02e60f2e77
The scheduler should set critical/excess register usage thresholds that are guided by the maximum possible occupancy for the function. This change is focused on setting proper lower bounds on register usage which we would typically only see when a specific number of maximum waves is requested with the "waves-per-eu" attribute, or by setting "amdgpu-num-vgpr|sgpr" directly. This was broken previously. I have a follow-on patch that will address issues with the scheduler not targeting correct upper bounds on register usage which is typical with launch bounds and min "waves-per-eu". Changes by this patch: Set the initial critical register usage thresholds to minimum values that are determined by the maximum possible occupancy for the function, or the number of allocatable registers, whichever is lower. Avoid unisgned overflow if register limits are lower than the register tracking "ErrorMargin", I.e. when using stress-regalloc=2. Reviewed By: arsenm Differential Revision: https://reviews.llvm.org/D112373
111 lines
6.7 KiB
LLVM
111 lines
6.7 KiB
LLVM
; REQUIRES: asserts
|
|
|
|
; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs -debug-only=machine-scheduler -o /dev/null < %s 2>&1 | FileCheck %s
|
|
|
|
; We are only targeting one wave. Check that the machine scheduler doesn't use
|
|
; register pressure heuristics to prioritize any candidate instruction.
|
|
|
|
; CHECK-NOT: REG-CRIT
|
|
; CHECK-NOT: REG-EXCESS
|
|
|
|
define amdgpu_kernel void @load_fma_store(float addrspace(3)* nocapture readonly %arg, float addrspace(3)* nocapture %arg1) #1 {
|
|
bb:
|
|
%tmp0 = getelementptr inbounds float, float addrspace(3)* %arg, i32 1
|
|
%tmp1 = load float, float addrspace(3)* %tmp0, align 4
|
|
%tmp2 = getelementptr inbounds float, float addrspace(3)* %arg, i32 2
|
|
%tmp3 = load float, float addrspace(3)* %tmp2, align 4
|
|
%tmp4 = getelementptr inbounds float, float addrspace(3)* %arg, i32 3
|
|
%tmp5 = load float, float addrspace(3)* %tmp4, align 4
|
|
%tmp6 = getelementptr inbounds float, float addrspace(3)* %arg, i32 4
|
|
%tmp7 = load float, float addrspace(3)* %tmp6, align 4
|
|
%tmp8 = getelementptr inbounds float, float addrspace(3)* %arg, i32 5
|
|
%tmp9 = load float, float addrspace(3)* %tmp8, align 4
|
|
%tmp10 = getelementptr inbounds float, float addrspace(3)* %arg, i32 6
|
|
%tmp11 = load float, float addrspace(3)* %tmp10, align 4
|
|
%tmp12 = getelementptr inbounds float, float addrspace(3)* %arg, i32 7
|
|
%tmp13 = load float, float addrspace(3)* %tmp12, align 4
|
|
%tmp14 = getelementptr inbounds float, float addrspace(3)* %arg, i32 8
|
|
%tmp15 = load float, float addrspace(3)* %tmp14, align 4
|
|
%tmp16 = getelementptr inbounds float, float addrspace(3)* %arg, i32 9
|
|
%tmp17 = load float, float addrspace(3)* %tmp16, align 4
|
|
%tmp18 = getelementptr inbounds float, float addrspace(3)* %arg, i32 10
|
|
%tmp19 = load float, float addrspace(3)* %tmp18, align 4
|
|
%tmp20 = getelementptr inbounds float, float addrspace(3)* %arg, i32 11
|
|
%tmp21 = load float, float addrspace(3)* %tmp20, align 4
|
|
%tmp22 = getelementptr inbounds float, float addrspace(3)* %arg, i32 12
|
|
%tmp23 = load float, float addrspace(3)* %tmp22, align 4
|
|
%tmp24 = getelementptr inbounds float, float addrspace(3)* %arg, i32 13
|
|
%tmp25 = load float, float addrspace(3)* %tmp24, align 4
|
|
%tmp26 = getelementptr inbounds float, float addrspace(3)* %arg, i32 14
|
|
%tmp27 = load float, float addrspace(3)* %tmp26, align 4
|
|
%tmp28 = getelementptr inbounds float, float addrspace(3)* %arg, i32 15
|
|
%tmp29 = load float, float addrspace(3)* %tmp28, align 4
|
|
%tmp30 = getelementptr inbounds float, float addrspace(3)* %arg, i32 16
|
|
%tmp31 = load float, float addrspace(3)* %tmp30, align 4
|
|
%tmp32 = getelementptr inbounds float, float addrspace(3)* %arg, i32 17
|
|
%tmp33 = load float, float addrspace(3)* %tmp32, align 4
|
|
%tmp34 = getelementptr inbounds float, float addrspace(3)* %arg, i32 18
|
|
%tmp35 = load float, float addrspace(3)* %tmp34, align 4
|
|
%tmp36 = getelementptr inbounds float, float addrspace(3)* %arg, i32 19
