llvm/test/CodeGen/AMDGPU/propagate-flat-work-group-size.ll

   1 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --function-signature --check-globals
   2 ; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -amdgpu-attributor %s | FileCheck %s
   3
   4 ; Check propagation of amdgpu-flat-work-group-size attribute.
   5
   6 ; Called from a single kernel with 1,256
   7 define internal void @default_to_1_256() {
   8 ; CHECK-LABEL: define {{[^@]+}}@default_to_1_256
   9 ; CHECK-SAME: () #[[ATTR0:[0-9]+]] {
  10 ; CHECK-NEXT:    ret void
  11 ;
  12   ret void
  13 }
  14
  15 define amdgpu_kernel void @kernel_1_256() #0 {
  16 ; CHECK-LABEL: define {{[^@]+}}@kernel_1_256
  17 ; CHECK-SAME: () #[[ATTR0]] {
  18 ; CHECK-NEXT:    call void @default_to_1_256()
  19 ; CHECK-NEXT:    ret void
  20 ;
  21   call void @default_to_1_256()
  22   ret void
  23 }
  24
  25 ; Called from a single kernel with 64,128
  26 define internal void @default_to_64_128() {
  27 ; CHECK-LABEL: define {{[^@]+}}@default_to_64_128
  28 ; CHECK-SAME: () #[[ATTR1:[0-9]+]] {
  29 ; CHECK-NEXT:    ret void
  30 ;
  31   ret void
  32 }
  33
  34 define amdgpu_kernel void @kernel_64_128() #1 {
  35 ; CHECK-LABEL: define {{[^@]+}}@kernel_64_128
  36 ; CHECK-SAME: () #[[ATTR1]] {
  37 ; CHECK-NEXT:    call void @default_to_64_128()
  38 ; CHECK-NEXT:    call void @flat_group_64_64()
  39 ; CHECK-NEXT:    call void @default_to_64_256()
  40 ; CHECK-NEXT:    call void @flat_group_128_256()
  41 ; CHECK-NEXT:    ret void
  42 ;
  43   call void @default_to_64_128()
  44   call void @flat_group_64_64()
  45   call void @default_to_64_256()
  46   call void @flat_group_128_256()
  47   ret void
  48 }
  49
  50 ; Called from kernels with 128,512 and 512,512
  51 define internal void @default_to_128_512() {
  52 ; CHECK-LABEL: define {{[^@]+}}@default_to_128_512
  53 ; CHECK-SAME: () #[[ATTR2:[0-9]+]] {
  54 ; CHECK-NEXT:    ret void
  55 ;
  56   ret void
  57 }
  58
  59 ; This already has a strict bounds, but called from kernels with wider
  60 ; bounds, and should not be changed.
  61 define internal void @flat_group_64_64() #2 {
  62 ; CHECK-LABEL: define {{[^@]+}}@flat_group_64_64
  63 ; CHECK-SAME: () #[[ATTR3:[0-9]+]] {
  64 ; CHECK-NEXT:    ret void
  65 ;
  66   ret void
  67 }
  68
  69 ; 128,256 -> 128,128
  70 define internal void @flat_group_128_256() #3 {
  71 ; CHECK-LABEL: define {{[^@]+}}@flat_group_128_256
  72 ; CHECK-SAME: () #[[ATTR4:[0-9]+]] {
  73 ; CHECK-NEXT:    ret void
  74 ;
  75   ret void
  76 }
  77
  78 define internal void @flat_group_512_1024() #4 {
  79 ; CHECK-LABEL: define {{[^@]+}}@flat_group_512_1024
  80 ; CHECK-SAME: () #[[ATTR5:[0-9]+]] {
  81 ; CHECK-NEXT:    ret void
  82 ;
  83   ret void
  84 }
  85
  86 define amdgpu_kernel void @kernel_128_512() #5 {
  87 ; CHECK-LABEL: define {{[^@]+}}@kernel_128_512
  88 ; CHECK-SAME: () #[[ATTR2]] {
  89 ; CHECK-NEXT:    call void @default_to_128_512()
  90 ; CHECK-NEXT:    call void @flat_group_64_64()
  91 ; CHECK-NEXT:    ret void
  92 ;
  93   call void @default_to_128_512()
  94   call void @flat_group_64_64()
  95   ret void
  96 }
  97
  98 define amdgpu_kernel void @kernel_512_512() #6 {
  99 ; CHECK-LABEL: define {{[^@]+}}@kernel_512_512
 100 ; CHECK-SAME: () #[[ATTR5]] {
 101 ; CHECK-NEXT:    call void @default_to_128_512()
 102 ; CHECK-NEXT:    call void @flat_group_512_1024()
 103 ; CHECK-NEXT:    ret void
 104 ;
 105   call void @default_to_128_512()
 106   call void @flat_group_512_1024()
 107   ret void
 108 }
 109
 110 ; Called from kernels with 128,256 and 64,128 => 64,256
 111 define internal void @default_to_64_256() {
 112 ; CHECK-LABEL: define {{[^@]+}}@default_to_64_256
 113 ; CHECK-SAME: () #[[ATTR6:[0-9]+]] {
 114 ; CHECK-NEXT:    ret void
 115 ;
 116   ret void
 117 }
 118
 119 ; The kernel's lower bound is higher than the callee's lower bound, so
 120 ; this should probably be illegal.
