
|
|
%tid = call i32 @llvm.amdgcn.workitem.id.x()
|
|
|
|
|
%tid.ext = sext i32 %tid to i64
|
|
|
|
|
%a.gep = getelementptr inbounds float, float addrspace(1)* %a.ptr, i64 %tid.ext
|
|
|
|
|
%b.gep = getelementptr inbounds float, float addrspace(1)* %b.ptr, i64 %tid.ext
|
|
|
|
|
%out.gep = getelementptr inbounds <2 x half>, <2 x half> addrspace(1)* %out, i64 %tid.ext
|
|
|
|
|
%a = load volatile float, float addrspace(1)* %a.gep
|
|
|
|
|
%b = load volatile float, float addrspace(1)* %b.gep
|
|
|
|
|
%cvt = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float %a, float %b)
|
|
|
|
|
store <2 x half> %cvt, <2 x half> addrspace(1)* %out.gep
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @v_cvt_pkrtz_v2f16_f32_reg_imm(<2 x half> addrspace(1)* %out, float addrspace(1)* %a.ptr) #0 {
|
2019-07-19 23:43:56 +08:00
|
|
|
; SI-LABEL: v_cvt_pkrtz_v2f16_f32_reg_imm:
|
|
|
|
|
; SI: ; %bb.0:
|
|
|
|
|
; SI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
|
|
|
|
|
; SI-NEXT: s_mov_b32 s7, 0xf000
|
|
|
|
|
; SI-NEXT: s_mov_b32 s6, 0
|
|
|
|
|
; SI-NEXT: v_lshlrev_b32_e32 v0, 2, v0
|
|
|
|
|
; SI-NEXT: v_mov_b32_e32 v1, 0
|
|
|
|
|
; SI-NEXT: s_waitcnt lgkmcnt(0)
|
|
|
|
|
; SI-NEXT: s_mov_b64 s[4:5], s[2:3]
|
|
|
|
|
; SI-NEXT: buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
|
|
|
|
|
; SI-NEXT: s_mov_b64 s[2:3], s[6:7]
|
|
|
|
|
; SI-NEXT: s_waitcnt vmcnt(0)
|
|
|
|
|
; SI-NEXT: v_cvt_pkrtz_f16_f32_e64 v2, v2, 1.0
|
|
|
|
|
; SI-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
|
|
|
|
|
; SI-NEXT: s_endpgm
|
|
|
|
|
;
|
|
|
|
|
; VI-LABEL: v_cvt_pkrtz_v2f16_f32_reg_imm:
|
|
|
|
|
; VI: ; %bb.0:
|
|
|
|
|
; VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
|
|
|
|
|
; VI-NEXT: v_lshlrev_b32_e32 v2, 2, v0
|
|
|
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
|
|
|
; VI-NEXT: v_mov_b32_e32 v1, s3
|
|
|
|
|
; VI-NEXT: v_add_u32_e32 v0, vcc, s2, v2
|
|
|
|
|
; VI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
|
[AMDGPU] Remove dubious logic in bidirectional list scheduler
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
|
|
|
; VI-NEXT: flat_load_dword v0, v[0:1]
|
|
|
|
|
; VI-NEXT: v_mov_b32_e32 v3, s1
|
|
|
|
|
; VI-NEXT: v_add_u32_e32 v2, vcc, s0, v2
|
|
|
|
|
; VI-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc
|
2019-07-19 23:43:56 +08:00
|
|
|
; VI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
[AMDGPU] Remove dubious logic in bidirectional list scheduler
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
|
|
|
; VI-NEXT: v_cvt_pkrtz_f16_f32 v0, v0, 1.0
|
|
|
|
|
; VI-NEXT: flat_store_dword v[2:3], v0
|
2019-07-19 23:43:56 +08:00
|
|
|
; VI-NEXT: s_endpgm
|
|
|
|
|
;
|
|
|
|
|
; GFX9-LABEL: v_cvt_pkrtz_v2f16_f32_reg_imm:
|
|
|
|
|
; GFX9: ; %bb.0:
|
|
|
|
|
; GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
|
|
|
|
|
; GFX9-NEXT: v_lshlrev_b32_e32 v2, 2, v0
|
|
|
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
|
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, s3
|
|
|
|
|
; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, s2, v2
|
|
|
|
|
; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
|
[AMDGPU] Remove dubious logic in bidirectional list scheduler
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
|
|
|
; GFX9-NEXT: global_load_dword v0, v[0:1], off
|
|
|
|
|
; GFX9-NEXT: v_mov_b32_e32 v3, s1
|
|
|
|
|
; GFX9-NEXT: v_add_co_u32_e32 v2, vcc, s0, v2
|
|
|
|
|
