Merge fixes.

NVIDIA · Nov 4, 2024 · 097674d · 097674d
1 parent d372ae2
commit 097674d
Show file tree

Hide file tree

Showing 4 changed files with 25 additions and 25 deletions.
diff --git a/tests/cpp/test_combined_inner_outer_reduction.cpp b/tests/cpp/test_combined_inner_outer_reduction.cpp
@@ -982,8 +982,8 @@ TEST_F(CombinedSchedulerTest, SharedMemoryPersistentVectFactor) {
   heuristic_params->as<ReductionParams>()->smem_persistent_buffers =
       std::vector<TensorView*>{tv1};
   scheduler->schedule(&fusion, heuristic_params.get());
-  FusionExecutor fe;
-  fe.compileFusion(&fusion, aten_inputs);
+  KernelExecutor ke;
+  ke.compile(&fusion, aten_inputs);
 
   for (auto tv : fusion.allTvs()) {
     if (tv->getMemoryType() == MemoryType::Shared) {
@@ -992,8 +992,8 @@ TEST_F(CombinedSchedulerTest, SharedMemoryPersistentVectFactor) {
       }
     }
   }
-  auto cg_outputs = fe.runFusion(
-      aten_inputs, heuristic_params->as<ReductionParams>()->lparams);
+  auto cg_outputs =
+      ke.run(aten_inputs, heuristic_params->as<ReductionParams>()->lparams);
   testValidate(&fusion_copy, cg_outputs, aten_inputs, __LINE__, __FILE__);
 }
 

diff --git a/tests/cpp/test_gpu3.cpp b/tests/cpp/test_gpu3.cpp
@@ -8875,9 +8875,9 @@ TEST_F(NVFuserTest, CpAsyncDataTypeBool) {
   //    "r"((uint32_t)((!b3)))
   // );
   // If not correctly lowered, would trigger error in compile
-  FusionExecutor fe;
-  fe.compileFusion(&fusion, {t0});
-  auto cg_outputs = fe.runFusion({t0});
+  KernelExecutor ke;
+  ke.compile(&fusion, {t0});
+  auto cg_outputs = ke.run({t0});
   testValidate(&fusion, cg_outputs, {t0}, __LINE__, __FILE__);
 }
 // Test file size should be up to 10K LoC. Create a new file for more tests.

diff --git a/tests/cpp/test_matmul.cpp b/tests/cpp/test_matmul.cpp
@@ -185,19 +185,19 @@ TEST_P(MatmulTestWithLayout, AmpereMatmulBroadcastBatch) {
 
   auto inputs = matmulAtInput3DTuring(M, N, K, layout);
 
-  FusionExecutor fe;
+  KernelExecutor ke;
   NVFUSER_TEST_CUDA_ARCH_COMPILE_CHECK(
       8,
       0,
-      fe.compileFusion(
+      ke.compile(
           &fusion,
           {inputs.first, inputs.second},
           LaunchParams(),
           matmul_cparams));
-  ASSERT_TRUE(getBankConflictInfo(fe.kernel()).empty());
+  ASSERT_TRUE(getBankConflictInfo(ke.kernel()).empty());
   ASSERT_FALSE(
-      PredicatedChecker::isCpAsyncMmaPredicatedByIfThenElse(fe.kernel()));
-  auto cg_outputs = fe.runFusion({inputs.first, inputs.second});
+      PredicatedChecker::isCpAsyncMmaPredicatedByIfThenElse(ke.kernel()));
+  auto cg_outputs = ke.run({inputs.first, inputs.second});
   auto tref =
       atMatmul(
           inputs.first.to(at::kFloat), inputs.second.to(at::kFloat), layout)
@@ -2700,7 +2700,7 @@ TEST_F(MatmulTest, AmpereMatmulSmemEpiloguePromotionRequiredA100) {
   SchedulerEntry::makeSchedulerInstance(SchedulerType::Matmul)
       ->schedule(&fusion, &mparams);
 
-  // KernelExecutor::compileFusion would fail otherwise.
+  // KernelExecutor::compile would fail otherwise.
   SKIP_IF_INSUFFICIENT_SMEM(&mparams, data_types);
 
   at::manual_seed(0);

diff --git a/tests/cpp/test_matmul_scheduler.cpp b/tests/cpp/test_matmul_scheduler.cpp
@@ -3140,13 +3140,13 @@ TEST_F(MatmulSchedulerTest, HSH_TT) {
   //! TODO Disabled because hopper multiple matmul scheduler is currently a copy
   //! of ampere scheduler.
   /*
-  FusionExecutor fe;
-  fe.compileFusion(
+  KernelExecutor ke;
+  ke.compile(
       fusion.get(),
       {inputs.first, inputs.second},
       LaunchParams(),
       matmul_cparams);
-  auto cg_outputs = fe.runFusion({inputs.first, inputs.second});
+  auto cg_outputs = ke.run({inputs.first, inputs.second});
   auto tref = atMatmul(inputs.first.squeeze(), inputs.second.squeeze(), layout);
   EXPECT_TRUE(at::allclose(cg_outputs[0], tref, 1e-5, 1e-5));
   */
@@ -3211,14 +3211,14 @@ TEST_F(MatmulSchedulerTest, HSH_TN) {
   auto inputs =
       matmulAtInput3DHopperSS(M, N, K, layout, data_type_to_aten(dtype));
 
-  FusionExecutor fe;
-  fe.compileFusion(
+  KernelExecutor ke;
+  ke.compile(
       fusion.get(),
       {inputs.first, inputs.second},
       LaunchParams(),
       matmul_cparams);
 
-  auto cg_outputs = fe.runFusion({inputs.first, inputs.second});
+  auto cg_outputs = ke.run({inputs.first, inputs.second});
   auto tref = atMatmul(inputs.first.squeeze(), inputs.second.squeeze(), layout);
   EXPECT_TRUE(at::allclose(cg_outputs[0], tref, 1e-5, 1e-5));
 }
@@ -3286,14 +3286,14 @@ TEST_F(MatmulSchedulerTest, HSH_NT) {
   auto inputs =
       matmulAtInput3DHopperSS(M, N, K, layout, data_type_to_aten(dtype));
 
-  FusionExecutor fe;
-  fe.compileFusion(
+  KernelExecutor ke;
+  ke.compile(
       fusion.get(),
       {inputs.first, inputs.second},
       LaunchParams(),
       matmul_cparams);
 
-  auto cg_outputs = fe.runFusion({inputs.first, inputs.second});
+  auto cg_outputs = ke.run({inputs.first, inputs.second});
   auto tref = atMatmul(inputs.first.squeeze(), inputs.second.squeeze(), layout);
   EXPECT_TRUE(at::allclose(cg_outputs[0], tref, 1e-5, 1e-5));
 }
@@ -3363,13 +3363,13 @@ TEST_F(MatmulSchedulerTest, HSH_NN) {
   // TODO Disabled because hopper multiple matmul scheduler is currently a copy
   // of ampere scheduler.
   /*
-  FusionExecutor fe;
-  fe.compileFusion(
+  KernelExecutor ke;
+  ke.compile(
       fusion.get(),
       {inputs.first, inputs.second},
       LaunchParams(),
       matmul_cparams);
-  auto cg_outputs = fe.runFusion({inputs.first, inputs.second});
+  auto cg_outputs = ke.run({inputs.first, inputs.second});
   auto tref = atMatmul(inputs.first.squeeze(), inputs.second.squeeze(), layout);
   EXPECT_TRUE(at::allclose(cg_outputs[0], tref, 1e-5, 1e-5));
   */