SWDEV-508869 - Fix Linux build error for HIP on PAL (#176)

linqun · web-flow · commit 3b44884a57b6 · 2025-06-27T07:51:22.000+08:00
[ROCm/clr commit: 9699cc3]
diff --git a/projects/clr/rocclr/device/device.hpp b/projects/clr/rocclr/device/device.hpp
@@ -981,7 +981,7 @@ class Memory : public amd::HeapObject {
   //! not a physical map. When a memory object does not use USE_HOST_PTR we
   //! can use a remote resource and DMA, avoiding the additional CPU memcpy.
   amd::Memory* mapMemory_;            //!< Memory used as map target buffer
-  volatile size_t indirectMapCount_;  //!< Number of maps
+  std::atomic<size_t> indirectMapCount_;  //!< Number of maps
   std::unordered_map<const void*, WriteMapInfo>
       writeMapInfo_;  //!< Saved write map info for partial unmap
 
diff --git a/projects/clr/rocclr/device/pal/paldevice.cpp b/projects/clr/rocclr/device/pal/paldevice.cpp
@@ -426,7 +426,7 @@ void NullDevice::fillDeviceInfo(const Pal::DeviceProperties& palProp,
                                       uint64_t(heaps[Pal::GpuHeapInvisible].physicalSize));
   }
 
-#if defined(ATI_OS_WIN)
+#if IS_WINDOWS
   if (settings().apuSystem_) {
     info_.maxMemAllocSize_ = std::max(
         (static_cast<uint64_t>(heaps[Pal::GpuHeapGartUswc].logicalSize) * uswcPercentAvailable) /
@@ -673,8 +673,10 @@ void NullDevice::fillDeviceInfo(const Pal::DeviceProperties& palProp,
   info_.vgprsPerSimd_ = palProp.gfxipProperties.shaderCore.vgprsPerSimd;
   info_.sgprsPerSimd_ = palProp.gfxipProperties.shaderCore.sgprsPerSimd;
   info_.availableRegistersPerCU_ = info_.vgprsPerSimd_ * info_.simdPerCU_ * 32;
+#if IS_WINDOWS
   info_.luidLowPart_ = palProp.osProperties.luidLowPart;
   info_.luidHighPart_ = palProp.osProperties.luidHighPart;
+#endif
   // Setup the node mask for MGPU only case from the original PAL list of all devices
   if ((gNumDevices > 1) && (pal_device != nullptr)) {
     for (uint32_t i = 0; i < gNumDevices; ++i) {
@@ -2058,6 +2060,7 @@ bool Device::globalFreeMemory(size_t* freeMemory) const {
   Pal::gpusize system_memory = allocedMem[Pal::GpuHeapGartCacheable] +
     allocedMem[Pal::GpuHeapGartUswc] + cache_group_local - resourceCache().cacheSize();
 
+#if IS_WINDOWS
   // Second, query OS for overall memory usage on the system
 
   if (properties().osProperties.supportMemoryBudgetQuery) {
@@ -2086,7 +2089,7 @@ bool Device::globalFreeMemory(size_t* freeMemory) const {
       system_memory = system_total_alloced;
     }
   }
-
+#endif
   // Third, finalize reported free memory
 
   // Fill free memory info
@@ -2799,12 +2802,12 @@ bool Device::createBlitProgram() {
         if (asm_program->load()) {
           trap_handler_ = asm_program;
         } else {
-          DevLogPrintfError("Could not load the trap handler \n");
+          DevLogError("Could not load the trap handler \n");
           asm_program->release();
         }
       }
     } else {
-      DevLogPrintfError("Trap handler creation failed\n");
+      DevLogError("Trap handler creation failed\n");
     }
   }
 
