2019-05-10 08:09:01 +08:00
|
|
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
|
|
|
; RUN: llc < %s -mtriple=amdgcn-amd-mesa3d -mcpu=gfx900 -verify-machineinstrs | FileCheck -check-prefix=GFX9 %s
|
|
|
|
; RUN: llc < %s -mtriple=amdgcn-amd-mesa3d -mcpu=gfx1010 -verify-machineinstrs | FileCheck -check-prefix=GFX10 %s
|
|
|
|
|
|
|
|
; ===================================================================================
|
|
|
|
; V_XOR3_B32
|
|
|
|
; ===================================================================================
|
|
|
|
|
|
|
|
define amdgpu_ps float @xor3(i32 %a, i32 %b, i32 %c) {
|
|
|
|
; GFX9-LABEL: xor3:
|
|
|
|
; GFX9: ; %bb.0:
|
|
|
|
; GFX9-NEXT: v_xor_b32_e32 v0, v0, v1
|
|
|
|
; GFX9-NEXT: v_xor_b32_e32 v0, v0, v2
|
|
|
|
; GFX9-NEXT: ; return to shader part epilog
|
|
|
|
;
|
|
|
|
; GFX10-LABEL: xor3:
|
|
|
|
; GFX10: ; %bb.0:
|
|
|
|
; GFX10-NEXT: v_xor3_b32 v0, v0, v1, v2
|
2019-06-20 23:08:34 +08:00
|
|
|
; GFX10-NEXT: ; implicit-def: $vcc_hi
|
2019-05-10 08:09:01 +08:00
|
|
|
; GFX10-NEXT: ; return to shader part epilog
|
|
|
|
%x = xor i32 %a, %b
|
|
|
|
%result = xor i32 %x, %c
|
|
|
|
%bc = bitcast i32 %result to float
|
|
|
|
ret float %bc
|
|
|
|
}
|
|
|
|
|
|
|
|
define amdgpu_ps float @xor3_vgpr_b(i32 inreg %a, i32 %b, i32 inreg %c) {
|
|
|
|
; GFX9-LABEL: xor3_vgpr_b:
|
|
|
|
; GFX9: ; %bb.0:
|
|
|
|
; GFX9-NEXT: v_xor_b32_e32 v0, s2, v0
|
|
|
|
; GFX9-NEXT: v_xor_b32_e32 v0, s3, v0
|
|
|
|
; GFX9-NEXT: ; return to shader part epilog
|
|
|
|
;
|
|
|
|
; GFX10-LABEL: xor3_vgpr_b:
|
|
|
|
; GFX10: ; %bb.0:
|
|
|
|
; GFX10-NEXT: v_xor3_b32 v0, s2, v0, s3
|
2019-06-20 23:08:34 +08:00
|
|
|
; GFX10-NEXT: ; implicit-def: $vcc_hi
|
2019-05-10 08:09:01 +08:00
|
|
|
; GFX10-NEXT: ; return to shader part epilog
|
|
|
|
%x = xor i32 %a, %b
|
|
|
|
%result = xor i32 %x, %c
|
|
|
|
%bc = bitcast i32 %result to float
|
|
|
|
ret float %bc
|
|
|
|
}
|
|
|
|
|
|
|
|
define amdgpu_ps float @xor3_vgpr_all2(i32 %a, i32 %b, i32 %c) {
|
|
|
|
; GFX9-LABEL: xor3_vgpr_all2:
|
|
|
|
; GFX9: ; %bb.0:
|
|
|
|
; GFX9-NEXT: v_xor_b32_e32 v1, v1, v2
|
|
|
|
; GFX9-NEXT: v_xor_b32_e32 v0, v0, v1
|
|
|
|
; GFX9-NEXT: ; return to shader part epilog
|
|
|
|
;
|
|
|
|
; GFX10-LABEL: xor3_vgpr_all2:
|
|
|
|
; GFX10: ; %bb.0:
|
|
|
|
; GFX10-NEXT: v_xor3_b32 v0, v1, v2, v0
|
2019-06-20 23:08:34 +08:00
|
|
|
; GFX10-NEXT: ; implicit-def: $vcc_hi
|
2019-05-10 08:09:01 +08:00
|
|
|
; GFX10-NEXT: ; return to shader part epilog
|
|
|
|
%x = xor i32 %b, %c
|
|
|
|
