[NFC][OpenMP][MLIR] Add test for lowering parallel workshare GPU loop

[DominikAdamski]

This test checks if MLIR code is lowered according to schema presented below:

func1() {
call __kmpc_parallel_51(..., func2, ...)
}

func2() {
call __kmpc_for_static_loop_4u(..., func3, ...)
}

func3() {
//loop body
}

[llvmbot]

@llvm/pr-subscribers-mlir

@llvm/pr-subscribers-mlir-llvm

Author: Dominik Adamski (DominikAdamski)

Changes

This test checks if MLIR code is lowered according to schema presented below:

func1() {
call __kmpc_parallel_51(..., func2, ...)
}

func2() {
call __kmpc_for_static_loop_4u(..., func3, ...)
}

func3() {
//loop body
}

---

Full diff: https://github.com/llvm/llvm-project/pull/76144.diff

1 Files Affected:

• (added) mlir/test/Target/LLVMIR/omptarget-parallel-wsloop.mlir (+36)

diff --git a/mlir/test/Target/LLVMIR/omptarget-parallel-wsloop.mlir b/mlir/test/Target/LLVMIR/omptarget-parallel-wsloop.mlir
new file mode 100644
index 00000000000000..43d0934d3a931e
--- /dev/null
+++ b/mlir/test/Target/LLVMIR/omptarget-parallel-wsloop.mlir
@@ -0,0 +1,36 @@
+// RUN: mlir-translate -mlir-to-llvmir %s | FileCheck %s
+
+// The aim of the test is to check the GPU LLVM IR codegen
+// for nested omp do loop inside omp target region
+
+module attributes {dlti.dl_spec = #dlti.dl_spec<#dlti.dl_entry<"dlti.alloca_memory_space", 5 : ui32>>, llvm.data_layout = "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-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", llvm.target_triple = "amdgcn-amd-amdhsa", omp.is_gpu = true, omp.is_target_device = true } {
+  llvm.func @target_parallel_wsloop(%arg0: !llvm.ptr ){
+    omp.parallel {
+      %loop_ub = llvm.mlir.constant(9 : i32) : i32
+      %loop_lb = llvm.mlir.constant(0 : i32) : i32
+      %loop_step = llvm.mlir.constant(1 : i32) : i32
+      omp.wsloop for  (%loop_cnt) : i32 = (%loop_lb) to (%loop_ub) inclusive step (%loop_step) {
+        %gep = llvm.getelementptr %arg0[0, %loop_cnt] : (!llvm.ptr, i32) -> !llvm.ptr, !llvm.array<10 x i32>
+        llvm.store %loop_cnt, %gep : i32, !llvm.ptr
+        omp.yield
+      }
+     omp.terminator
+    }
+
+    llvm.return
+  }
+
+}
+// CHECK:      call void @__kmpc_parallel_51(ptr addrspacecast
+// CHECK-SAME:  (ptr addrspace(1) @[[GLOB:[0-9]+]] to ptr),
+// CHECK-SAME:  i32 %[[THREAD_NUM:.*]], i32 1, i32 -1, i32 -1,
+// CHECK-SAME:  ptr @[[PARALLEL_FUNC:.*]], ptr null, ptr %[[PARALLEL_ARGS:.*]], i64 1)
+
+// CHECK:      define internal void @[[PARALLEL_FUNC]]
+// CHECK-SAME:  (ptr noalias noundef %[[TID_ADDR:.*]], ptr noalias noundef %[[ZERO_ADDR:.*]],
+// CHECK-SAME:  ptr %[[ARG_PTR:.*]])
+// CHECK: call void @__kmpc_for_static_loop_4u(ptr addrspacecast (ptr addrspace(1) @[[GLOB]] to ptr),
+// CHECK-SAME:   ptr @[[LOOP_BODY_FUNC:.*]], ptr %[[LOO_BODY_FUNC_ARG:.*]], i32 10,
+// CHECK-SAME:   i32 %[[THREAD_NUM:.*]], i32 0)
+
+// CHECK:      define internal void @[[LOOP_BODY_FUNC]](i32 %[[CNT:.*]], ptr %[[LOOP_BODY_ARG_PTR:.*]]) {

[kiranchandramohan]

Could you share the command-line flags that have to be used with flang-new to get the llvm ir in this test?

[kiranchandramohan]

Could you share the command-line flags that have to be used with flang-new to get the llvm ir in this test?

I guess it is something like the following.

flang-new -fc1 -triple amdgcn-amd-amdhsa -emit-llvm -fopenmp -fopenmp-is-target-device do.f90

[DominikAdamski]

The MLIR test case was reduced by hand.

The initial code was similar to:

subroutine test(i, beta) bind(C)
!$omp declare target
        use ISO_C_BINDING
        integer (C_INT), dimension(*), intent(out) :: beta
        integer (C_INT), value :: i
!$omp parallel
!$omp do
      do i = 1, 10
        beta(i) = i
      end do
!$omp end do
!$omp end parallel
end subroutine

I am able to generate test-openmp-amdgcn-amd-amdhsa-gfx90a-llvmir.mlir file by command:
flang-new -save-temps -c -fopenmp --offload-arch=gfx90a test.f95

Or:
flang-new -fc1 -triple amdgcn-amd-amdhsa -emit-llvm-bc -fopenmp -fcolor-diagnostics -mrelocation-model pic -pic-level 2 -target-cpu gfx90a -fopenmp-host-ir-file-path test-host-x86_64-unknown-linux-gnu.bc -fopenmp-is-target-device -mframe-pointer=all -o test-openmp-amdgcn-amd-amdhsa-gfx90a.tmp.bc -save-temps=cwd -x f95 test-openmp-amdgcn-amd-amdhsa-gfx90a.i