diff --git a/projects/clr/rocclr/device/pal/paldevicegl.cpp b/projects/clr/rocclr/device/pal/paldevicegl.cpp
@@ -65,6 +65,15 @@ static PFNGlxGetProcAddress pfnGlxGetProcAddress = nullptr;
 typedef int(APIENTRYP PFNMesaGLInteropGLXQueryDeviceInfo)(Display* dpy, GLXContext context,
                                                           mesa_glinterop_device_info* out);
 static PFNMesaGLInteropGLXQueryDeviceInfo pfnMesaGLInteropGLXQueryDeviceInfo = nullptr;
+typedef Bool(APIENTRYP PFNGLXBEGINCLINTEROPAMD)(GLXContext context, GLuint flags);
+typedef Bool(APIENTRYP PFNGLXENDCLINTEROPAMD)(GLXContext context, GLuint flags);
+typedef Bool(APIENTRYP PFNGLXRESOURCEATTACHAMD)(GLXContext context, GLvoid* resource,
+                                                GLvoid* pResourceData);
+typedef Bool(APIENTRYP PFNGLXRESOURCEDETACHAMD)(GLXContext context, GLvoid* resource);
+typedef Bool(APIENTRYP PFNGLXRESOURCEDETACHAMD)(GLXContext context, GLvoid* resource);
+typedef Bool(APIENTRYP PFNGLXRESOURCEDETACHAMD)(GLXContext context, GLvoid* resource);
+typedef Bool(APIENTRYP PFNGLXGETCONTEXTMVPUINFOAMD)(GLXContext context, GLuint* deviceId,
+                                                    GLuint* chainMask);
 static PFNGLXBEGINCLINTEROPAMD glXBeginCLInteropAMD = nullptr;
 static PFNGLXENDCLINTEROPAMD glXEndCLInteropAMD = nullptr;
 static PFNGLXRESOURCEATTACHAMD glXResourceAttachAMD = nullptr;
@@ -90,8 +99,8 @@ static PFNWGLRESOURCEDETACHAMD wglResourceAcquireAMD = nullptr;
 static PFNWGLRESOURCEDETACHAMD wglResourceReleaseAMD = nullptr;
 static PFNWGLRESOURCEDETACHAMD wglResourceDetachAMD = nullptr;
 static PFNWGLGETCONTEXTGPUINFOAMD wglGetContextGPUInfoAMD = nullptr;
-bool gGlFuncInit = false;
 #endif
+bool gGlFuncInit = false;
 
 namespace amd::pal {
 
diff --git a/projects/clr/rocclr/device/pal/palmemory.cpp b/projects/clr/rocclr/device/pal/palmemory.cpp
@@ -1031,8 +1031,10 @@ Memory* Buffer::createBufferView(amd::Memory& subBufferOwner) const {
 // ================================================================================================
 bool Buffer::ExportHandle(void* handle) const {
   Pal::GpuMemoryExportInfo exportInfo = {};
+#if IS_WINDOWS
   // Set default flags in case they are not provided by application
   exportInfo.accessFlags = GENERIC_READ | GENERIC_WRITE;
+#endif
   *reinterpret_cast<Pal::OsExternalHandle*>(handle) = iMem()->ExportExternalHandle(exportInfo);
   return true;
 }
diff --git a/projects/clr/rocclr/device/pal/palresource.cpp b/projects/clr/rocclr/device/pal/palresource.cpp
@@ -792,7 +792,7 @@ bool Resource::CreateInterop(CreateParams* params) {
 
     if (!dev().resGLAssociate(oglRes->glPlatformContext_, oglRes->handle_, glType_,
                               &openInfo.hExternalResource, &glInteropMbRes_, &offset_, desc_.format_
-#ifdef ATI_OS_WIN
+#if IS_WINDOWS
                               ,
                               openInfo.doppDesktopInfo
 #endif
@@ -808,6 +808,7 @@ bool Resource::CreateInterop(CreateParams* params) {
     if (vparams->handle_) {
       openInfo.hExternalResource = vparams->handle_;
     } else if (vparams->name_) {
+#if IS_WINDOWS
       Pal::ExternalHandleInfo eHandleInfo = {};
       eHandleInfo.objectType = Pal::ExternalObjectType::Allocation;
       eHandleInfo.pNtObjectName = reinterpret_cast<const wchar_t*>(vparams->name_);
@@ -819,10 +820,13 @@ bool Resource::CreateInterop(CreateParams* params) {
           dev().iDev()->OpenExternalHandleFromName(eHandleInfo, &openInfo.hExternalResource)) {
         return false;
       }
+#else
+      return false;
+#endif
     }
     openInfo.flags.ntHandle = vparams->nt_handle_;
   }
-#ifdef ATI_OS_WIN
+#if IS_WINDOWS
   else {
     D3DInteropParams* d3dRes = reinterpret_cast<D3DInteropParams*>(params);
     openInfo.hExternalResource = d3dRes->handle_;
@@ -883,7 +887,7 @@ bool Resource::CreateInterop(CreateParams* params) {
       switch (misc) {
         case 1:  // NV12 or P010 formats
           switch (layer) {
-            case -1:
+            case std::numeric_limits<uint>::max():
             case 0:
               break;
             case 1:
@@ -899,7 +903,7 @@ bool Resource::CreateInterop(CreateParams* params) {
           break;
         case 2:  // YV12 format
           switch (layer) {
-            case -1:
+            case std::numeric_limits<uint>::max():
             case 0:
               break;
             case 1:
diff --git a/projects/clr/rocclr/device/pal/palvirtual.hpp b/projects/clr/rocclr/device/pal/palvirtual.hpp
@@ -21,6 +21,7 @@
 #pragma once
 