%result = xor i32 %a, %x
|
|
|
|
%bc = bitcast i32 %result to float
|
|
|
|
ret float %bc
|
|
|
|
}
|
|
|
|
|
|
|
|
define amdgpu_ps float @xor3_vgpr_bc(i32 inreg %a, i32 %b, i32 %c) {
|
|
|
|
; GFX9-LABEL: xor3_vgpr_bc:
|
|
|
|
; GFX9: ; %bb.0:
|
|
|
|
; GFX9-NEXT: v_xor_b32_e32 v0, s2, v0
|
|
|
|
; GFX9-NEXT: v_xor_b32_e32 v0, v0, v1
|
|
|
|
; GFX9-NEXT: ; return to shader part epilog
|
|
|
|
;
|
|
|
|
; GFX10-LABEL: xor3_vgpr_bc:
|
|
|
|
; GFX10: ; %bb.0:
|
|
|
|
; GFX10-NEXT: v_xor3_b32 v0, s2, v0, v1
|
2019-06-20 23:08:34 +08:00
|
|
|
; GFX10-NEXT: ; implicit-def: $vcc_hi
|
2019-05-10 08:09:01 +08:00
|
|
|
; GFX10-NEXT: ; return to shader part epilog
|
|
|
|
%x = xor i32 %a, %b
|
|
|
|
%result = xor i32 %x, %c
|
|
|
|
%bc = bitcast i32 %result to float
|
|
|
|
ret float %bc
|
|
|
|
}
|
|
|
|
|
|
|
|
define amdgpu_ps float @xor3_vgpr_const(i32 %a, i32 %b) {
|
|
|
|
; GFX9-LABEL: xor3_vgpr_const:
|
|
|
|
; GFX9: ; %bb.0:
|
|
|
|
; GFX9-NEXT: v_xor_b32_e32 v0, v0, v1
|
|
|
|
; GFX9-NEXT: v_xor_b32_e32 v0, 16, v0
|
|
|
|
; GFX9-NEXT: ; return to shader part epilog
|
|
|
|
;
|
|
|
|
; GFX10-LABEL: xor3_vgpr_const:
|
|
|
|
; GFX10: ; %bb.0:
|
|
|
|
; GFX10-NEXT: v_xor3_b32 v0, v0, v1, 16
|
2019-06-20 23:08:34 +08:00
|
|
|
; GFX10-NEXT: ; implicit-def: $vcc_hi
|
2019-05-10 08:09:01 +08:00
|
|
|
; GFX10-NEXT: ; return to shader part epilog
|
|
|
|
%x = xor i32 %a, %b
|
|
|
|
%result = xor i32 %x, 16
|
|
|
|
%bc = bitcast i32 %result to float
|
|
|
|
ret float %bc
|
|
|
|
}
|
|
|
|
|
|
|
|
define amdgpu_ps <2 x float> @xor3_multiuse_outer(i32 %a, i32 %b, i32 %c, i32 %x) {
|
|
|
|
; GFX9-LABEL: xor3_multiuse_outer:
|
|
|
|
; GFX9: ; %bb.0:
|
|
|
|
; GFX9-NEXT: v_xor_b32_e32 v0, v0, v1
|
|
|
|
; GFX9-NEXT: v_xor_b32_e32 v0, v0, v2
|
|
|
|
; GFX9-NEXT: v_mul_lo_u32 v1, v0, v3
|
|
|
|
; GFX9-NEXT: ; return to shader part epilog
|
|
|
|
;
|
|
|
|
; GFX10-LABEL: xor3_multiuse_outer:
|
|
|
|
; GFX10: ; %bb.0:
|
|
|
|
; GFX10-NEXT: v_xor3_b32 v0, v0, v1, v2
|
2019-06-20 23:08:34 +08:00
|
|
|
; GFX10-NEXT: ; implicit-def: $vcc_hi
|
2019-05-10 08:09:01 +08:00
|
|
|
; GFX10-NEXT: v_mul_lo_u32 v1, v0, v3
|
|
|
|
; GFX10-NEXT: ; return to shader part epilog
|
|
|
|
%inner = xor i32 %a, %b
|
|
|
|
%outer = xor i32 %inner, %c
|
|
|
|
%x1 = mul i32 %outer, %x
|
|
|
|
%r1 = insertelement <2 x i32> undef, i32 %outer, i32 0
|
|
|
|
%r0 = insertelement <2 x i32> %r1, i32 %x1, i32 1
|
|
|
|
%bc = bitcast <2 x i32> %r0 to <2 x float>
|
|
|
|
ret <2 x float> %bc
|
|
|
|
}
|
|
|
|
|
|
|
|