; GFX9-NEXT: v_addc_co_u32_e32 v3, vcc, 0, v3, vcc
|
2019-07-19 23:43:56 +08:00
|
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0)
|
[AMDGPU] Remove dubious logic in bidirectional list scheduler
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
|
|
|
; GFX9-NEXT: v_cvt_pkrtz_f16_f32 v0, v0, 1.0
|
|
|
|
|
; GFX9-NEXT: global_store_dword v[2:3], v0, off
|
2019-07-19 23:43:56 +08:00
|
|
|
; GFX9-NEXT: s_endpgm
|
2017-02-22 08:27:34 +08:00
|
|
|
%tid = call i32 @llvm.amdgcn.workitem.id.x()
|
|
|
|
|
%tid.ext = sext i32 %tid to i64
|
|
|
|
|
%a.gep = getelementptr inbounds float, float addrspace(1)* %a.ptr, i64 %tid.ext
|
|
|
|
|
%out.gep = getelementptr inbounds <2 x half>, <2 x half> addrspace(1)* %out, i64 %tid.ext
|
|
|
|
|
%a = load volatile float, float addrspace(1)* %a.gep
|
|
|
|
|
%cvt = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float %a, float 1.0)
|
|
|
|
|
store <2 x half> %cvt, <2 x half> addrspace(1)* %out.gep
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @v_cvt_pkrtz_v2f16_f32_imm_reg(<2 x half> addrspace(1)* %out, float addrspace(1)* %a.ptr) #0 {
|
2019-07-19 23:43:56 +08:00
|
|
|
; SI-LABEL: v_cvt_pkrtz_v2f16_f32_imm_reg:
|
|
|
|
|
; SI: ; %bb.0:
|
|
|
|
|
; SI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
|
|
|
|
|
; SI-NEXT: s_mov_b32 s7, 0xf000
|
|
|
|
|
; SI-NEXT: s_mov_b32 s6, 0
|
|
|
|
|
; SI-NEXT: v_lshlrev_b32_e32 v0, 2, v0
|
|
|
|
|
; SI-NEXT: v_mov_b32_e32 v1, 0
|
|
|
|
|
; SI-NEXT: s_waitcnt lgkmcnt(0)
|
|
|
|
|
; SI-NEXT: s_mov_b64 s[4:5], s[2:3]
|
|
|
|
|
; SI-NEXT: buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
|
|
|
|
|
; SI-NEXT: s_mov_b64 s[2:3], s[6:7]
|
|
|
|
|
; SI-NEXT: s_waitcnt vmcnt(0)
|
|
|
|
|
; SI-NEXT: v_cvt_pkrtz_f16_f32_e32 v2, 1.0, v2
|
|
|
|
|
; SI-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
|
|
|
|
|
; SI-NEXT: s_endpgm
|
|
|
|
|
;
|
|
|
|
|
; VI-LABEL: v_cvt_pkrtz_v2f16_f32_imm_reg:
|
|
|
|
|
; VI: ; %bb.0:
|
|
|
|
|
; VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
|
|
|
|
|
; VI-NEXT: v_lshlrev_b32_e32 v2, 2, v0
|
|
|
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
|
|
|
; VI-NEXT: v_mov_b32_e32 v1, s3
|
|
|
|
|
; VI-NEXT: v_add_u32_e32 v0, vcc, s2, v2
|
|
|
|
|
; VI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
|
[AMDGPU] Remove dubious logic in bidirectional list scheduler
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
|
|
|
; VI-NEXT: flat_load_dword v0, v[0:1]
|
|
|
|
|
; VI-NEXT: v_mov_b32_e32 v3, s1
|
|
|
|
|
; VI-NEXT: v_add_u32_e32 v2, vcc, s0, v2
|
|
|
|
|
; VI-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc
|
2019-07-19 23:43:56 +08:00
|
|
|
; VI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
[AMDGPU] Remove dubious logic in bidirectional list scheduler
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
|
|
|
; VI-NEXT: v_cvt_pkrtz_f16_f32 v0, 1.0, v0
|
|
|
|
|
; VI-NEXT: flat_store_dword v[2:3], v0
|
2019-07-19 23:43:56 +08:00
|
|
|
; VI-NEXT: s_endpgm
|
|
|
|
|
;
|
|
|
|
|
; GFX9-LABEL: v_cvt_pkrtz_v2f16_f32_imm_reg:
|
|
|
|
|
; GFX9: ; %bb.0:
|
|
|
|
|
; GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
|
|
|
|
|
; GFX9-NEXT: v_lshlrev_b32_e32 v2, 2, v0
|
|
|
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
|
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, s3
|
|
|
|
|
; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, s2, v2
|
|
|
|
|
; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
