130080fab1
Don't count register uses when determining the maximum number of registers used by a function. Count only the defs. This is really an underestimate of the true register usage, but in practice that's not a problem because if a function uses a register, then it has either defined it earlier, or some other function that executed before has defined it. In particular, the register counts are used: 1. When launching an entry function - in which case we're safe because the register counts of the entry function will include the register counts of all callees. 2. At function boundaries in dynamic VGPR mode. In this case it's safe because whenever we set the new VGPR allocation we take into account the outgoing_vgpr_count set by the middle-end. The main advantage of doing this is that the artificial VGPR arguments used only for preserving the inactive lanes when using the llvm.amdgcn.init.whole.wave intrinsic are no longer counted. This enables us to allocate only the registers we need in dynamic VGPR mode. --------- Co-authored-by: Thomas Symalla <5754458+tsymalla@users.noreply.github.com>
228 lines
7.9 KiB
LLVM
228 lines
7.9 KiB
LLVM
; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=fiji -verify-machineinstrs < %s | FileCheck %s
|
|
|
|
; Exactly 1 wave per execution unit.
|
|
; CHECK-LABEL: {{^}}empty_exactly_1:
|
|
; CHECK: SGPRBlocks: 12
|
|
; CHECK: VGPRBlocks: 32
|
|
; CHECK: NumSGPRsForWavesPerEU: 102
|
|
; CHECK: NumVGPRsForWavesPerEU: 129
|
|
define amdgpu_kernel void @empty_exactly_1() #0 {
|
|
entry:
|
|
ret void
|
|
}
|
|
attributes #0 = {"amdgpu-waves-per-eu"="1,1" "amdgpu-flat-work-group-size"="1,64" }
|
|
|
|
; Exactly 5 waves per execution unit.
|
|
; CHECK-LABEL: {{^}}empty_exactly_5:
|
|
; CHECK: SGPRBlocks: 12
|
|
; CHECK: VGPRBlocks: 10
|
|
; CHECK: NumSGPRsForWavesPerEU: 102
|
|
; CHECK: NumVGPRsForWavesPerEU: 41
|
|
define amdgpu_kernel void @empty_exactly_5() #1 {
|
|
entry:
|
|
ret void
|
|
}
|
|
attributes #1 = {"amdgpu-waves-per-eu"="5,5"}
|
|
|
|
; Exactly 10 waves per execution unit.
|
|
; CHECK-LABEL: {{^}}empty_exactly_10:
|
|
; CHECK: SGPRBlocks: 2
|
|
; CHECK: VGPRBlocks: 0
|
|
; CHECK: NumSGPRsForWavesPerEU: 1
|
|
; CHECK: NumVGPRsForWavesPerEU: 3
|
|
define amdgpu_kernel void @empty_exactly_10() #2 {
|
|
entry:
|
|
ret void
|
|
}
|
|
attributes #2 = {"amdgpu-waves-per-eu"="10,10"}
|
|
|
|
; At least 1 wave per execution unit.
|
|
; CHECK-LABEL: {{^}}empty_at_least_1:
|
|
; CHECK: SGPRBlocks: 2
|
|
; CHECK: VGPRBlocks: 0
|
|
; CHECK: NumSGPRsForWavesPerEU: 1
|
|
; CHECK: NumVGPRsForWavesPerEU: 3
|
|
define amdgpu_kernel void @empty_at_least_1() #3 {
|
|
entry:
|
|
ret void
|
|
}
|
|
attributes #3 = {"amdgpu-waves-per-eu"="1"}
|
|
|
|
; At least 5 waves per execution unit.
|
|
; CHECK-LABEL: {{^}}empty_at_least_5:
|
|
; CHECK: SGPRBlocks: 2
|
|
; CHECK: VGPRBlocks: 0
|
|
; CHECK: NumSGPRsForWavesPerEU: 1
|
|
; CHECK: NumVGPRsForWavesPerEU: 3
|
|
define amdgpu_kernel void @empty_at_least_5() #4 {
|
|
entry:
|
|
ret void
|
|
}
|
|
attributes #4 = {"amdgpu-waves-per-eu"="5"}
|
|
|
|
; At least 10 waves per execution unit.
|
|
; CHECK-LABEL: {{^}}empty_at_least_10:
|
|
; CHECK: SGPRBlocks: 2
|
|
; CHECK: VGPRBlocks: 0
|
|
; CHECK: NumSGPRsForWavesPerEU: 1
|
|
; CHECK: NumVGPRsForWavesPerEU: 3
|
|
define amdgpu_kernel void @empty_at_least_10() #5 {
|
|
entry:
|
|
ret void
|
|
}
|
|
attributes #5 = {"amdgpu-waves-per-eu"="10"}
|
|
|
|
; At most 1 wave per execution unit (same as @empty_exactly_1).