define amdgpu_ps <2 x float> @xor3_multiuse_inner(i32 %a, i32 %b, i32 %c) {
|
|
|
|
; GFX9-LABEL: xor3_multiuse_inner:
|
|
|
|
; GFX9: ; %bb.0:
|
|
|
|
; GFX9-NEXT: v_xor_b32_e32 v0, v0, v1
|
|
|
|
; GFX9-NEXT: v_xor_b32_e32 v1, v0, v2
|
|
|
|
; GFX9-NEXT: ; return to shader part epilog
|
|
|
|
;
|
|
|
|
; GFX10-LABEL: xor3_multiuse_inner:
|
|
|
|
; GFX10: ; %bb.0:
|
|
|
|
; GFX10-NEXT: v_xor_b32_e32 v0, v0, v1
|
2019-06-20 23:08:34 +08:00
|
|
|
; GFX10-NEXT: ; implicit-def: $vcc_hi
|
2019-05-10 08:09:01 +08:00
|
|
|
; GFX10-NEXT: v_xor_b32_e32 v1, v0, v2
|
|
|
|
; GFX10-NEXT: ; return to shader part epilog
|
|
|
|
%inner = xor i32 %a, %b
|
|
|
|
%outer = xor i32 %inner, %c
|
|
|
|
%r1 = insertelement <2 x i32> undef, i32 %inner, i32 0
|
|
|
|
%r0 = insertelement <2 x i32> %r1, i32 %outer, i32 1
|
|
|
|
%bc = bitcast <2 x i32> %r0 to <2 x float>
|
|
|
|
ret <2 x float> %bc
|
|
|
|
}
|
|
|
|
|
|
|
|
; A case where uniform values end up in VGPRs -- we could use v_xor3_b32 here,
|
|
|
|
; but we don't.
|
|
|
|
define amdgpu_ps float @xor3_uniform_vgpr(float inreg %a, float inreg %b, float inreg %c) {
|
|
|
|
; GFX9-LABEL: xor3_uniform_vgpr:
|
|
|
|
; GFX9: ; %bb.0:
|
|
|
|
; GFX9-NEXT: v_mov_b32_e32 v2, 0x40400000
|
|
|
|
; GFX9-NEXT: v_add_f32_e64 v0, s2, 1.0
|
|
|
|
; GFX9-NEXT: v_add_f32_e64 v1, s3, 2.0
|
|
|
|
; GFX9-NEXT: v_add_f32_e32 v2, s4, v2
|
|
|
|
; GFX9-NEXT: v_xor_b32_e32 v0, v0, v1
|
|
|
|
; GFX9-NEXT: v_xor_b32_e32 v0, v0, v2
|
|
|
|
; GFX9-NEXT: ; return to shader part epilog
|
|
|
|
;
|
|
|
|
; GFX10-LABEL: xor3_uniform_vgpr:
|
|
|
|
; GFX10: ; %bb.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
|
|
|
; GFX10-NEXT: v_add_f32_e64 v0, s2, 1.0
|
2019-05-10 08:09:01 +08:00
|
|
|
; GFX10-NEXT: v_add_f32_e64 v1, s3, 2.0
|
[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
|
|
|
; GFX10-NEXT: v_add_f32_e64 v2, 0x40400000, s4
|
2019-06-20 23:08:34 +08:00
|
|
|
; GFX10-NEXT: ; implicit-def: $vcc_hi
|
[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
|
|
|
; GFX10-NEXT: v_xor_b32_e32 v0, v0, v1
|
|
|
|
; GFX10-NEXT: v_xor_b32_e32 v0, v0, v2
|
2019-05-10 08:09:01 +08:00
|
|
|
; GFX10-NEXT: ; return to shader part epilog
|
|
|
|
%a1 = fadd float %a, 1.0
|
|
|
|
%b2 = fadd float %b, 2.0
|
|
|
|
%c3 = fadd float %c, 3.0
|
|
|
|
%bc.a = bitcast float %a1 to i32
|
|
|
|
%bc.b = bitcast float %b2 to i32
|
|
|
|
%bc.c = bitcast float %c3 to i32
|
|
|
|
%x = xor i32 %bc.a, %bc.b
|
|
|
|
%result = xor i32 %x, %bc.c
|
|
|
|
%bc = bitcast i32 %result to float
|
|
|
|
ret float %bc
|
|
|
|
}
|