 121 define amdgpu_kernel void @kernel_128_256() #3 {
 122 ; CHECK-LABEL: define {{[^@]+}}@kernel_128_256
 123 ; CHECK-SAME: () #[[ATTR7:[0-9]+]] {
 124 ; CHECK-NEXT:    call void @default_to_64_256()
 125 ; CHECK-NEXT:    ret void
 126 ;
 127   call void @default_to_64_256()
 128   ret void
 129 }
 130
 131 ; 64,128 -> 64,128
 132 define internal void @merge_cycle_0() #1 {
 133 ; CHECK-LABEL: define {{[^@]+}}@merge_cycle_0
 134 ; CHECK-SAME: () #[[ATTR1]] {
 135 ; CHECK-NEXT:    call void @merge_cycle_1()
 136 ; CHECK-NEXT:    ret void
 137 ;
 138   call void @merge_cycle_1()
 139   ret void
 140 }
 141
 142 ; 128,256 -> 128,128
 143 define internal void @merge_cycle_1() #3 {
 144 ; CHECK-LABEL: define {{[^@]+}}@merge_cycle_1
 145 ; CHECK-SAME: () #[[ATTR4]] {
 146 ; CHECK-NEXT:    call void @merge_cycle_0()
 147 ; CHECK-NEXT:    ret void
 148 ;
 149   call void @merge_cycle_0()
 150   ret void
 151 }
 152
 153 define amdgpu_kernel void @kernel_64_256() #7 {
 154 ; CHECK-LABEL: define {{[^@]+}}@kernel_64_256
 155 ; CHECK-SAME: () #[[ATTR6]] {
 156 ; CHECK-NEXT:    call void @merge_cycle_0()
 157 ; CHECK-NEXT:    call void @default_captured_address()
 158 ; CHECK-NEXT:    call void @externally_visible_default()
 159 ; CHECK-NEXT:    [[F32:%.*]] = call float @bitcasted_function()
 160 ; CHECK-NEXT:    ret void
 161 ;
 162   call void @merge_cycle_0()
 163   call void @default_captured_address()
 164   call void @externally_visible_default()
 165   %f32 = call float @bitcasted_function()
 166   ret void
 167 }
 168
 169 define internal void @default_captured_address() {
 170 ; CHECK-LABEL: define {{[^@]+}}@default_captured_address
 171 ; CHECK-SAME: () #[[ATTR8:[0-9]+]] {
 172 ; CHECK-NEXT:    store volatile ptr @default_captured_address, ptr undef, align 8
 173 ; CHECK-NEXT:    ret void
 174 ;
 175   store volatile ptr @default_captured_address, ptr undef, align 8
 176   ret void
 177 }
 178
 179 define void @externally_visible_default() {
 180 ; CHECK-LABEL: define {{[^@]+}}@externally_visible_default
 181 ; CHECK-SAME: () #[[ATTR8]] {
 182 ; CHECK-NEXT:    ret void
 183 ;
 184   ret void
 185 }
 186
 187 ; 1,1024 -> 64,256
 188 define internal i32 @bitcasted_function() {
 189 ; CHECK-LABEL: define {{[^@]+}}@bitcasted_function
 190 ; CHECK-SAME: () #[[ATTR6]] {
 191 ; CHECK-NEXT:    ret i32 0
 192 ;
 193   ret i32 0
 194 }
 195
 196 attributes #0 = { "amdgpu-flat-work-group-size"="1,256" }
 197 attributes #1 = { "amdgpu-flat-work-group-size"="64,128" }
 198 attributes #2 = { "amdgpu-flat-work-group-size"="64,64" }
 199 attributes #3 = { "amdgpu-flat-work-group-size"="128,256" }
 200 attributes #4 = { "amdgpu-flat-work-group-size"="512,1024" }
 201 attributes #5 = { "amdgpu-flat-work-group-size"="128,512" }
 202 attributes #6 = { "amdgpu-flat-work-group-size"="512,512" }
 203 attributes #7 = { "amdgpu-flat-work-group-size"="64,256" }
 204 ;.
 205 ; CHECK: attributes #[[ATTR0]] = { "amdgpu-flat-work-group-size"="1,256" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" }
 206 ; CHECK: attributes #[[ATTR1]] = { "amdgpu-flat-work-group-size"="64,128" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" }
 207 ; CHECK: attributes #[[ATTR2]] = { "amdgpu-flat-work-group-size"="128,512" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="2,10" "uniform-work-group-size"="false" }
 208 ; CHECK: attributes #[[ATTR3]] = { "amdgpu-flat-work-group-size"="64,64" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" }
 209 ; CHECK: attributes #[[ATTR4]] = { "amdgpu-flat-work-group-size"="128,128" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" }
 210 ; CHECK: attributes #[[ATTR5]] = { "amdgpu-flat-work-group-size"="512,512" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="2,10" "uniform-work-group-size"="false" }
 211 ; CHECK: attributes #[[ATTR6]] = { "amdgpu-flat-work-group-size"="64,256" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" }
 212 ; CHECK: attributes #[[ATTR7]] = { "amdgpu-flat-work-group-size"="128,256" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" }
 213 ; CHECK: attributes #[[ATTR8]] = { "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="4,10" "uniform-work-group-size"="false" }
 214 ;.