fix(dnn/cuda): fix int4 epilogue stg bug

GitOrigin-RevId: e86da9a8a8

dnn/test/common/checker.cpp

@@ -182,10 +182,10 @@ CheckerHelper::CheckerHelper(Handle *handle, bool check_dispatch):
     const char* env_p = std::getenv("MGB_NO_NAIVE_CHECK");
     if (env_p) {
         int no_naive_flag = atoi(env_p);
-        no_naive_and_check = no_naive_flag > 0 ? true : false;
+        m_no_naive_and_check = no_naive_flag > 0 ? true : false;
         check_dispatch = false;
     } else {
-        no_naive_and_check = false;
+        m_no_naive_and_check = false;
     }
     auto tmp_handle = create_cpu_handle(2, check_dispatch);
     m_handle_naive = std::move(tmp_handle);
@@ -282,7 +282,18 @@ void CheckerHelper::do_exec(const TensorLayoutArray &user_layouts,
         m_expect_exec_fail = {};
         return;
     }
-    if (no_naive_and_check){
+    if (m_stable_check) {
+        auto tensors_bak_host_storage =
+                alloc_tensors(m_handle_naive.get(), layouts, m_offset);
+        auto&& tensors_bak_host = *tensors_bak_host_storage;
+        copy_tensors_from_device(tensors_bak_host, tensors_cur);
+        for (int i = 0; i < 10; i++) {
+            exec_opr(tensors_cur);
+            copy_tensors_from_device(tensors_cur_host, tensors_cur);
+            check_tensors(tensors_bak_host, tensors_cur_host);
+        }
+    }
+    if (m_no_naive_and_check) {
         m_prev_succ = !::testing::Test::HasFailure();
         return;
     }

dnn/test/common/checker.h

@@ -76,7 +76,8 @@ protected:
     ExtraOprImpl m_extra_opr_impl;
     OutputCanonizer m_output_canonizer;
     TensorsConstriant m_tensor_constraint;
-    bool no_naive_and_check = false;
+    bool m_no_naive_and_check = false;
+    bool m_stable_check = false;
     /**
      * the offset from the start of malloc memory
      *
@@ -230,6 +231,17 @@ public:
         return *this;
     }
 
+    //! stable check will run many iter and compare result with first iter
+    Checker& set_stable_check(bool stable_check) {
+        m_stable_check = stable_check;
+        return *this;
+    }
+
+    Checker& set_no_naive_check(bool no_naive_and_check) {
+        m_no_naive_and_check = no_naive_and_check;
+        return *this;
+    }
+
     //! load input tensors from file for next run
     Checker& load_input_tensors(const char* fpath) {
         m_input_tensors_fpath = fpath;

dnn/test/common/conv_bias.cpp

@@ -731,9 +731,10 @@ std::vector<TestArg> get_int8_chwn4_tensorcore_args(size_t kernel_size) {
 void check_conv_bias(DType src_dtype, DType filter_dtype, DType bias_dtype,
                      DType dst_dtype, Handle* handle, const char* algo,
                      param::ConvBias::Format format,
-                     const std::vector<TestArg>& args, bool fuse_z) {
+                     const std::vector<TestArg>& args, bool fuse_z,
+                     bool stable_test) {
     megdnn_assert(src_dtype.enumv() == filter_dtype.enumv());
-    Checker<ConvBiasForward> checker(handle);
+    Checker<ConvBiasForward> checker(handle, !stable_test);
     if (algo) {
         checker.set_before_exec_callback(
                 ConvBiasAlgoChecker<ConvBiasForward>(algo));
@@ -823,6 +824,10 @@ void check_conv_bias(DType src_dtype, DType filter_dtype, DType bias_dtype,
             .set_rng(1, rng.get())
             .set_rng(2, bias_rng.get())
             .set_rng(3, rng.get());
+    if (stable_test) {
+        checker.set_stable_check(true);
+        checker.set_no_naive_check(true);
+    }
     if (args.empty()) {
         std::vector<TestArg> default_args;
         if (format == Format::NCHW4) {

dnn/test/common/conv_bias.h

@@ -69,7 +69,7 @@ void check_conv_bias(
         DType src_dtype, DType filter_dtype, DType bias_dtype, DType dst_dtype,
         Handle* handle, const char* algo = nullptr,
         param::ConvBias::Format format = param::ConvBias::Format::NCHW4,
-        const std::vector<TestArg>& args = {}, bool fuse_z = false);
+        const std::vector<TestArg>& args = {}, bool fuse_z = false, bool stable_test = false);
 
 #if MEGDNN_WITH_BENCHMARK
 std::vector<conv_bias::TestArg> get_winograd_benchmark_args(

dnn/test/common/warp_perspective.h

@@ -71,7 +71,7 @@ public:
         auto rand_real2 = [&](double range) {
             return rand_real(-range, range);
         };
-        dt_float32 res;
+        dt_float32 res = 0;
         switch (idx) {
             case 0:
                 rot = rand_real(0, M_PI * 2);

dnn/test/cuda/utils.cpp

@@ -28,6 +28,13 @@ bool check_compute_capability_eq(int major, int minor) {
     cuda_check(cudaGetDeviceProperties(&prop, dev));
     return (prop.major == major && prop.minor == minor);
 }
+const cudaDeviceProp current_cuda_device_prop() {
+    int dev;
+    cuda_check(cudaGetDevice(&dev));
+    cudaDeviceProp prop;
+    cuda_check(cudaGetDeviceProperties(&prop, dev));
+    return prop;
+}
 }  // namespace test
 }  // namespace megdnn
 

dnn/test/cuda/utils.h

@@ -27,6 +27,7 @@ namespace megdnn {
 namespace test {
 bool check_compute_capability(int major, int minor);
 bool check_compute_capability_eq(int major, int minor);
+const cudaDeviceProp current_cuda_device_prop();
 }  // namespace test
 }  // namespace megdnn