|
|
|
|
; At most 5 waves per execution unit.
|
|
; CHECK-LABEL: {{^}}empty_at_most_5:
|
|
; CHECK: SGPRBlocks: 12
|
|
; CHECK: VGPRBlocks: 10
|
|
; CHECK: NumSGPRsForWavesPerEU: 102
|
|
; CHECK: NumVGPRsForWavesPerEU: 41
|
|
define amdgpu_kernel void @empty_at_most_5() #6 {
|
|
entry:
|
|
ret void
|
|
}
|
|
attributes #6 = {"amdgpu-waves-per-eu"="1,5" "amdgpu-flat-work-group-size"="1,64"}
|
|
|
|
; At most 10 waves per execution unit.
|
|
; CHECK-LABEL: {{^}}empty_at_most_10:
|
|
; CHECK: SGPRBlocks: 2
|
|
; CHECK: VGPRBlocks: 0
|
|
; CHECK: NumSGPRsForWavesPerEU: 1
|
|
; CHECK: NumVGPRsForWavesPerEU: 3
|
|
define amdgpu_kernel void @empty_at_most_10() #7 {
|
|
entry:
|
|
ret void
|
|
}
|
|
attributes #7 = {"amdgpu-waves-per-eu"="1,10"}
|
|
|
|
; Between 1 and 5 waves per execution unit (same as @empty_at_most_5).
|
|
|
|
; Between 5 and 10 waves per execution unit.
|
|
; CHECK-LABEL: {{^}}empty_between_5_and_10:
|
|
; CHECK: SGPRBlocks: 2
|
|
; CHECK: VGPRBlocks: 0
|
|
; CHECK: NumSGPRsForWavesPerEU: 1
|
|
; CHECK: NumVGPRsForWavesPerEU: 3
|
|
define amdgpu_kernel void @empty_between_5_and_10() #8 {
|
|
entry:
|
|
ret void
|
|
}
|
|
attributes #8 = {"amdgpu-waves-per-eu"="5,10"}
|
|
|
|
@var = addrspace(1) global float 0.0
|
|
|
|
; Exactly 10 waves per execution unit.
|
|
; CHECK-LABEL: {{^}}exactly_10:
|
|
; CHECK: SGPRBlocks: 3
|
|
; CHECK: VGPRBlocks: 5
|
|
; CHECK: NumSGPRsForWavesPerEU: 30
|
|
; CHECK: NumVGPRsForWavesPerEU: 24
|
|
define amdgpu_kernel void @exactly_10() #9 {
|
|
%val0 = load volatile float, ptr addrspace(1) @var
|
|
%val1 = load volatile float, ptr addrspace(1) @var
|
|
%val2 = load volatile float, ptr addrspace(1) @var
|
|
%val3 = load volatile float, ptr addrspace(1) @var
|
|
%val4 = load volatile float, ptr addrspace(1) @var
|
|
%val5 = load volatile float, ptr addrspace(1) @var
|
|
%val6 = load volatile float, ptr addrspace(1) @var
|
|
%val7 = load volatile float, ptr addrspace(1) @var
|
|
%val8 = load volatile float, ptr addrspace(1) @var
|
|
%val9 = load volatile float, ptr addrspace(1) @var
|
|
%val10 = load volatile float, ptr addrspace(1) @var
|
|
%val11 = load volatile float, ptr addrspace(1) @var
|
|
%val12 = load volatile float, ptr addrspace(1) @var
|
|
%val13 = load volatile float, ptr addrspace(1) @var
|
|
%val14 = load volatile float, ptr addrspace(1) @var
|
|
%val15 = load volatile float, ptr addrspace(1) @var
|
|
%val16 = load volatile float, ptr addrspace(1) @var
|
|
%val17 = load volatile float, ptr addrspace(1) @var
|
|
%val18 = load volatile float, ptr addrspace(1) @var
|
|
%val19 = load volatile float, ptr addrspace(1) @var
|
|
%val20 = load volatile float, ptr addrspace(1) @var
|
|
%val21 = load volatile float, ptr addrspace(1) @var
|
|
%val22 = load volatile float, ptr addrspace(1) @var
|
|
%val23 = load volatile float, ptr addrspace(1) @var
|
|
%val24 = load volatile float, ptr addrspace(1) @var
|
|
%val25 = load volatile float, ptr addrspace(1) @var
|
|
%val26 = load volatile float, ptr addrspace(1) @var
|
|
%val27 = load volatile float, ptr addrspace(1) @var
|
|
%val28 = load volatile float, ptr addrspace(1) @var
|
|
%val29 = load volatile float, ptr addrspace(1) @var
|
|
%val30 = load volatile float, ptr addrspace(1) @var
|
|
|
|
