[Mlir-commits] [mlir] [NFC][MLIR][OpenMP] Add test to check lowering omp.wsloop inside target region (PR #74857)
Dominik Adamski
llvmlistbot at llvm.org
Mon Dec 11 04:35:11 PST 2023
https://github.com/DominikAdamski updated https://github.com/llvm/llvm-project/pull/74857
>From 160484e8b1584a2334bcb5dcf75b5c81e76da879 Mon Sep 17 00:00:00 2001
From: Dominik Adamski <dominik.adamski at amd.com>
Date: Fri, 8 Dec 2023 10:09:47 -0600
Subject: [PATCH 1/2] [NFC][MLIR][OpenMP] Add test to check lowering omp.wsloop
inside target region
This test checks if proper OpenMP device RTL function is called to handle
workshare loop inside target region.
The code generation for GPU worksharing loops is implemented by the patch:
https://github.com/llvm/llvm-project/pull/73360
---
mlir/test/Target/LLVMIR/omptarget-wsloop.mlir | 52 +++++++++++++++++++
1 file changed, 52 insertions(+)
create mode 100644 mlir/test/Target/LLVMIR/omptarget-wsloop.mlir
diff --git a/mlir/test/Target/LLVMIR/omptarget-wsloop.mlir b/mlir/test/Target/LLVMIR/omptarget-wsloop.mlir
new file mode 100644
index 0000000000000..973465003720f
--- /dev/null
+++ b/mlir/test/Target/LLVMIR/omptarget-wsloop.mlir
@@ -0,0 +1,52 @@
+// 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, omp.target = #omp.target<target_cpu = "gfx90a", target_features = "">} {
+ llvm.func @writeindex_(%arg0: !llvm.ptr {fir.bindc_name = "int_array"}) attributes {fir.internal_name = "_QPwriteindex", omp.declare_target = #omp.declaretarget<device_type = (host), capture_clause = (to)>} {
+ %0 = llvm.mlir.constant(9 : index) : i64
+ %1 = llvm.mlir.constant(0 : index) : i64
+ %2 = llvm.mlir.constant(1 : index) : i64
+ %3 = llvm.mlir.constant(10 : index) : i64
+ %6 = omp.bounds lower_bound(%1 : i64) upper_bound(%0 : i64) extent(%3 : i64) stride(%2 : i64) start_idx(%2 : i64)
+ %7 = omp.map_info var_ptr(%arg0 : !llvm.ptr, !llvm.array<10 x i32>) map_clauses(from) capture(ByRef) bounds(%6) -> !llvm.ptr {name = "int_array"}
+ omp.target map_entries(%7 -> %arg1 : !llvm.ptr) {
+ ^bb0(%arg1: !llvm.ptr):
+ %9 = llvm.mlir.constant(1 : i64) : i64
+ %10 = llvm.mlir.constant(10 : i32) : i32
+ %11 = llvm.mlir.constant(1 : i32) : i32
+ %12 = llvm.mlir.constant(1 : i64) : i64
+ %13 = llvm.alloca %12 x i32 {pinned} : (i64) -> !llvm.ptr
+ omp.wsloop for (%arg4) : i32 = (%11) to (%10) inclusive step (%11) {
+ llvm.store %arg4, %13 : i32, !llvm.ptr
+ %14 = llvm.load %13 : !llvm.ptr -> i32
+ %15 = llvm.sext %14 : i32 to i64
+ %16 = llvm.sub %15, %9 : i64
+ %17 = llvm.getelementptr %arg1[0, %16] : (!llvm.ptr, i64) -> !llvm.ptr, !llvm.array<10 x i32>
+ llvm.store %14, %17 : i32, !llvm.ptr
+ omp.yield
+ }
+ omp.terminator
+ }
+ llvm.return
+ }
+}
+
+// CHECK: define weak_odr protected amdgpu_kernel void @[[FUNC0:.*]](
+// CHECK-SAME: ptr %[[ARG0:.*]], ptr %[[ARG1:.*]])
+// CHECK: %[[STRUCTARG:.*]] = alloca { ptr, ptr }, align 8, addrspace(5)
