P4 to Git Change 1094113 by gandryey@gera-dev-w7 on 2014/11/05 18:22:34
ECR #304775 - CL-GL depth buffer interop - Make the changes necessary for the depth buffer interop support. Currently the extension is disabled, because the conformance tests have multiple bugs. Affected files ... ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_gl.cpp#45 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_memobj.cpp#71 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/device.hpp#231 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuresource.cpp#197 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpusettings.cpp#292 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gslbe/src/rt/GSLContext.cpp#66 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gslbe/src/rt/GSLDevice.cpp#101 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gslbe/src/rt/GSLDeviceGL.cpp#20 edit ... //depot/stg/opencl/drivers/opencl/runtime/platform/memory.cpp#113 edit ... //depot/stg/opencl/drivers/opencl/runtime/utils/flags.hpp#217 edit
Этот коммит содержится в:
@@ -584,13 +584,12 @@ public:
|
||||
uint supportRA_: 1; //!< Support RA channel order format
|
||||
uint largeHostMemAlloc_: 1; //!< Allow large host mem allocations (> maxSingleAlloc)
|
||||
uint waitCommand_: 1; //!< Enables a wait for every submitted command
|
||||
uint depthMSAAInterop_: 1; //!< Enable depth-stencil and MSAA interop CL-GL inteorp
|
||||
uint customHostAllocator_: 1;//!< True if device has custom host allocator
|
||||
// that replaces generic OS allocation routines
|
||||
uint customSvmAllocator_: 1; //!< True if device has custom SVM allocator
|
||||
uint supportDepthsRGB_: 1; //!< Support DEPTH and sRGB channel order format
|
||||
uint assumeAliases_: 1; //!< Assume aliases in the compilation process
|
||||
uint reserved_: 23;
|
||||
uint reserved_: 24;
|
||||
};
|
||||
uint value_;
|
||||
};
|
||||
|
||||
@@ -243,8 +243,12 @@ static uint32_t GetHSAILImageFormatType(cmSurfFmt format)
|
||||
case CM_SURF_FMT_RG32F:
|
||||
case CM_SURF_FMT_RGBA32F:
|
||||
case CM_SURF_FMT_DEPTH32F:
|
||||
case CM_SURF_FMT_DEPTH32F_X24_STEN8:
|
||||
formatType = HSA_EXT_IMAGE_CHANNEL_TYPE_FLOAT;
|
||||
break;
|
||||
case CM_SURF_FMT_DEPTH24_STEN8:
|
||||
formatType = HSA_EXT_IMAGE_CHANNEL_TYPE_UNORM_INT24;
|
||||
break;
|
||||
default:
|
||||
assert(false);
|
||||
}
|
||||
@@ -252,7 +256,7 @@ static uint32_t GetHSAILImageFormatType(cmSurfFmt format)
|
||||
return formatType;
|
||||
}
|
||||
|
||||
static uint32_t GetHSAILImageOrderType(gslChannelOrder chOrder)
|
||||
static uint32_t GetHSAILImageOrderType(gslChannelOrder chOrder, cmSurfFmt format)
|
||||
{
|
||||
uint32_t orderType = HSA_EXT_IMAGE_CHANNEL_ORDER_A;
|
||||
|
||||
@@ -301,7 +305,13 @@ static uint32_t GetHSAILImageOrderType(gslChannelOrder chOrder)
|
||||
orderType = HSA_EXT_IMAGE_CHANNEL_ORDER_LUMINANCE;
|
||||
break;
|
||||
case GSL_CHANNEL_ORDER_REPLICATE_R:
|
||||
orderType = HSA_EXT_IMAGE_CHANNEL_ORDER_DEPTH;
|
||||
if ((format == CM_SURF_FMT_DEPTH32F_X24_STEN8) ||
|
||||
(format == CM_SURF_FMT_DEPTH24_STEN8)) {
|
||||
orderType = HSA_EXT_IMAGE_CHANNEL_ORDER_DEPTH_STENCIL;
|
||||
}
|
||||
else {
|
||||
orderType = HSA_EXT_IMAGE_CHANNEL_ORDER_DEPTH;
|
||||
}
|
||||
break;
|
||||
default:
|
||||
assert(false);
|
||||
@@ -976,7 +986,7 @@ Resource::create(MemoryType memType, CreateParams* params, bool heap)
|
||||
}
|
||||
dev().fillImageHwState(gslResource, hwState_, 8 * sizeof(uint32_t));
|
||||
hwState_[8] = GetHSAILImageFormatType(cal()->format_);
|
||||
hwState_[9] = GetHSAILImageOrderType(cal()->channelOrder_);
|
||||
hwState_[9] = GetHSAILImageOrderType(cal()->channelOrder_, cal()->format_);
|
||||
hwState_[10] = static_cast<uint32_t>(cal()->width_);
|
||||
// Workaround for depth view, change tileIndex to 0 for depth view
|
||||