|
[AMDGPU] Remove dubious logic in bidirectional list scheduler
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
|
|
|
; GFX9-NEXT: global_load_dword v0, v[0:1], off
|
|
|
|
|
; GFX9-NEXT: v_mov_b32_e32 v3, s1
|
|
|
|
|
; GFX9-NEXT: v_add_co_u32_e32 v2, vcc, s0, v2
|
|
|
|
|
; GFX9-NEXT: v_addc_co_u32_e32 v3, vcc, 0, v3, vcc
|
2019-07-19 23:43:56 +08:00
|
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0)
|
[AMDGPU] Remove dubious logic in bidirectional list scheduler
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
|
|
|
; GFX9-NEXT: v_cvt_pkrtz_f16_f32 v0, 1.0, v0
|
|
|
|
|
; GFX9-NEXT: global_store_dword v[2:3], v0, off
|
2019-07-19 23:43:56 +08:00
|
|
|
; GFX9-NEXT: s_endpgm
|
2017-02-22 08:27:34 +08:00
|
|
|
%tid = call i32 @llvm.amdgcn.workitem.id.x()
|
|
|
|
|
%tid.ext = sext i32 %tid to i64
|
|
|
|
|
%a.gep = getelementptr inbounds float, float addrspace(1)* %a.ptr, i64 %tid.ext
|
|
|
|
|
%out.gep = getelementptr inbounds <2 x half>, <2 x half> addrspace(1)* %out, i64 %tid.ext
|
|
|
|
|
%a = load volatile float, float addrspace(1)* %a.gep
|
|
|
|
|
%cvt = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float 1.0, float %a)
|
|
|
|
|
store <2 x half> %cvt, <2 x half> addrspace(1)* %out.gep
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @v_cvt_pkrtz_v2f16_f32_fneg_lo(<2 x half> addrspace(1)* %out, float addrspace(1)* %a.ptr, float addrspace(1)* %b.ptr) #0 {
|
2019-07-19 23:43:56 +08:00
|
|
|
; SI-LABEL: v_cvt_pkrtz_v2f16_f32_fneg_lo:
|
|
|
|
|
; SI: ; %bb.0:
|
|
|
|
|
; SI-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x9
|
|
|
|
|
; SI-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
|
|
|
|
|
; SI-NEXT: s_mov_b32 s3, 0xf000
|
|
|
|
|
; SI-NEXT: s_mov_b32 s2, 0
|
|
|
|
|
; SI-NEXT: v_lshlrev_b32_e32 v0, 2, v0
|
|
|
|
|
; SI-NEXT: v_mov_b32_e32 v1, 0
|
|
|
|
|
; SI-NEXT: s_mov_b64 s[6:7], s[2:3]
|
|
|
|
|
; SI-NEXT: s_waitcnt lgkmcnt(0)
|
|
|
|
|
; SI-NEXT: s_mov_b64 s[0:1], s[10:11]
|
|
|
|
|
; SI-NEXT: buffer_load_dword v2, v[0:1], s[0:3], 0 addr64
|
|
|
|
|
; SI-NEXT: buffer_load_dword v3, v[0:1], s[4:7], 0 addr64
|
|
|
|
|
; SI-NEXT: s_mov_b64 s[10:11], s[2:3]
|
|
|
|
|
; SI-NEXT: s_waitcnt vmcnt(0)
|
|
|
|
|
; SI-NEXT: v_cvt_pkrtz_f16_f32_e64 v2, -v2, v3
|
|
|
|
|
; SI-NEXT: buffer_store_dword v2, v[0:1], s[8:11], 0 addr64
|
|
|
|
|
; SI-NEXT: s_endpgm
|
|
|
|
|
;
|
|
|
|
|
; VI-LABEL: v_cvt_pkrtz_v2f16_f32_fneg_lo:
|
|
|
|
|
; VI: ; %bb.0:
|
|
|
|
|
; VI-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
|
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
|
|
|
|
|
; VI-NEXT: v_lshlrev_b32_e32 v4, 2, v0
|
|
|
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
|
|
|
; VI-NEXT: v_mov_b32_e32 v1, s7
|
|
|
|
|
; VI-NEXT: v_add_u32_e32 v0, vcc, s6, v4
|
|
|
|
|
; VI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
|
|
|
|
|
; VI-NEXT: v_mov_b32_e32 v3, s1
|
|
|
|
|
; VI-NEXT: v_add_u32_e32 v2, vcc, s0, v4
|
|
|
|
|
; VI-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc
|
[AMDGPU] Remove dubious logic in bidirectional list scheduler
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
|
|
|
; VI-NEXT: flat_load_dword v0, v[0:1]
|
|
|
|
|
; VI-NEXT: flat_load_dword v1, v[2:3]
|
|
|
|
|
; VI-NEXT: v_mov_b32_e32 v5, s5
|
|
|
|
|
; VI-NEXT: v_add_u32_e32 v4, vcc, s4, v4
|
|
|
|
|
; VI-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc
|
2019-07-19 23:43:56 +08:00
|
|
|
