18f8106f31
This patch implements an LLVM IR pass, named kernel-info, that reports various statistics for codes compiled for GPUs. The ultimate goal of these statistics to help identify bad code patterns and ways to mitigate them. The pass operates at the LLVM IR level so that it can, in theory, support any LLVM-based compiler for programming languages supporting GPUs. It has been tested so far with LLVM IR generated by Clang for OpenMP offload codes targeting NVIDIA GPUs and AMD GPUs. By default, the pass runs at the end of LTO, and options like ``-Rpass=kernel-info`` enable its remarks. Example `opt` and `clang` command lines appear in `llvm/docs/KernelInfo.rst`. Remarks include summary statistics (e.g., total size of static allocas) and individual occurrences (e.g., source location of each alloca). Examples of its output appear in tests in `llvm/test/Analysis/KernelInfo`.
84 lines
4.6 KiB
LLVM
84 lines
4.6 KiB
LLVM
; Check info on launch bounds for AMD GPU.
|
|
|
|
; REQUIRES: amdgpu-registered-target
|
|
|
|
; RUN: opt -pass-remarks=kernel-info -passes=kernel-info \
|
|
; RUN: -disable-output %s 2>&1 | \
|
|
; RUN: FileCheck -match-full-lines %s
|
|
|
|
target datalayout = "e-p:64:64-p1:64:64-p2:32:32-p3:32:32-p4:64:64-p5:32:32-p6:32:32-p7:160:256:256:32-p8:128:128-p9:192:256:256:32-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64-S32-A5-G1-ni:7:8:9"
|
|
target triple = "amdgcn-amd-amdhsa"
|
|
|
|
; CHECK: remark: test.c:10:0: in artificial function 'all', omp_target_num_teams = 100
|
|
; CHECK: remark: test.c:10:0: in artificial function 'all', omp_target_thread_limit = 101
|
|
; CHECK: remark: test.c:10:0: in artificial function 'all', amdgpu-max-num-workgroups[0] = 200
|
|
; CHECK: remark: test.c:10:0: in artificial function 'all', amdgpu-max-num-workgroups[1] = 201
|
|
; CHECK: remark: test.c:10:0: in artificial function 'all', amdgpu-max-num-workgroups[2] = 202
|
|
; CHECK: remark: test.c:10:0: in artificial function 'all', amdgpu-flat-work-group-size[0] = 210
|
|
; CHECK: remark: test.c:10:0: in artificial function 'all', amdgpu-flat-work-group-size[1] = 211
|
|
; CHECK: remark: test.c:10:0: in artificial function 'all', amdgpu-waves-per-eu[0] = 2
|
|
; CHECK: remark: test.c:10:0: in artificial function 'all', amdgpu-waves-per-eu[1] = 9
|
|
define void @all() #0 !dbg !5 {
|
|
entry:
|
|
ret void
|
|
}
|
|
|
|
; CHECK-NOT: remark: test.c:11:0: in function 'none', omp_target_num_teams = {{.*}}
|
|
; CHECK-NOT: remark: test.c:11:0: in function 'none', omp_target_thread_limit = {{.*}}
|
|
; CHECK: remark: test.c:11:0: in function 'none', amdgpu-max-num-workgroups[0] = 4294967295
|
|
; CHECK: remark: test.c:11:0: in function 'none', amdgpu-max-num-workgroups[1] = 4294967295
|
|
; CHECK: remark: test.c:11:0: in function 'none', amdgpu-max-num-workgroups[2] = 4294967295
|
|
; CHECK: remark: test.c:11:0: in function 'none', amdgpu-flat-work-group-size[0] = 1
|
|
; CHECK: remark: test.c:11:0: in function 'none', amdgpu-flat-work-group-size[1] = 1024
|
|
; CHECK: remark: test.c:11:0: in function 'none', amdgpu-waves-per-eu[0] = 4
|
|
; CHECK: remark: test.c:11:0: in function 'none', amdgpu-waves-per-eu[1] = 10
|
|
define void @none() !dbg !6 {
|
|
entry:
|
|
ret void
|
|
}
|
|
|
|
; CHECK: remark: test.c:12:0: in function 'bogus', omp_target_num_teams = 987654321
|
|
; CHECK: remark: test.c:12:0: in function 'bogus', omp_target_thread_limit = 987654321
|
|
; CHECK: remark: test.c:12:0: in function 'bogus', amdgpu-max-num-workgroups[0] = 987654321
|
|
; CHECK: remark: test.c:12:0: in function 'bogus', amdgpu-max-num-workgroups[1] = 987654321
|
|
; CHECK: remark: test.c:12:0: in function 'bogus', amdgpu-max-num-workgroups[2] = 987654321
|
|
; CHECK: remark: test.c:12:0: in function 'bogus', amdgpu-flat-work-group-size[0] = 1
|
|
; CHECK: remark: test.c:12:0: in function 'bogus', amdgpu-flat-work-group-size[1] = 1024
|
|
; CHECK: remark: test.c:12:0: in function 'bogus', amdgpu-waves-per-eu[0] = 4
|
|
; CHECK: remark: test.c:12:0: in function 'bogus', amdgpu-waves-per-eu[1] = 10
|
|
define void @bogus() #1 !dbg !7 {
|
|
entry:
|
|
ret void
|
|
}
|
|
|
|
attributes #0 = {
|
|
"omp_target_num_teams"="100"
|
|
"omp_target_thread_limit"="101"
|
|
"amdgpu-max-num-workgroups"="200,201,202"
|
|
"amdgpu-flat-work-group-size"="210,211"
|
|
"amdgpu-waves-per-eu"="2,9"
|
|
}
|
|
|
|
; We choose values that are small enough to parse successfully but that are
|
|
; impossibly large. For values that are validated, we check that they are
|
|
; overridden with realistic values.
|
|
attributes #1 = {
|
|
"omp_target_num_teams"="987654321"
|
|
"omp_target_thread_limit"="987654321"
|
|
"amdgpu-max-num-workgroups"="987654321,987654321,987654321"
|
|
"amdgpu-flat-work-group-size"="987654321,987654321"
|
|
"amdgpu-waves-per-eu"="987654321,987654321"
|
|
}
|
|
|
|
!llvm.module.flags = !{!0}
|
|
!llvm.dbg.cu = !{!1}
|
|
|
|
!0 = !{i32 2, !"Debug Info Version", i32 3}
|
|
!1 = distinct !DICompileUnit(language: DW_LANG_C11, file: !2, producer: "clang version 19.0.0git", isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, splitDebugInlining: false, nameTableKind: None)
|
|
!2 = !DIFile(filename: "test.c", directory: "/tmp")
|
|
!3 = !{}
|
|
!4 = !DISubroutineType(types: !3)
|
|
!5 = distinct !DISubprogram(name: "all", scope: !2, file: !2, line: 10, type: !4, scopeLine: 10, flags: DIFlagArtificial | DIFlagPrototyped, spFlags: DISPFlagLocalToUnit | DISPFlagDefinition, unit: !1, retainedNodes: !3)
|
|
!6 = distinct !DISubprogram(name: "none", scope: !2, file: !2, line: 11, type: !4, scopeLine: 11, flags: DIFlagPrototyped, spFlags: DISPFlagLocalToUnit | DISPFlagDefinition, unit: !1, retainedNodes: !3)
|
|
!7 = distinct !DISubprogram(name: "bogus", scope: !2, file: !2, line: 12, type: !4, scopeLine: 12, flags: DIFlagPrototyped, spFlags: DISPFlagLocalToUnit | DISPFlagDefinition, unit: !1, retainedNodes: !3)
|