 #include <queue>
+#include <inttypes.h>
 #include "device/pal/paldefs.hpp"
 #include "device/pal/palconstbuf.hpp"
 #include "device/pal/palprintf.hpp"
@@ -734,7 +735,7 @@ class VirtualGPU : public device::VirtualDevice {
 inline void VirtualGPU::logVmMemory(const std::string name, const Memory* memory) {
   if (PAL_EMBED_KERNEL_MD || (AMD_LOG_LEVEL >= amd::LOG_INFO)) {
     char buf[256];
-    sprintf(buf, "%s = ptr:[%p-%p] size:[%llu] heap[%d]", name.c_str(),
+    sprintf(buf, "%s = ptr:[%p-%p] size:[%" PRIu64 "] heap[%d]", name.c_str(),
             reinterpret_cast<void*>(memory->vmAddress()),
             reinterpret_cast<void*>(memory->vmAddress() + memory->size()),
             memory->iMem()->Desc().size, memory->iMem()->Desc().heaps[0]);
diff --git a/projects/clr/rocclr/include/GL/gl_interop.h b/projects/clr/rocclr/include/GL/gl_interop.h
@@ -21,7 +21,6 @@
 #ifndef GL_INTEROP_H_
 #define GL_INTEROP_H_
 
-#ifdef _WIN32
 #define GL_RESOURCE_ATTACH_TEXTURE_AMD                                  0x12a000
 #define GL_RESOURCE_ATTACH_FRAMEBUFFER_AMD                              0x12a001
 #define GL_RESOURCE_ATTACH_RENDERBUFFER_AMD                             0x12a002
@@ -88,6 +87,7 @@ typedef struct GLResourceDataRec {
 
 } GLResourceData;
 
+#ifdef _WIN32
 typedef BOOL (WINAPI* PFNWGLBEGINCLINTEROPAMD)(HGLRC hglrc, GLuint flags);
 typedef BOOL (WINAPI* PFNWGLENDCLINTEROPAMD) (HGLRC hglrc,  GLuint flags);
 typedef BOOL (WINAPI* PFNWGLRESOURCEATTACHAMD) (HGLRC hglrc, GLvoid* resource, GLvoid* pResourceData);
diff --git a/projects/clr/rocclr/utils/debug.hpp b/projects/clr/rocclr/utils/debug.hpp
@@ -237,6 +237,7 @@ inline void warning(const char* msg) { amd::report_warning(msg); }
 #else /*CL_LOG*/
 #define ClPrint(level, mask, format, ...) (void)(0)
 #define ClCondPrint(level, mask, condition, format, ...) (void)(0)
+#define HIPPrintDuration(level, mask, startTimeUs, format, ...) (void)(0)
 #endif /*CL_LOG*/
 
 #define ClTrace(level, mask) ClPrint(level, mask, "%s", __func__)
@@ -245,13 +246,13 @@ inline void warning(const char* msg) { amd::report_warning(msg); }
 #define LogError(msg) ClPrint(amd::LOG_ERROR, amd::LOG_ALWAYS, msg)
 #define LogWarning(msg) ClPrint(amd::LOG_WARNING, amd::LOG_ALWAYS, msg)
 
-#define LogPrintfDebug(format, ...) ClPrint(amd::LOG_DEBUG, amd::LOG_ALWAYS, format, __VA_ARGS__)
-#define LogPrintfError(format, ...) ClPrint(amd::LOG_ERROR, amd::LOG_ALWAYS, format, __VA_ARGS__)
-#define LogPrintfWarning(format, ...) ClPrint(amd::LOG_WARNING, amd::LOG_ALWAYS, format, __VA_ARGS__)
-#define LogPrintfInfo(format, ...) ClPrint(amd::LOG_INFO, amd::LOG_ALWAYS, format, __VA_ARGS__)
+#define LogPrintfDebug(format, ...) ClPrint(amd::LOG_DEBUG, amd::LOG_ALWAYS, format, ##__VA_ARGS__)
+#define LogPrintfError(format, ...) ClPrint(amd::LOG_ERROR, amd::LOG_ALWAYS, format, ##__VA_ARGS__)
+#define LogPrintfWarning(format, ...) ClPrint(amd::LOG_WARNING, amd::LOG_ALWAYS, format, ##__VA_ARGS__)
+#define LogPrintfInfo(format, ...) ClPrint(amd::LOG_INFO, amd::LOG_ALWAYS, format, ##__VA_ARGS__)
 
 #if (defined(DEBUG) || defined(DEV_LOG_ENABLE))
-  #define DevLogPrintfError(format, ...) LogPrintfError(format, __VA_ARGS__)
+  #define DevLogPrintfError(format, ...)  LogPrintfError(format, ##__VA_ARGS__)
   #define DevLogError(msg) LogError(msg)
 #else
   #define DevLogPrintfError(format, ...)