; VI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
[AMDGPU] Remove dubious logic in bidirectional list scheduler
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
|
|
|
; VI-NEXT: v_cvt_pkrtz_f16_f32 v0, -v0, v1
|
|
|
|
|
; VI-NEXT: flat_store_dword v[4:5], v0
|
2019-07-19 23:43:56 +08:00
|
|
|
; VI-NEXT: s_endpgm
|
|
|
|
|
;
|
|
|
|
|
; GFX9-LABEL: v_cvt_pkrtz_v2f16_f32_fneg_lo:
|
|
|
|
|
; GFX9: ; %bb.0:
|
|
|
|
|
; GFX9-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
|
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
|
|
|
|
|
; GFX9-NEXT: v_lshlrev_b32_e32 v4, 2, v0
|
|
|
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
|
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, s7
|
|
|
|
|
; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, s6, v4
|
|
|
|
|
; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
|
|
|
|
|
; GFX9-NEXT: v_mov_b32_e32 v3, s1
|
|
|
|
|
; GFX9-NEXT: v_add_co_u32_e32 v2, vcc, s0, v4
|
|
|
|
|
; GFX9-NEXT: v_addc_co_u32_e32 v3, vcc, 0, v3, vcc
|
[AMDGPU] Remove dubious logic in bidirectional list scheduler
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
|
|
|
; GFX9-NEXT: global_load_dword v0, v[0:1], off
|
|
|
|
|
; GFX9-NEXT: global_load_dword v1, v[2:3], off
|
|
|
|
|
; GFX9-NEXT: v_mov_b32_e32 v5, s5
|
|
|
|
|
; GFX9-NEXT: v_add_co_u32_e32 v4, vcc, s4, v4
|
|
|
|
|
; GFX9-NEXT: v_addc_co_u32_e32 v5, vcc, 0, v5, vcc
|
2019-07-19 23:43:56 +08:00
|
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0)
|
[AMDGPU] Remove dubious logic in bidirectional list scheduler
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
|
|
|
; GFX9-NEXT: v_cvt_pkrtz_f16_f32 v0, -v0, v1
|
|
|
|
|
; GFX9-NEXT: global_store_dword v[4:5], v0, off
|
2019-07-19 23:43:56 +08:00
|
|
|
; GFX9-NEXT: s_endpgm
|
2017-02-22 08:27:34 +08:00
|
|
|
%tid = call i32 @llvm.amdgcn.workitem.id.x()
|
|
|
|
|
%tid.ext = sext i32 %tid to i64
|
|
|
|
|
%a.gep = getelementptr inbounds float, float addrspace(1)* %a.ptr, i64 %tid.ext
|
|
|
|
|
%b.gep = getelementptr inbounds float, float addrspace(1)* %b.ptr, i64 %tid.ext
|
|
|
|
|
%out.gep = getelementptr inbounds <2 x half>, <2 x half> addrspace(1)* %out, i64 %tid.ext
|
|
|
|
|
%a = load volatile float, float addrspace(1)* %a.gep
|
|
|
|
|
%b = load volatile float, float addrspace(1)* %b.gep
|
|
|
|
|
%neg.a = fsub float -0.0, %a
|
|
|
|
|
%cvt = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float %neg.a, float %b)
|
|
|
|
|
store <2 x half> %cvt, <2 x half> addrspace(1)* %out.gep
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @v_cvt_pkrtz_v2f16_f32_fneg_hi(<2 x half> addrspace(1)* %out, float addrspace(1)* %a.ptr, float addrspace(1)* %b.ptr) #0 {
|
2019-07-19 23:43:56 +08:00
|
|
|
; SI-LABEL: v_cvt_pkrtz_v2f16_f32_fneg_hi:
|
|
|
|
|
; SI: ; %bb.0:
|
|
|
|
|
; SI-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x9
|
|
|
|
|
; SI-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
|
|
|
|
|
; SI-NEXT: s_mov_b32 s3, 0xf000
|
|
|
|
|
; SI-NEXT: s_mov_b32 s2, 0
|
|
|
|
|
; SI-NEXT: v_lshlrev_b32_e32 v0, 2, v0
|
|
|
|
|
; SI-NEXT: v_mov_b32_e32 v1, 0
|
|
|
|
|
; SI-NEXT: s_mov_b64 s[6:7], s[2:3]
|
|
|
|
|
; SI-NEXT: s_waitcnt lgkmcnt(0)
|
|
|
|
|
; SI-NEXT: s_mov_b64 s[0:1], s[10:11]
|
|
|
|
|
; SI-NEXT: buffer_load_dword v2, v[0:1], s[0:3], 0 addr64
|
|
|
|
|
; SI-NEXT: buffer_load_dword v3, v[0:1], s[4:7], 0 addr64
|
|
|
|
|
; SI-NEXT: s_mov_b64 s[10:11], s[2:3]
|
|
|
|
