1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-23 19:23:23 +01:00
llvm-mirror/test/CodeGen/AMDGPU/fmax_legacy.f64.ll
Jay Foad 7622b45921 [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
2020-02-28 21:35:34 +00:00

219 lines
8.3 KiB
LLVM

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=amdgcn-- -mcpu=tahiti < %s | FileCheck -check-prefix=SI %s
; RUN: llc -mtriple=amdgcn-- -mcpu=fiji < %s | FileCheck -check-prefix=VI %s
; Make sure we don't try to form FMAX_LEGACY nodes with f64
define amdgpu_kernel void @test_fmax_legacy_uge_f64(double addrspace(1)* %out, double addrspace(1)* %in) #0 {
; SI-LABEL: test_fmax_legacy_uge_f64:
; SI: ; %bb.0:
; SI-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT: s_mov_b32 s3, 0xf000
; SI-NEXT: s_mov_b32 s10, 0
; SI-NEXT: s_mov_b32 s11, s3
; SI-NEXT: v_lshlrev_b32_e32 v0, 3, v0
; SI-NEXT: s_waitcnt lgkmcnt(0)
; SI-NEXT: s_mov_b64 s[8:9], s[6:7]
; SI-NEXT: v_mov_b32_e32 v1, 0
; SI-NEXT: buffer_load_dwordx4 v[0:3], v[0:1], s[8:11], 0 addr64
; SI-NEXT: s_mov_b32 s2, -1
; SI-NEXT: s_mov_b32 s0, s4
; SI-NEXT: s_mov_b32 s1, s5
; SI-NEXT: s_waitcnt vmcnt(0)
; SI-NEXT: v_cmp_nlt_f64_e32 vcc, v[0:1], v[2:3]
; SI-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc
; SI-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; SI-NEXT: s_endpgm
;
; VI-LABEL: test_fmax_legacy_uge_f64:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
; VI-NEXT: v_lshlrev_b32_e32 v0, 3, v0
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v1, s3
; VI-NEXT: v_add_u32_e32 v0, vcc, s2, v0
; VI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT: flat_load_dwordx4 v[0:3], v[0:1]
; VI-NEXT: v_mov_b32_e32 v4, s0
; VI-NEXT: v_mov_b32_e32 v5, s1
; VI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_nlt_f64_e32 vcc, v[0:1], v[2:3]
; VI-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc
; VI-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
; VI-NEXT: flat_store_dwordx2 v[4:5], v[0:1]
; VI-NEXT: s_endpgm
%tid = call i32 @llvm.amdgcn.workitem.id.x() #1
%gep.0 = getelementptr double, double addrspace(1)* %in, i32 %tid
%gep.1 = getelementptr double, double addrspace(1)* %gep.0, i32 1
%a = load double, double addrspace(1)* %gep.0, align 8
%b = load double, double addrspace(1)* %gep.1, align 8
%cmp = fcmp uge double %a, %b
%val = select i1 %cmp, double %a, double %b
store double %val, double addrspace(1)* %out, align 8
ret void
}
define amdgpu_kernel void @test_fmax_legacy_oge_f64(double addrspace(1)* %out, double addrspace(1)* %in) #0 {
; SI-LABEL: test_fmax_legacy_oge_f64:
; SI: ; %bb.0:
; SI-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT: s_mov_b32 s3, 0xf000
; SI-NEXT: s_mov_b32 s10, 0
; SI-NEXT: s_mov_b32 s11, s3
; SI-NEXT: v_lshlrev_b32_e32 v0, 3, v0
; SI-NEXT: s_waitcnt lgkmcnt(0)
; SI-NEXT: s_mov_b64 s[8:9], s[6:7]
; SI-NEXT: v_mov_b32_e32 v1, 0
; SI-NEXT: buffer_load_dwordx4 v[0:3], v[0:1], s[8:11], 0 addr64
; SI-NEXT: s_mov_b32 s2, -1
; SI-NEXT: s_mov_b32 s0, s4
; SI-NEXT: s_mov_b32 s1, s5
; SI-NEXT: s_waitcnt vmcnt(0)
; SI-NEXT: v_cmp_ge_f64_e32 vcc, v[0:1], v[2:3]
; SI-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc
; SI-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; SI-NEXT: s_endpgm
;
; VI-LABEL: test_fmax_legacy_oge_f64:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
; VI-NEXT: v_lshlrev_b32_e32 v0, 3, v0
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v1, s3
; VI-NEXT: v_add_u32_e32 v0, vcc, s2, v0
; VI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT: flat_load_dwordx4 v[0:3], v[0:1]
; VI-NEXT: v_mov_b32_e32 v4, s0
; VI-NEXT: v_mov_b32_e32 v5, s1
; VI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_ge_f64_e32 vcc, v[0:1], v[2:3]
; VI-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc
; VI-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
; VI-NEXT: flat_store_dwordx2 v[4:5], v[0:1]
; VI-NEXT: s_endpgm
%tid = call i32 @llvm.amdgcn.workitem.id.x() #1
%gep.0 = getelementptr double, double addrspace(1)* %in, i32 %tid
%gep.1 = getelementptr double, double addrspace(1)* %gep.0, i32 1
%a = load double, double addrspace(1)* %gep.0, align 8