+// CHECK: %[[STRUCTARG_ASCAST:.*]] = addrspacecast ptr addrspace(5) %[[STRUCTARG]] to ptr
+// CHECK: store ptr %[[ARG1]], ptr %[[TMP0:.*]], align 8
+// CHECK: %[[TMP1:.*]] = load ptr, ptr %[[TMP0]], align 8
+// CHECK: %[[GEP:.*]] = getelementptr { ptr, ptr }, ptr addrspace(5) %[[STRUCTARG]], i32 0, i32 1
+// CHECK: store ptr %[[TMP1]], ptr addrspace(5) %[[GEP]], align 8
+// CHECK: %[[NUM_THREADS:.*]] = call i32 @omp_get_num_threads()
+// CHECK: call void @__kmpc_for_static_loop_4u(ptr addrspacecast (ptr addrspace(1) @[[GLOB1:[0-9]+]] to ptr), ptr @[[LOOP_BODY_FN:.*]], ptr %[[STRUCTARG_ASCAST]], i32 10, i32 %[[NUM_THREADS]], i32 0)
+
+// CHECK: define internal void @[[LOOP_BODY_FN]](i32 %[[LOOP_CNT:.*]], ptr %[[LOOP_BODY_ARG:.*]])
+// CHECK: %[[GEP2:.*]] = getelementptr { ptr, ptr }, ptr %[[LOOP_BODY_ARG]], i32 0, i32 1
+// CHECK: %[[LOADGEP:.*]] = load ptr, ptr %[[GEP2]], align 8
+// CHECK: %[[GEP3:.*]] = getelementptr [10 x i32], ptr %[[LOADGEP]], i32 0, i64 %[[TMP2:.*]]
+// CHECK: store i32 %[[VAL0:.*]], ptr %[[GEP3]], align 4
+
>From 7f9cf902931d47709f16dba5295334d6d4c02c3d Mon Sep 17 00:00:00 2001
From: Dominik Adamski <dominik.adamski at amd.com>
Date: Mon, 11 Dec 2023 06:28:40 -0600
Subject: [PATCH 2/2] Applied remarks
---
mlir/test/Target/LLVMIR/omptarget-wsloop.mlir | 44 +++++++++----------
1 file changed, 22 insertions(+), 22 deletions(-)
diff --git a/mlir/test/Target/LLVMIR/omptarget-wsloop.mlir b/mlir/test/Target/LLVMIR/omptarget-wsloop.mlir
index 973465003720f..0040c5b887089 100644
--- a/mlir/test/Target/LLVMIR/omptarget-wsloop.mlir
+++ b/mlir/test/Target/LLVMIR/omptarget-wsloop.mlir
@@ -4,27 +4,27 @@
// 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, omp.target = #omp.target<target_cpu = "gfx90a", target_features = "">} {
- llvm.func @writeindex_(%arg0: !llvm.ptr {fir.bindc_name = "int_array"}) attributes {fir.internal_name = "_QPwriteindex", omp.declare_target = #omp.declaretarget<device_type = (host), capture_clause = (to)>} {
- %0 = llvm.mlir.constant(9 : index) : i64
- %1 = llvm.mlir.constant(0 : index) : i64
- %2 = llvm.mlir.constant(1 : index) : i64
- %3 = llvm.mlir.constant(10 : index) : i64
- %6 = omp.bounds lower_bound(%1 : i64) upper_bound(%0 : i64) extent(%3 : i64) stride(%2 : i64) start_idx(%2 : i64)
+ // This MLIR function corresponds to OpenMP code:
+ // #pragma omp target map(from: a)
+ // {
+ // #pragma omp for
+ // for (int i = 0; i <9; ++i)
+ // a[i] = i;
+ // }
+ llvm.func @target_wsloop(%arg0: !llvm.ptr ) attributes { omp.declare_target = #omp.declaretarget<device_type = (host), capture_clause = (to)>} {
+ %arr_ub = llvm.mlir.constant(9 : index) : i64
+ %arr_lb = llvm.mlir.constant(0 : index) : i64
+ %arr_stride = llvm.mlir.constant(1 : index) : i64
+ %6 = omp.bounds lower_bound(%arr_lb : i64) upper_bound(%arr_ub : i64) extent(%arr_ub : i64) stride(%arr_stride : i64) start_idx(%arr_lb : i64)