|
; SI-NEXT: s_waitcnt vmcnt(0)
|
|
|
|
|
; SI-NEXT: v_cvt_pkrtz_f16_f32_e64 v2, v2, -v3
|
|
|
|
|
; SI-NEXT: buffer_store_dword v2, v[0:1], s[8:11], 0 addr64
|
|
|
|
|
; SI-NEXT: s_endpgm
|
|
|
|
|
;
|
|
|
|
|
; VI-LABEL: v_cvt_pkrtz_v2f16_f32_fneg_hi:
|
|
|
|
|
; VI: ; %bb.0:
|
|
|
|
|
; VI-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
|
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
|
|
|
|
|
; VI-NEXT: v_lshlrev_b32_e32 v4, 2, v0
|
|
|
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
|
|
|
; VI-NEXT: v_mov_b32_e32 v1, s7
|
|
|
|
|
; VI-NEXT: v_add_u32_e32 v0, vcc, s6, v4
|
|
|
|
|
; VI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
|
|
|
|
|
; VI-NEXT: v_mov_b32_e32 v3, s1
|
|
|
|
|
; VI-NEXT: v_add_u32_e32 v2, vcc, s0, v4
|
|
|
|
|
; VI-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc
|
[AMDGPU] Remove dubious logic in bidirectional list scheduler
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
|
|
|
; VI-NEXT: flat_load_dword v0, v[0:1]
|
|
|
|
|
; VI-NEXT: flat_load_dword v1, v[2:3]
|
|
|
|
|
; VI-NEXT: v_mov_b32_e32 v5, s5
|
|
|
|
|
; VI-NEXT: v_add_u32_e32 v4, vcc, s4, v4
|
|
|
|
|
; VI-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc
|
2019-07-19 23:43:56 +08:00
|
|
|
; VI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
[AMDGPU] Remove dubious logic in bidirectional list scheduler
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
|
|
|
; VI-NEXT: v_cvt_pkrtz_f16_f32 v0, v0, -v1
|
|
|
|
|
; VI-NEXT: flat_store_dword v[4:5], v0
|
2019-07-19 23:43:56 +08:00
|
|
|
; VI-NEXT: s_endpgm
|
|
|
|
|
;
|
|
|
|
|
; GFX9-LABEL: v_cvt_pkrtz_v2f16_f32_fneg_hi:
|
|
|
|
|
; GFX9: ; %bb.0:
|
|
|
|
|
; GFX9-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
|
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
|
|
|
|
|
; GFX9-NEXT: v_lshlrev_b32_e32 v4, 2, v0
|
|
|
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
|
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, s7
|
|
|
|
|
; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, s6, v4
|
|
|
|
|
; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
|
|
|
|
|
; GFX9-NEXT: v_mov_b32_e32 v3, s1
|
|
|
|
|
; GFX9-NEXT: v_add_co_u32_e32 v2, vcc, s0, v4
|
|
|
|
|
; GFX9-NEXT: v_addc_co_u32_e32 v3, vcc, 0, v3, vcc
|
[AMDGPU] Remove dubious logic in bidirectional list scheduler
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
|
|
|
; GFX9-NEXT: global_load_dword v0, v[0:1], off
|
|
|
|
|
; GFX9-NEXT: global_load_dword v1, v[2:3], off
|
|
|
|
|
; GFX9-NEXT: v_mov_b32_e32 v5, s5
|
|
|
|
|
; GFX9-NEXT: v_add_co_u32_e32 v4, vcc, s4, v4
|
|
|
|
|
; GFX9-NEXT: v_addc_co_u32_e32 v5, vcc, 0, v5, vcc
|
2019-07-19 23:43:56 +08:00
|
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0)
|
[AMDGPU] Remove dubious logic in bidirectional list scheduler
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
|
|
|
; GFX9-NEXT: v_cvt_pkrtz_f16_f32 v0, v0, -v1
|
|
|
|
|
; GFX9-NEXT: global_store_dword v[4:5], v0, off
|
2019-07-19 23:43:56 +08:00
|
|
|
; GFX9-NEXT: s_endpgm
|
2017-02-22 08:27:34 +08:00
|
|
|
%tid = call i32 @llvm.amdgcn.workitem.id.x()
|
|
|
|
|
%tid.ext = sext i32 %tid to i64
|
|
|
|
|
%a.gep = getelementptr inbounds float, float addrspace(1)* %a.ptr, i64 %tid.ext
|
|
|
|
|
%b.gep = getelementptr inbounds float, float addrspace(1)* %b.ptr, i64 %tid.ext
|
|
|
|
|
%out.gep = getelementptr inbounds <2 x half>, <2 x half> addrspace(1)* %out, i64 %tid.ext
