Merge remote-tracking branch 'origin/develop' into bwroblew/direct_loads

docs/sphinx/requirements.txt

@@ -48,6 +48,12 @@ idna==3.4
     # via requests
 imagesize==1.4.1
     # via sphinx
+importlib-metadata==6.8.0
+    # via
+    #   sphinx
+    #   sphinxcontrib-bibtex
+importlib-resources==6.1.0
+    # via rocm-docs-core
 jinja2==3.1.2
     # via
     #   myst-parser
@@ -90,9 +96,13 @@ pygments==2.14.0
     #   pydata-sphinx-theme
     #   sphinx
 pyjwt[crypto]==2.6.0
-    # via pygithub
+    # via
+    #   pygithub
+    #   pyjwt
 pynacl==1.5.0
     # via pygithub
+pytz==2023.3.post1
+    # via babel
 pyyaml==6.0
     # via
     #   myst-parser
@@ -103,7 +113,7 @@ requests==2.28.2
     # via
     #   pygithub
     #   sphinx
-rocm-docs-core==0.26.0
+rocm-docs-core==0.27.0
     # via -r requirements.in
 six==1.16.0
     # via
@@ -157,3 +167,7 @@ urllib3==1.26.15
     # via requests
 wrapt==1.15.0
     # via deprecated
+zipp==3.17.0
+    # via
+    #   importlib-metadata
+    #   importlib-resources

include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_abd_xdl_cshuffle.hpp

@@ -836,7 +836,7 @@ struct DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle
             // check if it's 1x1, stride=1 conv
             for(index_t i = 0; i < NDimSpatial; ++i)
             {
-                const index_t X          = arg.b_g_k_c_xs_lengths_[i + 2];
+                const index_t X          = arg.b_g_k_c_xs_lengths_[i + 3];
                 const index_t ConvStride = arg.conv_filter_strides_[i];
                 const index_t LeftPad    = arg.input_left_pads_[i];
                 const index_t RightPad   = arg.input_right_pads_[i];
@@ -853,7 +853,7 @@ struct DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle
             // check if it's 1x1 conv
             for(index_t i = 0; i < NDimSpatial; ++i)
             {
-                const index_t X        = arg.b_g_k_c_xs_lengths_[i + 2];
+                const index_t X        = arg.b_g_k_c_xs_lengths_[i + 3];
                 const index_t LeftPad  = arg.input_left_pads_[i];
                 const index_t RightPad = arg.input_right_pads_[i];
 
@@ -1094,6 +1094,7 @@ struct DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle
             << NXdlPerWave << ", "
             << ABlockTransferSrcScalarPerVector << ", "
             << BBlockTransferSrcScalarPerVector << ", "
+            << CDEBlockTransferScalarPerVector_NPerBlock << ", "
             << CShuffleMXdlPerWavePerShuffle << ", "
             << CShuffleNXdlPerWavePerShuffle
             << ">";

include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r_xdl_cshuffle.hpp

@@ -834,7 +834,7 @@ struct DeviceGroupedConvFwdMultipleDMultipleR_Xdl_CShuffle
             // check if it's 1x1, stride=1 conv
             for(index_t i = 0; i < NDimSpatial; ++i)
             {
-                const index_t X          = arg.b_g_k_c_xs_lengths_[i + 2];
+                const index_t X          = arg.b_g_k_c_xs_lengths_[i + 3];
                 const index_t ConvStride = arg.conv_filter_strides_[i];
                 const index_t LeftPad    = arg.input_left_pads_[i];
                 const index_t RightPad   = arg.input_right_pads_[i];
@@ -851,7 +851,7 @@ struct DeviceGroupedConvFwdMultipleDMultipleR_Xdl_CShuffle
             // check if it's 1x1 conv
             for(index_t i = 0; i < NDimSpatial; ++i)
             {
-                const index_t X        = arg.b_g_k_c_xs_lengths_[i + 2];
+                const index_t X        = arg.b_g_k_c_xs_lengths_[i + 3];
                 const index_t LeftPad  = arg.input_left_pads_[i];
                 const index_t RightPad = arg.input_right_pads_[i];