%7 = omp.map_info var_ptr(%arg0 : !llvm.ptr, !llvm.array<10 x i32>) map_clauses(from) capture(ByRef) bounds(%6) -> !llvm.ptr {name = "int_array"}
omp.target map_entries(%7 -> %arg1 : !llvm.ptr) {
^bb0(%arg1: !llvm.ptr):
- %9 = llvm.mlir.constant(1 : i64) : i64
- %10 = llvm.mlir.constant(10 : i32) : i32
- %11 = llvm.mlir.constant(1 : i32) : i32
- %12 = llvm.mlir.constant(1 : i64) : i64
- %13 = llvm.alloca %12 x i32 {pinned} : (i64) -> !llvm.ptr
- omp.wsloop for (%arg4) : i32 = (%11) to (%10) inclusive step (%11) {
- llvm.store %arg4, %13 : i32, !llvm.ptr
- %14 = llvm.load %13 : !llvm.ptr -> i32
- %15 = llvm.sext %14 : i32 to i64
- %16 = llvm.sub %15, %9 : i64
- %17 = llvm.getelementptr %arg1[0, %16] : (!llvm.ptr, i64) -> !llvm.ptr, !llvm.array<10 x i32>
- llvm.store %14, %17 : i32, !llvm.ptr
+ %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 (%arg4) : i32 = (%loop_lb) to (%loop_ub) inclusive step (%loop_step) {
+ %17 = llvm.getelementptr %arg1[0, %arg4] : (!llvm.ptr, i32) -> !llvm.ptr, !llvm.array<10 x i32>
+ llvm.store %arg4, %17 : i32, !llvm.ptr
omp.yield
}
omp.terminator
@@ -35,18 +35,18 @@ module attributes {dlti.dl_spec = #dlti.dl_spec<#dlti.dl_entry<"dlti.alloca_memo
// CHECK: define weak_odr protected amdgpu_kernel void @[[FUNC0:.*]](
// CHECK-SAME: ptr %[[ARG0:.*]], ptr %[[ARG1:.*]])
-// CHECK: %[[STRUCTARG:.*]] = alloca { ptr, ptr }, align 8, addrspace(5)
+// CHECK: %[[STRUCTARG:.*]] = alloca { ptr }, align 8, addrspace(5)
// CHECK: %[[STRUCTARG_ASCAST:.*]] = addrspacecast ptr addrspace(5) %[[STRUCTARG]] to ptr
// CHECK: store ptr %[[ARG1]], ptr %[[TMP0:.*]], align 8
// CHECK: %[[TMP1:.*]] = load ptr, ptr %[[TMP0]], align 8
-// CHECK: %[[GEP:.*]] = getelementptr { ptr, ptr }, ptr addrspace(5) %[[STRUCTARG]], i32 0, i32 1
+// CHECK: %[[GEP:.*]] = getelementptr { ptr }, ptr addrspace(5) %[[STRUCTARG]], i32 0, i32 0
// CHECK: store ptr %[[TMP1]], ptr addrspace(5) %[[GEP]], align 8
// CHECK: %[[NUM_THREADS:.*]] = call i32 @omp_get_num_threads()
// CHECK: call void @__kmpc_for_static_loop_4u(ptr addrspacecast (ptr addrspace(1) @[[GLOB1:[0-9]+]] to ptr), ptr @[[LOOP_BODY_FN:.*]], ptr %[[STRUCTARG_ASCAST]], i32 10, i32 %[[NUM_THREADS]], i32 0)
// CHECK: define internal void @[[LOOP_BODY_FN]](i32 %[[LOOP_CNT:.*]], ptr %[[LOOP_BODY_ARG:.*]])
-// CHECK: %[[GEP2:.*]] = getelementptr { ptr, ptr }, ptr %[[LOOP_BODY_ARG]], i32 0, i32 1
+// CHECK: %[[GEP2:.*]] = getelementptr { ptr }, ptr %[[LOOP_BODY_ARG]], i32 0, i32 0
// CHECK: %[[LOADGEP:.*]] = load ptr, ptr %[[GEP2]], align 8
-// CHECK: %[[GEP3:.*]] = getelementptr [10 x i32], ptr %[[LOADGEP]], i32 0, i64 %[[TMP2:.*]]
+// CHECK: %[[GEP3:.*]] = getelementptr [10 x i32], ptr %[[LOADGEP]], i32 0, i32 %[[TMP2:.*]]
// CHECK: store i32 %[[VAL0:.*]], ptr %[[GEP3]], align 4
More information about the Mlir-commits
mailing list