|
|
|
|
|
%a = load volatile float, float addrspace(1)* %a.gep
|
|
|
|
|
%b = load volatile float, float addrspace(1)* %b.gep
|
|
|
|
|
%neg.b = fsub float -0.0, %b
|
|
|
|
|
%cvt = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float %a, float %neg.b)
|
|
|
|
|
store <2 x half> %cvt, <2 x half> addrspace(1)* %out.gep
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @v_cvt_pkrtz_v2f16_f32_fneg_lo_hi(<2 x half> addrspace(1)* %out, float addrspace(1)* %a.ptr, float addrspace(1)* %b.ptr) #0 {
|
2019-07-19 23:43:56 +08:00
|
|
|
; SI-LABEL: v_cvt_pkrtz_v2f16_f32_fneg_lo_hi:
|
|
|
|
|
; SI: ; %bb.0:
|
|
|
|
|
; SI-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x9
|
|
|
|
|
; SI-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
|
|
|
|
|
; SI-NEXT: s_mov_b32 s3, 0xf000
|
|
|
|
|
; SI-NEXT: s_mov_b32 s2, 0
|
|
|
|
|
; SI-NEXT: v_lshlrev_b32_e32 v0, 2, v0
|
|
|
|
|
; SI-NEXT: v_mov_b32_e32 v1, 0
|
|
|
|
|
; SI-NEXT: s_mov_b64 s[6:7], s[2:3]
|
|
|
|
|
; SI-NEXT: s_waitcnt lgkmcnt(0)
|
|
|
|
|
; SI-NEXT: s_mov_b64 s[0:1], s[10:11]
|
|
|
|
|
; SI-NEXT: buffer_load_dword v2, v[0:1], s[0:3], 0 addr64
|
|
|
|
|
; SI-NEXT: buffer_load_dword v3, v[0:1], s[4:7], 0 addr64
|
|
|
|
|
; SI-NEXT: s_mov_b64 s[10:11], s[2:3]
|
|
|
|
|
; SI-NEXT: s_waitcnt vmcnt(0)
|
|
|
|
|
; SI-NEXT: v_cvt_pkrtz_f16_f32_e64 v2, -v2, -v3
|
|
|
|
|
; SI-NEXT: buffer_store_dword v2, v[0:1], s[8:11], 0 addr64
|
|
|
|
|
; SI-NEXT: s_endpgm
|
|
|
|
|
;
|
|
|
|
|
; VI-LABEL: v_cvt_pkrtz_v2f16_f32_fneg_lo_hi:
|
|
|
|
|
; VI: ; %bb.0:
|
|
|
|
|
; VI-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
|
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
|
|
|
|
|
; VI-NEXT: v_lshlrev_b32_e32 v4, 2, v0
|
|
|
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
|
|
|
; VI-NEXT: v_mov_b32_e32 v1, s7
|
|
|
|
|
; VI-NEXT: v_add_u32_e32 v0, vcc, s6, v4
|
|
|
|
|
; VI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
|
|
|
|
|
; VI-NEXT: v_mov_b32_e32 v3, s1
|
|
|
|
|
; VI-NEXT: v_add_u32_e32 v2, vcc, s0, v4
|
|
|
|
|
; VI-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc
|
[AMDGPU] Remove dubious logic in bidirectional list scheduler
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
|
|
|
; VI-NEXT: flat_load_dword v0, v[0:1]
|
|
|
|
|
; VI-NEXT: flat_load_dword v1, v[2:3]
|
|
|
|
|
; VI-NEXT: v_mov_b32_e32 v5, s5
|
|
|
|
|
; VI-NEXT: v_add_u32_e32 v4, vcc, s4, v4
|
|
|
|
|
; VI-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc
|
2019-07-19 23:43:56 +08:00
|
|
|
; VI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
[AMDGPU] Remove dubious logic in bidirectional list scheduler
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
|
|
|
; VI-NEXT: v_cvt_pkrtz_f16_f32 v0, -v0, -v1
|
|
|
|
|
; VI-NEXT: flat_store_dword v[4:5], v0
|
2019-07-19 23:43:56 +08:00
|
|
|
; VI-NEXT: s_endpgm
|
|
|
|
|
;
|
|
|
|
|
; GFX9-LABEL: v_cvt_pkrtz_v2f16_f32_fneg_lo_hi:
|
|
|
|
|
; GFX9: ; %bb.0:
|
|
|
|
|
; GFX9-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
|
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
|
|
|
|
|
; GFX9-NEXT: v_lshlrev_b32_e32 v4, 2, v0
|
|
|
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
|
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, s7
|
|
|
|
|
; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, s6, v4
|
|
|
|
|
; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
|
|
|
|
|
; GFX9-NEXT: v_mov_b32_e32 v3, s1
|
|
|
|