store volatile float %val0, ptr addrspace(1) @var
|
|
store volatile float %val1, ptr addrspace(1) @var
|
|
store volatile float %val2, ptr addrspace(1) @var
|
|
store volatile float %val3, ptr addrspace(1) @var
|
|
store volatile float %val4, ptr addrspace(1) @var
|
|
store volatile float %val5, ptr addrspace(1) @var
|
|
store volatile float %val6, ptr addrspace(1) @var
|
|
store volatile float %val7, ptr addrspace(1) @var
|
|
store volatile float %val8, ptr addrspace(1) @var
|
|
store volatile float %val9, ptr addrspace(1) @var
|
|
store volatile float %val10, ptr addrspace(1) @var
|
|
store volatile float %val11, ptr addrspace(1) @var
|
|
store volatile float %val12, ptr addrspace(1) @var
|
|
store volatile float %val13, ptr addrspace(1) @var
|
|
store volatile float %val14, ptr addrspace(1) @var
|
|
store volatile float %val15, ptr addrspace(1) @var
|
|
store volatile float %val16, ptr addrspace(1) @var
|
|
store volatile float %val17, ptr addrspace(1) @var
|
|
store volatile float %val18, ptr addrspace(1) @var
|
|
store volatile float %val19, ptr addrspace(1) @var
|
|
store volatile float %val20, ptr addrspace(1) @var
|
|
store volatile float %val21, ptr addrspace(1) @var
|
|
store volatile float %val22, ptr addrspace(1) @var
|
|
store volatile float %val23, ptr addrspace(1) @var
|
|
store volatile float %val24, ptr addrspace(1) @var
|
|
store volatile float %val25, ptr addrspace(1) @var
|
|
store volatile float %val26, ptr addrspace(1) @var
|
|
store volatile float %val27, ptr addrspace(1) @var
|
|
store volatile float %val28, ptr addrspace(1) @var
|
|
store volatile float %val29, ptr addrspace(1) @var
|
|
store volatile float %val30, ptr addrspace(1) @var
|
|
|
|
ret void
|
|
}
|
|
attributes #9 = {"amdgpu-waves-per-eu"="10,10"}
|
|
|
|
; Exactly 256 workitems and exactly 2 waves.
|
|
; CHECK-LABEL: {{^}}empty_workitems_exactly_256_waves_exactly_2:
|
|
; CHECK: SGPRBlocks: 12
|
|
; CHECK: VGPRBlocks: 21
|
|
; CHECK: NumSGPRsForWavesPerEU: 102
|
|
; CHECK: NumVGPRsForWavesPerEU: 85
|
|
define amdgpu_kernel void @empty_workitems_exactly_256_waves_exactly_2() #10 {
|
|
entry:
|
|
ret void
|
|
}
|
|
attributes #10 = {"amdgpu-flat-work-group-size"="256,256" "amdgpu-waves-per-eu"="2,2"}
|
|
|
|
; Minimum 2 waves, maximum limited by LDS usage.
|
|
; CHECK-LABEL: {{^}}empty_at_least_2_lds_limited:
|
|
; CHECK: SGPRBlocks: 12
|
|
; CHECK: VGPRBlocks: 12
|
|
; CHECK: NumSGPRsForWavesPerEU: 102
|
|
; CHECK: NumVGPRsForWavesPerEU: 49
|
|
define amdgpu_kernel void @empty_at_least_2_lds_limited() #11 {
|
|
entry:
|
|
ret void
|
|
}
|
|
attributes #11 = {"amdgpu-flat-work-group-size"="1,256" "amdgpu-waves-per-eu"="2" "amdgpu-lds-size"="16384"}
|
|
|
|
; Minimum 2 waves, maximum limited by LDS usage. Requested maximum within spec
|
|
; but above achievable occupancy has no effect.
|
|
; CHECK-LABEL: {{^}}empty_at_least_2_lds_limited_max_above_achievable:
|
|
; CHECK: SGPRBlocks: 12
|
|
; CHECK: VGPRBlocks: 12
|
|
; CHECK: NumSGPRsForWavesPerEU: 102
|
|
; CHECK: NumVGPRsForWavesPerEU: 49
|
|
define amdgpu_kernel void @empty_at_least_2_lds_limited_max_above_achievable() #12 {
|
|
entry:
|
|
ret void
|
|
}
|
|
attributes #12 = {"amdgpu-flat-work-group-size"="1,256" "amdgpu-waves-per-eu"="2,10" "amdgpu-lds-size"="16384"}
|