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`.
1.1 KiB
1.1 KiB
The tests in this directory check that basic KernelInfoPrinter functionality behaves reasonably for LLVM IR produced by Clang OpenMP codegen.
So that these tests are straightforward to maintain and faithfully represent Clang OpenMP codegen, do not tweak or reduce the LLVM IR in them. Other tests more exhaustively check KernelInfoPrinter features using reduced LLVM IR.
The LLVM IR in each test file $TEST can be regenerated as follows in the case
that Clang OpenMP codegen changes or it becomes desirable to adjust the source
OpenMP program below. First, remove the existing LLVM IR from $TEST. Then,
where $TARGET (e.g., nvptx64-nvidia-cuda-sm_70 or amdgcn-amd-amdhsa-gfx906)
depends on $TEST:
$ cd /tmp
$ cat test.c
#pragma omp declare target
void f();
void g() {
int i;
int a[2];
f();
g();
}
#pragma omp end declare target
void h(int i) {
#pragma omp target map(tofrom:i)
{
int i;
int a[2];
f();
g();
}
}
$ clang -g -fopenmp --offload-arch=native -save-temps -c test.c
$ llvm-dis test-openmp-$TARGET.bc
$ cat test-openmp-$TARGET.ll >> $TEST