|
; GFX9-NEXT: v_add_co_u32_e32 v2, vcc, s0, v4
|
|
|
|
|
; GFX9-NEXT: v_addc_co_u32_e32 v3, vcc, 0, v3, vcc
|
[AMDGPU] Remove dubious logic in bidirectional list scheduler
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
|
|
|
; GFX9-NEXT: global_load_dword v0, v[0:1], off
|
|
|
|
|
; GFX9-NEXT: global_load_dword v1, v[2:3], off
|
|
|
|
|
; GFX9-NEXT: v_mov_b32_e32 v5, s5
|
|
|
|
|
; GFX9-NEXT: v_add_co_u32_e32 v4, vcc, s4, v4
|
|
|
|
|
; GFX9-NEXT: v_addc_co_u32_e32 v5, vcc, 0, v5, vcc
|
2019-07-19 23:43:56 +08:00
|
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0)
|
[AMDGPU] Remove dubious logic in bidirectional list scheduler
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
|
|
|
; GFX9-NEXT: v_cvt_pkrtz_f16_f32 v0, -v0, -v1
|
|
|
|
|
; GFX9-NEXT: global_store_dword v[4:5], v0, off
|
2019-07-19 23:43:56 +08:00
|
|
|
; GFX9-NEXT: s_endpgm
|
2017-02-22 08:27:34 +08:00
|
|
|
%tid = call i32 @llvm.amdgcn.workitem.id.x()
|
|
|
|
|
%tid.ext = sext i32 %tid to i64
|
|
|
|
|
%a.gep = getelementptr inbounds float, float addrspace(1)* %a.ptr, i64 %tid.ext
|
|
|
|
|
%b.gep = getelementptr inbounds float, float addrspace(1)* %b.ptr, i64 %tid.ext
|
|
|
|
|
%out.gep = getelementptr inbounds <2 x half>, <2 x half> addrspace(1)* %out, i64 %tid.ext
|
|
|
|
|
%a = load volatile float, float addrspace(1)* %a.gep
|
|
|
|
|
%b = load volatile float, float addrspace(1)* %b.gep
|
|
|
|
|
%neg.a = fsub float -0.0, %a
|
|
|
|
|
%neg.b = fsub float -0.0, %b
|
|
|
|
|
%cvt = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float %neg.a, float %neg.b)
|
|
|
|
|
store <2 x half> %cvt, <2 x half> addrspace(1)* %out.gep
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @v_cvt_pkrtz_v2f16_f32_fneg_fabs_lo_fneg_hi(<2 x half> addrspace(1)* %out, float addrspace(1)* %a.ptr, float addrspace(1)* %b.ptr) #0 {
|
2019-07-19 23:43:56 +08:00
|
|
|
; SI-LABEL: v_cvt_pkrtz_v2f16_f32_fneg_fabs_lo_fneg_hi:
|
|
|
|
|
; SI: ; %bb.0:
|
|
|
|
|
; SI-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x9
|
|
|
|
|
; SI-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
|
|
|
|
|
; SI-NEXT: s_mov_b32 s3, 0xf000
|
|
|
|
|
; SI-NEXT: s_mov_b32 s2, 0
|
|
|
|
|
; SI-NEXT: v_lshlrev_b32_e32 v0, 2, v0
|
|
|
|
|
; SI-NEXT: v_mov_b32_e32 v1, 0
|
|
|
|
|
; SI-NEXT: s_mov_b64 s[6:7], s[2:3]
|
|
|
|
|
; SI-NEXT: s_waitcnt lgkmcnt(0)
|
|
|
|
|
; SI-NEXT: s_mov_b64 s[0:1], s[10:11]
|
|
|
|
|
; SI-NEXT: buffer_load_dword v2, v[0:1], s[0:3], 0 addr64
|
|
|
|
|
; SI-NEXT: buffer_load_dword v3, v[0:1], s[4:7], 0 addr64
|
|
|
|
|
; SI-NEXT: s_mov_b64 s[10:11], s[2:3]
|
|
|
|
|
; SI-NEXT: s_waitcnt vmcnt(0)
|
|
|
|
|
; SI-NEXT: v_cvt_pkrtz_f16_f32_e64 v2, -|v2|, -v3
|
|
|
|
|
; SI-NEXT: buffer_store_dword v2, v[0:1], s[8:11], 0 addr64
|
|
|
|
|
; SI-NEXT: s_endpgm
|
|
|
|
|
;
|
|
|
|
|
; VI-LABEL: v_cvt_pkrtz_v2f16_f32_fneg_fabs_lo_fneg_hi:
|
|
|
|
|
; VI: ; %bb.0:
|
|
|
|
|
; VI-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
|
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
|
|
|
|
|
; VI-NEXT: v_lshlrev_b32_e32 v4, 2, v0
|
|
|
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
|
|
|
; VI-NEXT: v_mov_b32_e32 v1, s7
|
|
|
|
|
; VI-NEXT: v_add_u32_e32 v0, vcc, s6, v4
|
|
|
|
|
; VI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
|
|
|
|
|
; VI-NEXT: v_mov_b32_e32 v3, s1
|
|
|
|
|
; VI-NEXT: v_add_u32_e32 v2, vcc, s0, v4
|
|
|
|
|
; VI-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc
