Change CUDA tests to use print_memref.

Swap dimensions in all-reduce-op test. PiperOrigin-RevId: 281791744
llvm · Nov 21, 2019 · d7c1719 · d7c1719
1 parent c621e64
commit d7c1719
Show file tree

Hide file tree

Showing 4 changed files with 18 additions and 31 deletions.
diff --git a/mlir/test/mlir-cuda-runner/all-reduce-op.mlir b/mlir/test/mlir-cuda-runner/all-reduce-op.mlir
@@ -1,13 +1,13 @@
-// RUN: mlir-cuda-runner %s --shared-libs=%cuda_wrapper_library_dir/libcuda-runtime-wrappers%shlibext --entry-point-result=void | FileCheck %s
+// RUN: mlir-cuda-runner %s --shared-libs=%cuda_wrapper_library_dir/libcuda-runtime-wrappers%shlibext,%linalg_test_lib_dir/libmlir_runner_utils%shlibext --entry-point-result=void | FileCheck %s
 
-// CHECK: [5.356000e+03, 5.356000e+03, {{.*}}, 5.356000e+03, 5.356000e+03]
+// CHECK-COUNT-8: [{{(5356, ){12}5356}}]
 func @main() {
-  %arg = alloc() : memref<13x4x2xf32>
-  %dst = memref_cast %arg : memref<13x4x2xf32> to memref<?x?x?xf32>
+  %arg = alloc() : memref<2x4x13xf32>
+  %dst = memref_cast %arg : memref<2x4x13xf32> to memref<?x?x?xf32>
   %one = constant 1 : index
-  %sx = dim %dst, 0 : memref<?x?x?xf32>
+  %sx = dim %dst, 2 : memref<?x?x?xf32>
   %sy = dim %dst, 1 : memref<?x?x?xf32>
-  %sz = dim %dst, 2 : memref<?x?x?xf32>
+  %sz = dim %dst, 0 : memref<?x?x?xf32>
   call @mcuMemHostRegisterMemRef3dFloat(%dst) : (memref<?x?x?xf32>) -> ()
   gpu.launch blocks(%bx, %by, %bz) in (%grid_x = %one, %grid_y = %one, %grid_z = %one)
              threads(%tx, %ty, %tz) in (%block_x = %sx, %block_y = %sy, %block_z = %sz)
@@ -19,12 +19,12 @@ func @main() {
     %t3 = index_cast %idx : index to i32
     %val = sitofp %t3 : i32 to f32
     %sum = "gpu.all_reduce"(%val) ({}) { op = "add" } : (f32) -> (f32)
-    store %sum, %kernel_dst[%tx, %ty, %tz] : memref<?x?x?xf32>
+    store %sum, %kernel_dst[%tz, %ty, %tx] : memref<?x?x?xf32>
     gpu.return
   }
-  call @mcuPrintFloat(%dst) : (memref<?x?x?xf32>) -> ()
+  call @print_memref_3d_f32(%dst) : (memref<?x?x?xf32>) -> ()
   return
 }
 
 func @mcuMemHostRegisterMemRef3dFloat(%ptr : memref<?x?x?xf32>)
-func @mcuPrintFloat(%ptr : memref<?x?x?xf32>)
+func @print_memref_3d_f32(%ptr : memref<?x?x?xf32>)
diff --git a/mlir/test/mlir-cuda-runner/all-reduce-region.mlir b/mlir/test/mlir-cuda-runner/all-reduce-region.mlir
@@ -1,6 +1,6 @@
-// RUN: mlir-cuda-runner %s --shared-libs=%cuda_wrapper_library_dir/libcuda-runtime-wrappers%shlibext --entry-point-result=void | FileCheck %s
+// RUN: mlir-cuda-runner %s --shared-libs=%cuda_wrapper_library_dir/libcuda-runtime-wrappers%shlibext,%linalg_test_lib_dir/libmlir_runner_utils%shlibext --entry-point-result=void | FileCheck %s
 