|
|
%tmp37 = load float, float addrspace(3)* %tmp36, align 4
|
|
%tmp38 = getelementptr inbounds float, float addrspace(3)* %arg, i32 20
|
|
%tmp39 = load float, float addrspace(3)* %tmp38, align 4
|
|
%tmp40 = getelementptr inbounds float, float addrspace(3)* %arg, i32 21
|
|
%tmp41 = load float, float addrspace(3)* %tmp40, align 4
|
|
%tmp42 = getelementptr inbounds float, float addrspace(3)* %arg, i32 22
|
|
%tmp43 = load float, float addrspace(3)* %tmp42, align 4
|
|
%tmp44 = getelementptr inbounds float, float addrspace(3)* %arg, i32 23
|
|
%tmp45 = load float, float addrspace(3)* %tmp44, align 4
|
|
%tmp46 = getelementptr inbounds float, float addrspace(3)* %arg, i32 24
|
|
%tmp47 = load float, float addrspace(3)* %tmp46, align 4
|
|
%tmp48 = getelementptr inbounds float, float addrspace(3)* %arg, i32 25
|
|
%tmp49 = load float, float addrspace(3)* %tmp48, align 4
|
|
%tmp50 = getelementptr inbounds float, float addrspace(3)* %arg, i32 26
|
|
%tmp51 = load float, float addrspace(3)* %tmp50, align 4
|
|
%tmp52 = getelementptr inbounds float, float addrspace(3)* %arg, i32 27
|
|
%tmp53 = load float, float addrspace(3)* %tmp52, align 4
|
|
%tmp54 = getelementptr inbounds float, float addrspace(3)* %arg, i32 28
|
|
%tmp55 = load float, float addrspace(3)* %tmp54, align 4
|
|
%tmp56 = getelementptr inbounds float, float addrspace(3)* %arg, i32 29
|
|
%tmp57 = load float, float addrspace(3)* %tmp56, align 4
|
|
%tmp58 = getelementptr inbounds float, float addrspace(3)* %arg, i32 30
|
|
%tmp59 = load float, float addrspace(3)* %tmp58, align 4
|
|
%tmp60 = tail call float @llvm.fmuladd.f32(float %tmp1, float %tmp3, float %tmp5)
|
|
%tmp61 = tail call float @llvm.fmuladd.f32(float %tmp7, float %tmp9, float %tmp11)
|
|
%tmp62 = tail call float @llvm.fmuladd.f32(float %tmp13, float %tmp15, float %tmp17)
|
|
%tmp63 = tail call float @llvm.fmuladd.f32(float %tmp19, float %tmp21, float %tmp23)
|
|
%tmp64 = tail call float @llvm.fmuladd.f32(float %tmp25, float %tmp27, float %tmp29)
|
|
%tmp65 = tail call float @llvm.fmuladd.f32(float %tmp31, float %tmp33, float %tmp35)
|
|
%tmp66 = tail call float @llvm.fmuladd.f32(float %tmp37, float %tmp39, float %tmp41)
|
|
%tmp67 = tail call float @llvm.fmuladd.f32(float %tmp43, float %tmp45, float %tmp47)
|
|
%tmp68 = tail call float @llvm.fmuladd.f32(float %tmp49, float %tmp51, float %tmp53)
|
|
%tmp69 = tail call float @llvm.fmuladd.f32(float %tmp55, float %tmp57, float %tmp59)
|
|
%tmp70 = getelementptr inbounds float, float addrspace(3)* %arg1, i64 1
|
|
store float %tmp60, float addrspace(3)* %tmp70, align 4
|
|
%tmp71 = getelementptr inbounds float, float addrspace(3)* %arg1, i64 2
|
|
store float %tmp61, float addrspace(3)* %tmp71, align 4
|
|
%tmp72 = getelementptr inbounds float, float addrspace(3)* %arg1, i64 3
|
|
store float %tmp62, float addrspace(3)* %tmp72, align 4
|
|
%tmp73 = getelementptr inbounds float, float addrspace(3)* %arg1, i64 4
|
|
store float %tmp63, float addrspace(3)* %tmp73, align 4
|
|
%tmp74 = getelementptr inbounds float, float addrspace(3)* %arg1, i64 5
|
|
store float %tmp64, float addrspace(3)* %tmp74, align 4
|
|
%tmp75 = getelementptr inbounds float, float addrspace(3)* %arg1, i64 6
|
|
store float %tmp65, float addrspace(3)* %tmp75, align 4
|
|
%tmp76 = getelementptr inbounds float, float addrspace(3)* %arg1, i64 7
|
|
store float %tmp66, float addrspace(3)* %tmp76, align 4
|
|
%tmp77 = getelementptr inbounds float, float addrspace(3)* %arg1, i64 8
|
|
store float %tmp67, float addrspace(3)* %tmp77, align 4
|
|
%tmp78 = getelementptr inbounds float, float addrspace(3)* %arg1, i64 9
|
|
store float %tmp68, float addrspace(3)* %tmp78, align 4
|
|
%tmp79 = getelementptr inbounds float, float addrspace(3)* %arg1, i64 10
|
|
store float %tmp69, float addrspace(3)* %tmp79, align 4
|
|
ret void
|
|
}
|
|
|
|
; Function Attrs: nounwind readnone
|
|
declare float @llvm.fmuladd.f32(float, float, float) #0
|
|
|
|
attributes #0 = { nounwind readnone }
|
|
attributes #1 = { "amdgpu-waves-per-eu"="1,1" "amdgpu-flat-work-group-size"="1,256" }
|