|
[AMDGPU] Remove dubious logic in bidirectional list scheduler
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
|
|
|
; VI-NEXT: flat_load_dword v0, v[0:1]
|
|
|
|
|
; VI-NEXT: flat_load_dword v1, v[2:3]
|
|
|
|
|
; VI-NEXT: v_mov_b32_e32 v5, s5
|
|
|
|
|
; VI-NEXT: v_add_u32_e32 v4, vcc, s4, v4
|
|
|
|
|
; VI-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc
|
2019-07-19 23:43:56 +08:00
|
|
|
; VI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
[AMDGPU] Remove dubious logic in bidirectional list scheduler
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
|
|
|
; VI-NEXT: v_cvt_pkrtz_f16_f32 v0, -|v0|, -v1
|
|
|
|
|
; VI-NEXT: flat_store_dword v[4:5], v0
|
2019-07-19 23:43:56 +08:00
|
|
|
; VI-NEXT: s_endpgm
|
|
|
|
|
;
|
|
|
|
|
; GFX9-LABEL: v_cvt_pkrtz_v2f16_f32_fneg_fabs_lo_fneg_hi:
|
|
|
|
|
; GFX9: ; %bb.0:
|
|
|
|
|
; GFX9-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
|
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
|
|
|
|
|
; GFX9-NEXT: v_lshlrev_b32_e32 v4, 2, v0
|
|
|
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
|
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, s7
|
|
|
|
|
; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, s6, v4
|
|
|
|
|
; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
|
|
|
|
|
; GFX9-NEXT: v_mov_b32_e32 v3, s1
|
|
|
|
|
; GFX9-NEXT: v_add_co_u32_e32 v2, vcc, s0, v4
|
|
|
|
|
; GFX9-NEXT: v_addc_co_u32_e32 v3, vcc, 0, v3, vcc
|
[AMDGPU] Remove dubious logic in bidirectional list scheduler
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
|
|
|
; GFX9-NEXT: global_load_dword v0, v[0:1], off
|
|
|
|
|
; GFX9-NEXT: global_load_dword v1, v[2:3], off
|
|
|
|
|
; GFX9-NEXT: v_mov_b32_e32 v5, s5
|
|
|
|
|
; GFX9-NEXT: v_add_co_u32_e32 v4, vcc, s4, v4
|
|
|
|
|
; GFX9-NEXT: v_addc_co_u32_e32 v5, vcc, 0, v5, vcc
|
2019-07-19 23:43:56 +08:00
|
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0)
|
[AMDGPU] Remove dubious logic in bidirectional list scheduler
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
|
|
|
; GFX9-NEXT: v_cvt_pkrtz_f16_f32 v0, -|v0|, -v1
|
|
|
|
|
; GFX9-NEXT: global_store_dword v[4:5], v0, off
|
2019-07-19 23:43:56 +08:00
|
|
|
; GFX9-NEXT: s_endpgm
|
2017-02-22 08:27:34 +08:00
|
|
|
%tid = call i32 @llvm.amdgcn.workitem.id.x()
|
|
|
|
|
%tid.ext = sext i32 %tid to i64
|
|
|
|
|
%a.gep = getelementptr inbounds float, float addrspace(1)* %a.ptr, i64 %tid.ext
|
|
|
|
|
%b.gep = getelementptr inbounds float, float addrspace(1)* %b.ptr, i64 %tid.ext
|
|
|
|
|
%out.gep = getelementptr inbounds <2 x half>, <2 x half> addrspace(1)* %out, i64 %tid.ext
|
|
|
|
|
%a = load volatile float, float addrspace(1)* %a.gep
|
|
|
|
|
%b = load volatile float, float addrspace(1)* %b.gep
|
|
|
|
|
%fabs.a = call float @llvm.fabs.f32(float %a)
|
|
|
|
|
%neg.fabs.a = fsub float -0.0, %fabs.a
|
|
|
|
|
%neg.b = fsub float -0.0, %b
|
|
|
|
|
%cvt = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float %neg.fabs.a, float %neg.b)
|
|
|
|
|
store <2 x half> %cvt, <2 x half> addrspace(1)* %out.gep
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
declare <2 x half> @llvm.amdgcn.cvt.pkrtz(float, float) #1
|
|
|
|
|
declare float @llvm.fabs.f32(float) #1
|
|
|
|
|
declare i32 @llvm.amdgcn.workitem.id.x() #1
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
attributes #0 = { nounwind }
|
|
|
|
|
attributes #1 = { nounwind readnone }
|