-// CHECK: [3.500000e+01, 3.500000e+01, {{.*}}, 3.500000e+01, 3.500000e+01]
+// CHECK: [{{(35, ){34}35}}]
 func @main() {
   %arg = alloc() : memref<35xf32>
   %dst = memref_cast %arg : memref<35xf32> to memref<?xf32>
@@ -20,9 +20,9 @@ func @main() {
     store %res, %kernel_dst[%tx] : memref<?xf32>
     gpu.return
   }
-  call @mcuPrintFloat(%dst) : (memref<?xf32>) -> ()
+  call @print_memref_1d_f32(%dst) : (memref<?xf32>) -> ()
   return
 }
 
 func @mcuMemHostRegisterMemRef1dFloat(%ptr : memref<?xf32>)
-func @mcuPrintFloat(%ptr : memref<?xf32>)
+func @print_memref_1d_f32(memref<?xf32>)
diff --git a/mlir/test/mlir-cuda-runner/gpu-to-cubin.mlir b/mlir/test/mlir-cuda-runner/gpu-to-cubin.mlir
@@ -1,4 +1,4 @@
-// RUN: mlir-cuda-runner %s --shared-libs=%cuda_wrapper_library_dir/libcuda-runtime-wrappers%shlibext --entry-point-result=void | FileCheck %s
+// RUN: mlir-cuda-runner %s --shared-libs=%cuda_wrapper_library_dir/libcuda-runtime-wrappers%shlibext,%linalg_test_lib_dir/libmlir_runner_utils%shlibext --entry-point-result=void | FileCheck %s
 
 func @other_func(%arg0 : f32, %arg1 : memref<?xf32>) {
   %cst = constant 1 : index
@@ -12,18 +12,18 @@ func @other_func(%arg0 : f32, %arg1 : memref<?xf32>) {
   return
 }
 
-// CHECK: [1.000000e+00, 1.000000e+00, 1.000000e+00, 1.000000e+00, 1.000000e+00]
+// CHECK: [1, 1, 1, 1, 1]
 func @main() {
   %arg0 = alloc() : memref<5xf32>
   %21 = constant 5 : i32
   %22 = memref_cast %arg0 : memref<5xf32> to memref<?xf32>
   call @mcuMemHostRegisterMemRef1dFloat(%22) : (memref<?xf32>) -> ()
-  call @mcuPrintFloat(%22) : (memref<?xf32>) -> ()
+  call @print_memref_1d_f32(%22) : (memref<?xf32>) -> ()
   %24 = constant 1.0 : f32
   call @other_func(%24, %22) : (f32, memref<?xf32>) -> ()
-  call @mcuPrintFloat(%22) : (memref<?xf32>) -> ()
+  call @print_memref_1d_f32(%22) : (memref<?xf32>) -> ()
   return
 }
 
 func @mcuMemHostRegisterMemRef1dFloat(%ptr : memref<?xf32>)
-func @mcuPrintFloat(%ptr : memref<?xf32>)
+func @print_memref_1d_f32(memref<?xf32>)
diff --git a/mlir/tools/mlir-cuda-runner/cuda-runtime-wrappers.cpp b/mlir/tools/mlir-cuda-runner/cuda-runtime-wrappers.cpp
@@ -113,16 +113,3 @@ extern "C" void
 mcuMemHostRegisterMemRef3dFloat(const MemRefType<float, 3> *arg) {
   mcuMemHostRegisterMemRef(arg, 1.23f);
 }
-
-/// Prints the given float array to stderr.
-extern "C" void mcuPrintFloat(const MemRefType<float, 1> *arg) {
-  if (arg->sizes[0] == 0) {
-    llvm::outs() << "[]\n";
-    return;
-  }
-  llvm::outs() << "[" << arg->data[0];
-  for (int pos = 1; pos < arg->sizes[0]; pos++) {
-    llvm::outs() << ", " << arg->data[pos];
-  }
-  llvm::outs() << "]\n";
-}