SWDEV-558845 - Support image in rocr on Windows (#1582)

Enable image build in Windows. Remove some useless codes that fail building in Windows. Some minor improvement. Temporarily exclude mipmap test files. Prevent negative tests affect some tests. Move some catch info log codes into failed cases.
2025-11-05 09:33:41 -05:00
@@ -261,15 +261,10 @@ if (${PC_SAMPLING_SUPPORT})
  target_sources( ${CORE_RUNTIME_TARGET} PRIVATE ${PCS_SRCS} )
 endif()

-if (UNIX)
-  if ( NOT DEFINED IMAGE_SUPPORT AND CMAKE_SYSTEM_PROCESSOR MATCHES "i?86|x86_64|amd64|AMD64|loongarch64" )
-    set ( IMAGE_SUPPORT ON )
-  endif()
-set ( IMAGE_SUPPORT ${IMAGE_SUPPORT} CACHE BOOL "Build with image support (default: ON for x86, OFF elsewise)." )
-else()
-  # Force IMAGE_SUPPORT to be OFF
-  set(IMAGE_SUPPORT OFF CACHE BOOL "Build with image support (forced to OFF)" FORCE)
+if (NOT DEFINED IMAGE_SUPPORT AND CMAKE_SYSTEM_PROCESSOR MATCHES "i?86|x86_64|amd64|AMD64|loongarch64")
+  set (IMAGE_SUPPORT ON)
 endif()
+set (IMAGE_SUPPORT ${IMAGE_SUPPORT} CACHE BOOL "Build with image support (default: ON for x86, OFF elsewise).")

 ## Optional image module defintions.
 if(${IMAGE_SUPPORT})
@@ -289,7 +289,9 @@ void SurfaceGpuList(std::vector<int32_t>& gpu_list, bool xnack_mode, bool enable
        core::g_use_interrupt_wait = false;

      if (core::Runtime::runtime_singleton_->thunkLoader()->IsDXG()) {
+        core::Runtime::runtime_singleton_->flag().disable_image(true);
 #if defined(_WIN32)
+        core::Runtime::runtime_singleton_->flag().disable_image(false);
        if (node_prop.Capability2.ui32.AqlEmulationPm4_)
 #endif
        {
@@ -297,7 +299,6 @@ void SurfaceGpuList(std::vector<int32_t>& gpu_list, bool xnack_mode, bool enable
          core::Runtime::runtime_singleton_->flag().disable_scratch();
        }
        core::Runtime::runtime_singleton_->flag().set_sdma(false, false);
-        core::Runtime::runtime_singleton_->flag().disable_image(true);
        core::Runtime::runtime_singleton_->flag().disable_xnack();
        core::Runtime::runtime_singleton_->flag().disable_fine_grain_pcie();
        core::Runtime::runtime_singleton_->flag().set_ipc_mode_legacy(false);
@@ -41,11 +41,6 @@
 ////////////////////////////////////////////////////////////////////////////////

 #include "blit_kernel.h"
-
-#if (defined(WIN32) || defined(_WIN32))
-#define NOMINMAX
-#endif
-
 #include <algorithm>
 #include <atomic>
 #include <sstream>
@@ -67,10 +67,12 @@ endif()
 ##  Add custom command to generate a kernel code object file
 ##==========================================
 function(gen_kernel_bc TARGET_ID INPUT_FILE OUTPUT_FILE)
-
+  if(WIN32)
+    set (WIN_CL_COMPILE_OPTION "-fshort-wchar")
+  endif()
  separate_arguments(CLANG_ARG_LIST UNIX_COMMAND
    "-O2 -x cl -Xclang -finclude-default-header -cl-denorms-are-zero -cl-std=CL2.0
-    -target amdgcn-amd-amdhsa -mcpu=${TARGET_ID} -mcode-object-version=4
+    -target amdgcn-amd-amdhsa -mcpu=${TARGET_ID} -mcode-object-version=4 ${WIN_CL_COMPILE_OPTION}
    -o ${OUTPUT_FILE} ${INPUT_FILE}")

  ## Add custom command to produce a code object file.
@@ -128,9 +130,18 @@ function(generate_blit_file BFILE)

  ## Add a custom command that generates opencl_blit_objects.cpp
  ## This depends on all the generated code object files and the C++ generator script.
-  add_custom_command(OUTPUT ${BFILE}.cpp
+  if (UNIX)
+    add_custom_command(OUTPUT ${BFILE}.cpp
                     COMMAND ${CMAKE_CURRENT_SOURCE_DIR}/create_hsaco_ascii_file.sh ${CMAKE_CURRENT_BINARY_DIR}/${BFILE}.cpp
                     DEPENDS ${HSACO_TARG_LIST} create_hsaco_ascii_file.sh )
+  else()
+    find_package(Python3 COMPONENTS Interpreter REQUIRED)
+    add_custom_command(
+      OUTPUT ${BFILE}.cpp
+      COMMAND ${Python3_EXECUTABLE} ${CMAKE_CURRENT_SOURCE_DIR}/create_hsaco_ascii_file.py ${CMAKE_CURRENT_BINARY_DIR}/${BFILE}.cpp ${HSACO_TARG_LIST}
+      COMMENT "Collating blit shaders..."
+      DEPENDS ${HSACO_TARG_LIST} create_hsaco_ascii_file.py)
+  endif()