%b = load double, double addrspace(1)* %gep.1, align 8
%cmp = fcmp oge double %a, %b
%val = select i1 %cmp, double %a, double %b
store double %val, double addrspace(1)* %out, align 8
ret void
}
define amdgpu_kernel void @test_fmax_legacy_ugt_f64(double addrspace(1)* %out, double addrspace(1)* %in) #0 {
; SI-LABEL: test_fmax_legacy_ugt_f64:
; SI: ; %bb.0:
; SI-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT: s_mov_b32 s3, 0xf000
; SI-NEXT: s_mov_b32 s10, 0
; SI-NEXT: s_mov_b32 s11, s3
; SI-NEXT: v_lshlrev_b32_e32 v0, 3, v0
; SI-NEXT: s_waitcnt lgkmcnt(0)
; SI-NEXT: s_mov_b64 s[8:9], s[6:7]
; SI-NEXT: v_mov_b32_e32 v1, 0
; SI-NEXT: buffer_load_dwordx4 v[0:3], v[0:1], s[8:11], 0 addr64
; SI-NEXT: s_mov_b32 s2, -1
; SI-NEXT: s_mov_b32 s0, s4
; SI-NEXT: s_mov_b32 s1, s5
; SI-NEXT: s_waitcnt vmcnt(0)
; SI-NEXT: v_cmp_nle_f64_e32 vcc, v[0:1], v[2:3]
; SI-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc
; SI-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; SI-NEXT: s_endpgm
;
; VI-LABEL: test_fmax_legacy_ugt_f64:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
; VI-NEXT: v_lshlrev_b32_e32 v0, 3, v0
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v1, s3
; VI-NEXT: v_add_u32_e32 v0, vcc, s2, v0
; VI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT: flat_load_dwordx4 v[0:3], v[0:1]
; VI-NEXT: v_mov_b32_e32 v4, s0
; VI-NEXT: v_mov_b32_e32 v5, s1
; VI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_nle_f64_e32 vcc, v[0:1], v[2:3]
; VI-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc
; VI-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
; VI-NEXT: flat_store_dwordx2 v[4:5], v[0:1]
; VI-NEXT: s_endpgm
%tid = call i32 @llvm.amdgcn.workitem.id.x() #1
%gep.0 = getelementptr double, double addrspace(1)* %in, i32 %tid
%gep.1 = getelementptr double, double addrspace(1)* %gep.0, i32 1
%a = load double, double addrspace(1)* %gep.0, align 8
%b = load double, double addrspace(1)* %gep.1, align 8
%cmp = fcmp ugt double %a, %b
%val = select i1 %cmp, double %a, double %b
store double %val, double addrspace(1)* %out, align 8
ret void
}
define amdgpu_kernel void @test_fmax_legacy_ogt_f64(double addrspace(1)* %out, double addrspace(1)* %in) #0 {
; SI-LABEL: test_fmax_legacy_ogt_f64:
; SI: ; %bb.0:
; SI-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT: s_mov_b32 s3, 0xf000
; SI-NEXT: s_mov_b32 s10, 0
; SI-NEXT: s_mov_b32 s11, s3
; SI-NEXT: v_lshlrev_b32_e32 v0, 3, v0
; SI-NEXT: s_waitcnt lgkmcnt(0)
; SI-NEXT: s_mov_b64 s[8:9], s[6:7]
; SI-NEXT: v_mov_b32_e32 v1, 0
; SI-NEXT: buffer_load_dwordx4 v[0:3], v[0:1], s[8:11], 0 addr64
; SI-NEXT: s_mov_b32 s2, -1
; SI-NEXT: s_mov_b32 s0, s4
; SI-NEXT: s_mov_b32 s1, s5
; SI-NEXT: s_waitcnt vmcnt(0)
; SI-NEXT: v_cmp_gt_f64_e32 vcc, v[0:1], v[2:3]
; SI-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc
; SI-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; SI-NEXT: s_endpgm
;
; VI-LABEL: test_fmax_legacy_ogt_f64:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
; VI-NEXT: v_lshlrev_b32_e32 v0, 3, v0
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v1, s3
; VI-NEXT: v_add_u32_e32 v0, vcc, s2, v0
; VI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT: flat_load_dwordx4 v[0:3], v[0:1]
; VI-NEXT: v_mov_b32_e32 v4, s0
; VI-NEXT: v_mov_b32_e32 v5, s1
; VI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_gt_f64_e32 vcc, v[0:1], v[2:3]
; VI-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc
; VI-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
; VI-NEXT: flat_store_dwordx2 v[4:5], v[0:1]
; VI-NEXT: s_endpgm
%tid = call i32 @llvm.amdgcn.workitem.id.x() #1
%gep.0 = getelementptr double, double addrspace(1)* %in, i32 %tid
%gep.1 = getelementptr double, double addrspace(1)* %gep.0, i32 1
%a = load double, double addrspace(1)* %gep.0, align 8
%b = load double, double addrspace(1)* %gep.1, align 8
%cmp = fcmp ogt double %a, %b
%val = select i1 %cmp, double %a, double %b
store double %val, double addrspace(1)* %out, align 8
ret void
}
declare i32 @llvm.amdgcn.workitem.id.x() #1
attributes #0 = { nounwind }
attributes #1 = { nounwind readnone }