refactor(dnn/cuda): remove function name weiming

GitOrigin-RevId: f36495a46a36ab0976f3c70254af95e07cd92a80

dnn/src/cuda/local/forward.cpp

@@ -14,43 +14,7 @@
 #include "src/cuda/utils.h"
 #include "src/cuda/handle.h"
 
-namespace megdnn {
-namespace cuda {
-namespace local {
-
-void check_input(size_t N,
-        size_t IC, size_t IH, size_t IW,
-        size_t OC, size_t OH, size_t OW,
-        size_t FH, size_t FW,
-        size_t INs, size_t ONs,
-        size_t PH, size_t PW,
-        size_t SH, size_t SW,
-        bool is_xcorr)
-{
-    megdnn_ignore(N);
-    megdnn_ignore(IC);
-    megdnn_ignore(IH);
-    megdnn_ignore(IW);
-    megdnn_ignore(OC);
-    megdnn_ignore(OH);
-    megdnn_ignore(OW);
-    megdnn_ignore(FH);
-    megdnn_ignore(FW);
-    megdnn_ignore(INs);
-    megdnn_ignore(ONs);
-    megdnn_ignore(PH);
-    megdnn_ignore(PW);
-    megdnn_ignore(SH);
-    megdnn_ignore(SW);
-    megdnn_ignore(is_xcorr);
-    // shared memory constraint
-    megdnn_assert(IH*IW <= 768, "spatial size should not be larger than 768.");
-    // megdnn_assert(4 * 4 * 4 * IH * IW <= 49152);
-}
-
-} // namespace local
-} // namespace cuda
-} // namespace megdnn
+#include "src/common/utils.cuh"
 
 namespace megdnn {
 namespace cuda {
@@ -94,13 +58,9 @@ void LocalForwardImpl::exec(_megdnn_tensor_in src,
                 param().stride_h, param().stride_w,
                 cublas, stream,
                 one, zero);
-    } else {
-        local::check_input(N, IC, IH, IW, OC, OH, OW, FH, FW,
-                IC*IH*IW, OC*OH*OW,
-                param().pad_h, param().pad_w,
-                param().stride_h, param().stride_w,
-                is_xcorr);
-        local::forward_proxy_weiming(src.ptr<dt_float32>(),
+    } else if (local::forward_proxy_default_share_mem_in_bytes(IH, IW) <=
+               handle->device_prop().sharedMemPerBlock) {
+        local::forward_proxy_default(src.ptr<dt_float32>(),
                 filter.ptr<dt_float32>(),
                 dst.ptr<dt_float32>(),
                 N,
@@ -112,6 +72,11 @@ void LocalForwardImpl::exec(_megdnn_tensor_in src,
                 param().stride_h, param().stride_w,
                 is_xcorr,
                 stream);
+    } else {
+        megdnn_throw(ssprintf(
+                "No usable kernel for local conv, src: %s filter: %s \n",
+                src.layout.to_string().c_str(),
+                filter.layout.to_string().c_str()));
     }
 }
 

dnn/src/cuda/local/forward.cu

@@ -18,6 +18,12 @@ namespace megdnn {
 namespace cuda {
 namespace local {
 
+constexpr size_t Ns = 4, ICs = 4;
+
+size_t forward_proxy_default_share_mem_in_bytes(size_t IH, size_t IW) {
+    return Ns * ICs * sizeof(float) * IH * IW;
+}
+
 // blockIdx.y is OC*OH*OW/1024
 // blockIdx.x is N/4
 // threadIdx.x is [0, 1024)
@@ -96,7 +102,7 @@ __global__ void forward_kernel(const float * __restrict__ src,
     }
 }
 
-void forward_proxy_weiming(const float *src, const float *filter, float *dst,
+void forward_proxy_default(const float *src, const float *filter, float *dst,
         size_t N,
         size_t IC, size_t IH, size_t IW,
         size_t OC, size_t OH, size_t OW,
@@ -108,7 +114,6 @@ void forward_proxy_weiming(const float *src, const float *filter, float *dst,
         cudaStream_t stream)
 {
     size_t threads = 256;
-    const size_t Ns = 4, ICs = 4;
     dim3 blocks = dim3(DIVUP(N, Ns), DIVUP(OC*OH*OW, threads));
     if (is_xcorr) {
         forward_kernel<Ns, ICs, true><<<blocks, threads,

dnn/src/cuda/local/local.cuh

@@ -17,17 +17,10 @@ namespace megdnn {
 namespace cuda {
 namespace local {
 
-void check_input(size_t N, 
-        size_t IC, size_t IH, size_t IW,
-        size_t OC, size_t OH, size_t OW,
-        size_t FH, size_t FW,
-        size_t INs, size_t ONs,
-        size_t PH, size_t PW,
-        size_t SH, size_t SW,
-        bool is_xcorr);
+size_t forward_proxy_default_share_mem_in_bytes(size_t IH, size_t IW);
 
-void forward_proxy_weiming(const float *src, const float *filter, float *dst,
-        size_t N, 
+void forward_proxy_default(const float *src, const float *filter, float *dst,
+        size_t N,
         size_t IC, size_t IH, size_t IW,
         size_t OC, size_t OH, size_t OW,
         size_t FH, size_t FW,
@@ -39,7 +32,7 @@ void forward_proxy_weiming(const float *src, const float *filter, float *dst,
 
 /// forward
 
-bool can_forward_proxy_convnet(size_t N, 
+bool can_forward_proxy_convnet(size_t N,
         size_t IC, size_t IH, size_t IW,
         size_t OC, size_t OH, size_t OW,
         size_t FH, size_t FW,
@@ -70,7 +63,7 @@ size_t get_workspace_in_floats_forward_proxy_convnet(size_t N,
 
 /// bwd data
 
-bool can_backward_data_proxy_convnet(size_t N, 
+bool can_backward_data_proxy_convnet(size_t N,
         size_t IC, size_t IH, size_t IW,
         size_t OC, size_t OH, size_t OW,
         size_t FH, size_t FW,
@@ -78,7 +71,7 @@ bool can_backward_data_proxy_convnet(size_t N,
         size_t PH, size_t PW,
         size_t SH, size_t SW);
 
-void backward_data_proxy_convnet(const float *filter, 
+void backward_data_proxy_convnet(const float *filter,
         const float *diff,
         float *grad,
         float *workspace,
@@ -103,7 +96,7 @@ size_t get_workspace_in_floats_backward_data_proxy_convnet(size_t N,
 
 /// bwd filter
 
-bool can_backward_filter_proxy_convnet(size_t N, 
+bool can_backward_filter_proxy_convnet(size_t N,
         size_t IC, size_t IH, size_t IW,
         size_t OC, size_t OH, size_t OW,
         size_t FH, size_t FW,