P4 to Git Change 1159925 by rayxiao@alit_opencl_rayxiao on 2015/06/10 13:33:10

EPR #396242 - Add T_PAD type for padding to parameters, use struct instead of array for padding. Set name of padding struct as "__paddingFields" when created.

Affected files ...

... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_kernel.h#23 edit
... //depot/stg/opencl/drivers/opencl/compiler/clc/src/e2lType.cpp#53 edit
... //depot/stg/opencl/drivers/opencl/compiler/clc/src/oclMetallvm.cpp#46 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/cpu/cpumapping.cpp#3 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/cpu/cpuprogram.cpp#67 edit


[ROCm/clr commit: 5b6d359824]

projects/clr/rocclr/runtime/device/cpu/cpumapping.cpp

@@ -235,6 +235,38 @@ namespace cpu {
                 }
             }
         }
+        else if (desc[internal_index].type == T_PAD) {
+            //Struct has padding
+            hc_offset = init_offset;
+            if (desc[index_out+1].type != T_VOID) {
+                index_out++;
+                internal_index = index_out;
+                internal_field_map = new HCtoDCmap(desc, 0, internal_index, init_offset);
+                hc_size = internal_field_map->compute_map(desc, hc_alignment, dc_alignment, next_offset, inStruct, index_out);
+                // Adjust alignment to biggest member alignment
+                hc_alignment = 1;
+                dc_alignment = 1;
+                unsigned pad_size = hc_size;
+                internal_index = index_out;
+                if (desc[index_out+1].type != T_VOID) {
+                    //Still inside padding and not done
+                    index_out++;
+                    internal_index = index_out;
+                    next_field_map = new HCtoDCmap(desc, 0, internal_index, next_offset);
+                    pad_size = hc_size;
+                    pad_size += next_field_map->compute_map(desc, outer_hc_alignment, outer_dc_alignment, next_offset, inStruct, index_out);
+                    next_offset = max(next_field_map->hc_offset+next_field_map->hc_size, next_field_map->hc_offset+hc_alignment);
+                    // running count of padding dc_size = hc_size + size of next member
+                    return pad_size;
+                }
+                else {
+                    //Moving out of struct, go to next index
+                    index_out++;
+                    internal_index = index_out;
+                    return hc_size; //return last padding member size
+                }
+            }
+        }
         else {
             //Scalar parameter
             hc_offset = init_offset;
@@ -278,7 +310,7 @@ namespace cpu {
         }
         else {
             // Ignore alignment when a char occurs to account for padding
-            if (type != T_STRUCT && next_field_map->hc_size == 1 && map_param_size > 1 && inStruct_flag > 0) {
+            if (type == T_PAD) {
                 next_field_map->dc_offset = dc_offset + dc_size;
                 next_offset = current_offset + hc_size;
             }

projects/clr/rocclr/runtime/device/cpu/cpuprogram.cpp

@@ -172,7 +172,7 @@ getParamSizeImpl(bool cpuLayer, const clk_parameter_descriptor_t* desc,
                  size_t* alignment, unsigned* index_out)
 {
     size_t size = 0;
-    if(desc[index].type == T_STRUCT) {
+    if(desc[index].type == T_STRUCT || desc[index].type == T_PAD) {
         size_t maxAlignment = 0;
         size_t structSize = 0;
         size_t structAlignment = 0;
@@ -237,7 +237,7 @@ getNumTypeDescs(const clk_parameter_descriptor_t* desc)
   int numStruct = 0;
   unsigned i;
     for(i = 0; desc[i].type != T_VOID || numStruct > 0; ++i) {
-        if (desc[i].type == T_STRUCT)
+        if (desc[i].type == T_STRUCT || desc[i].type == T_PAD)
             numStruct++;
         if (desc[i].type == T_VOID)
             numStruct--;