  ## Export a target that builds (and depends on) opencl_blit_objects.cpp
  add_custom_target( ${BFILE} DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/${BFILE}.cpp )
@@ -0,0 +1,71 @@
+################################################################################
+##
+## Copyright (c) Advanced Micro Devices, Inc., or its affiliates. 
+## 
+## SPDX-License-Identifier: MIT
+##
+################################################################################
+import sys
+
+def GetSize(fileobject):
+    fileobject.seek(0,2) # move the cursor to the end of the file
+    size = fileobject.tell()
+    return size
+
+def DumpFile(header, input_name):
+  try:
+    with open(input_name, "rb") as binary_file:
+        # Read the entire content of the file as bytes
+        binary_data = binary_file.read()
+        file_size = GetSize(binary_file)
+        #print(f"Binary size: {file_size}")
+        # Reset file pointer
+        binary_file.seek(0)
+        parts = input_name.split('.')
+        file_name = parts[0]
+        content = f"unsigned char {file_name}""[] = {\n  "
+
+        header.write(content)
+        line = 0
+        count = 0
+        for byte_value in binary_data:
+          count += 1
+          padded_hex = '{:02x}'.format(byte_value)
+          if (count != file_size):
+            header.write(f"0x{padded_hex},")
+          else:
+            header.write(f"0x{padded_hex}")
+          line += 1
+          if (line == 12):
+            header.write(f"\n  ")
+            line = 0
+          else:
+            header.write(f" ")
+
+        header.write("\n};\nunsigned int "f"{file_name}_len = {file_size};\n")
+
+  except FileNotFoundError:
+    print(f"Error: The file {input_name} was not found.")
+  except Exception as e:
+    print(f"An error occurred: {e}")
+
+
+if len(sys.argv) > 1:
+  header_name = sys.argv[1];
+  with open(header_name, 'w') as header:
+    header.write("//==============================================================================\n")
+    header.write("//  This file is automatically generated during build process, don't modify it\n")
+    header.write("//==============================================================================\n\n")
+    header.write("namespace rocr {\n")
+    header.write("namespace image {\n\n")
+
+    for i, arg in enumerate(sys.argv):
+      if (i > 1):
+        #print(f"File {i}: {arg}\n")
+        DumpFile(header, arg)
+    header.write("} // namespace image\n")
+    header.write("} // namespace rocr\n\n")
+
+else:
+  print("Empty arguments!")
+  
@@ -52,11 +52,6 @@
 #include <climits>
 #include <cmath>

-#if (defined(WIN32) || defined(_WIN32))
-#define NOMINMAX
-__inline long int lrintf(float f) { return _mm_cvtss_si32(_mm_load_ss(&f)); }
-#endif
-
 namespace rocr {
 namespace image {

@@ -40,7 +40,6 @@
 //
 ////////////////////////////////////////////////////////////////////////////////

-#define NOMINMAX
 #include "image_manager_ai.h"

 #include <assert.h>
@@ -40,7 +40,6 @@
 //
 ////////////////////////////////////////////////////////////////////////////////

-#define NOMINMAX
 #include "image_manager_gfx11.h"

 #include <assert.h>
@@ -40,7 +40,6 @@
 //
 ////////////////////////////////////////////////////////////////////////////////

-#define NOMINMAX
 #include "image_manager_gfx12.h"

 #include <assert.h>
@@ -40,7 +40,6 @@
 //
 ////////////////////////////////////////////////////////////////////////////////

-#define NOMINMAX
 #include "image_manager_kv.h"

 #include <assert.h>
@@ -100,8 +99,8 @@ hsa_status_t ImageManagerKv::Initialize(hsa_agent_t agent_handle) {
  assert(status == HSA_STATUS_SUCCESS);

  HsaGpuTileConfig tileConfig = {0};
-  unsigned int tc[40];
-  unsigned int mtc[40];
+  unsigned int tc[40] = {0};
+  unsigned int mtc[40] = {0};
  tileConfig.TileConfig = &tc[0];
  tileConfig.NumTileConfigs = 40;
  tileConfig.MacroTileConfig = &mtc[0];
@@ -40,7 +40,6 @@
 //
 ////////////////////////////////////////////////////////////////////////////////

-#define NOMINMAX
 #include "image_manager_nv.h"

 #include <assert.h>
@@ -40,7 +40,6 @@
 //
 ////////////////////////////////////////////////////////////////////////////////

-#define NOMINMAX
 #include "image_runtime.h"

 #include <assert.h>
@@ -280,22 +280,6 @@ static __forceinline uint64_t NextPow2(uint64_t value) {

 static __forceinline bool strIsEmpty(const char* str) noexcept { return str[0] == '\0'; }

-static __forceinline std::string& ltrim(std::string& s) {
-  auto it = std::find_if(s.begin(), s.end(),
-                         [](char c) { return !std::isspace<char>(c, std::locale::classic()); });
-  s.erase(s.begin(), it);
-  return s;
-}
-
-static __forceinline std::string& rtrim(std::string& s) {
-  auto it = std::find_if(s.rbegin(), s.rend(),
-                         [](char c) { return !std::isspace<char>(c, std::locale::classic()); });
-  s.erase(it.base(), s.end());
-  return s;
-}
-
-static __forceinline std::string& trim(std::string& s) { return ltrim(rtrim(s)); }
-
 template<uint32_t lowBit, uint32_t highBit, typename T>
 static __forceinline uint32_t BitSelect(T p) {
  static_assert(sizeof(T) <= sizeof(uintptr_t), "Type out of range.");