93 lines · plain
1; RUN: llvm-split -o %t %s -j 3 -mtriple amdgcn-amd-amdhsa -amdgpu-module-splitting-max-depth=0 -amdgpu-module-splitting-large-threshold=1.2 -amdgpu-module-splitting-merge-threshold=0.52; RUN: llvm-dis -o - %t0 | FileCheck --check-prefix=CHECK0 --implicit-check-not=define %s3; RUN: llvm-dis -o - %t1 | FileCheck --check-prefix=CHECK1 --implicit-check-not=define %s4 5; RUN: llvm-split -o %t.nolarge %s -j 3 -mtriple amdgcn-amd-amdhsa -amdgpu-module-splitting-large-threshold=0 -amdgpu-module-splitting-max-depth=06; RUN: llvm-dis -o - %t.nolarge0 | FileCheck --check-prefix=NOLARGEKERNELS-CHECK0 --implicit-check-not=define %s7; RUN: llvm-dis -o - %t.nolarge1 | FileCheck --check-prefix=NOLARGEKERNELS-CHECK1 --implicit-check-not=define %s8; RUN: llvm-dis -o - %t.nolarge2 | FileCheck --check-prefix=NOLARGEKERNELS-CHECK2 --implicit-check-not=define %s9 10; 2 kernels (A/B) are large and share all their dependencies.11; They should go in the same partition, the remaining kernel should12; go somewhere else, and one partition should be empty.13;14; Also check w/o large kernels processing to verify they are indeed handled15; differently.16 17; Only 2 partitions for the first command.18 19; CHECK0: define internal void @HelperC()20; CHECK0: define amdgpu_kernel void @C21 22; CHECK1: define internal void @large2()23; CHECK1: define internal void @large1()24; CHECK1: define internal void @large0()25; CHECK1: define internal void @HelperA()26; CHECK1: define internal void @HelperB()27; CHECK1: define amdgpu_kernel void @A28; CHECK1: define amdgpu_kernel void @B29 30; NOLARGEKERNELS-CHECK0: define internal void @HelperC()31; NOLARGEKERNELS-CHECK0: define amdgpu_kernel void @C32 33; NOLARGEKERNELS-CHECK1: define internal void @large2()34; NOLARGEKERNELS-CHECK1: define internal void @large1()35; NOLARGEKERNELS-CHECK1: define internal void @large0()36; NOLARGEKERNELS-CHECK1: define internal void @HelperB()37; NOLARGEKERNELS-CHECK1: define amdgpu_kernel void @B38 39; NOLARGEKERNELS-CHECK2: define internal void @large2()40; NOLARGEKERNELS-CHECK2: define internal void @large1()41; NOLARGEKERNELS-CHECK2: define internal void @large0()42; NOLARGEKERNELS-CHECK2: define internal void @HelperA()43; NOLARGEKERNELS-CHECK2: define amdgpu_kernel void @A44 45 46define internal void @large2() {47 store volatile i32 42, ptr null48 call void @large2()49 ret void50}51 52define internal void @large1() {53 call void @large1()54 call void @large2()55 ret void56}57 58define internal void @large0() {59 call void @large0()60 call void @large1()61 call void @large2()62 ret void63}64 65define internal void @HelperA() {66 call void @large0()67 ret void68}69 70define internal void @HelperB() {71 call void @large0()72 ret void73}74 75define amdgpu_kernel void @A() {76 call void @HelperA()77 ret void78}79 80define amdgpu_kernel void @B() {81 call void @HelperB()82 ret void83}84 85define internal void @HelperC() {86 ret void87}88 89define amdgpu_kernel void @C() {90 call void @HelperC()91 ret void92}93