if ((memoryType() == ImageView) &&
|
||||
|
||||
@@ -345,9 +345,8 @@ Settings::create(
|
||||
|
||||
rectLinearDMA_ = true;
|
||||
|
||||
if (AMD_DEPTH_MSAA_INTEROP) {
|
||||
enableExtension(ClKhrGLDepthImages);
|
||||
depthMSAAInterop_ = true;
|
||||
if (oclVersion_ >= OpenCL20) {
|
||||
//enableExtension(ClKhrGLDepthImages);
|
||||
}
|
||||
if (AMD_THREAD_TRACE_ENABLE) {
|
||||
threadTraceEnable_ = true;
|
||||
@@ -445,7 +444,7 @@ Settings::create(
|
||||
|
||||
if (apuSystem_ &&
|
||||
((calAttr.totalVisibleHeap + calAttr.totalInvisibleHeap) < 150)) {
|
||||
remoteAlloc_ = true;
|
||||
remoteAlloc_ = true;
|
||||
}
|
||||
|
||||
// Override current device settings
|
||||
|
||||
@@ -300,6 +300,7 @@ CALGSLContext::setInput(uint32 physUnit, gslMemObject mem)
|
||||
m_textureResources[physUnit], physUnit);
|
||||
}
|
||||
|
||||
m_textureResources[physUnit]->updateDepthTextureParam(mem);
|
||||
m_textureResources[physUnit]->setMemObject(m_cs, GSL_COMPUTE_PROGRAM, mem);
|
||||
|
||||
if (mem != NULL)
|
||||
|
||||
@@ -1701,6 +1701,7 @@ CALGSLDevice::fillImageHwState(gslMemObject mem, void* hwState, uint32 hwStateSi
|
||||
amd::ScopedLock k(gslDeviceOps());
|
||||
intp channelOrder = mem->getAttribs().channelOrder;
|
||||
convertInputChannelOrder(&channelOrder);
|
||||
m_textureResource->updateDepthTextureParam(mem);
|
||||
m_textureResource->getTextureSrd(m_cs, mem, reinterpret_cast<const char*>(&channelOrder),
|
||||
hwState, hwStateSize);
|
||||
}
|
||||
|
||||
@@ -82,8 +82,8 @@ static const cmFormatXlateParams cmFormatXlateTable [] = {
|
||||
{CM_SURF_FMT_DEPTH24F_X8, (cmSurfFmt)500, GSL_CHANNEL_ORDER_R},
|
||||
{CM_SURF_FMT_DEPTH24_STEN8, CM_SURF_FMT_DEPTH24_STEN8, GSL_CHANNEL_ORDER_REPLICATE_R},
|
||||
{CM_SURF_FMT_DEPTH24F_STEN8, (cmSurfFmt)500, GSL_CHANNEL_ORDER_R},
|
||||
{CM_SURF_FMT_DEPTH32F_X24_STEN8, CM_SURF_FMT_RG32I, GSL_CHANNEL_ORDER_REPLICATE_R},
|
||||
{CM_SURF_FMT_DEPTH32F, CM_SURF_FMT_R32F, GSL_CHANNEL_ORDER_REPLICATE_R},
|
||||
{CM_SURF_FMT_DEPTH32F_X24_STEN8, CM_SURF_FMT_DEPTH32F_X24_STEN8, GSL_CHANNEL_ORDER_REPLICATE_R},
|
||||
{CM_SURF_FMT_DEPTH32F, CM_SURF_FMT_DEPTH32F, GSL_CHANNEL_ORDER_REPLICATE_R},
|
||||
{CM_SURF_FMT_sR11_sG11_sB10, (cmSurfFmt)500, GSL_CHANNEL_ORDER_R},
|
||||
{CM_SURF_FMT_sU16, CM_SURF_FMT_sU16, GSL_CHANNEL_ORDER_R},
|
||||
{CM_SURF_FMT_sUV16, CM_SURF_FMT_sUV16, GSL_CHANNEL_ORDER_RG},
|
||||
|
||||
@@ -1102,11 +1102,11 @@ Image::getSupportedFormats(
|
||||
bool
|
||||
Image::Format::isSupported(const Context& context, cl_mem_object_type image_type) const
|
||||
{
|
||||
bool supportDepthMSAA = true;
|
||||
bool supportDepth = true;
|
||||
const std::vector<amd::Device*>& devices = context.devices();
|
||||
for (size_t i = 0; i < devices.size(); i++) {
|
||||
if (!devices[i]->settings().depthMSAAInterop_) {
|
||||
supportDepthMSAA = false;
|
||||
if (!devices[i]->settings().checkExtension(ClKhrGLDepthImages)) {
|
||||
supportDepth = false;
|
||||
}
|
||||
}
|
||||
|
||||
@@ -1129,7 +1129,7 @@ Image::Format::isSupported(const Context& context, cl_mem_object_type image_type
|
||||
|
||||
delete image_formats;
|
||||
|
||||
if (supportDepthMSAA) {
|
||||
if (supportDepth) {
|
||||
for (cl_uint i = 0; i < sizeof(depthFormats) / sizeof(cl_image_format); i++) {
|
||||
if (*this == depthFormats[i]) {
|
||||
return true;
|
||||
|
||||
@@ -152,8 +152,6 @@ release(bool, GPU_USE_DEVICE_QUEUE, false, \
|
||||
"Use a dedicated device queue for the actual submissions") \
|
||||
release(bool, GPU_ENABLE_LARGE_ALLOCATION, false, \
|
||||
"Enable >4GB single allocations") \
|
||||
release(bool, AMD_DEPTH_MSAA_INTEROP, false, \
|
||||
"Enable depth stencil and MSAA buffer interop") \
|
||||
release(bool, AMD_THREAD_TRACE_ENABLE, false, \
|
||||
"Enable thread trace extension") \
|
||||
release(uint, OPENCL_VERSION, 200, \
|
||||
|
||||
Ссылка в новой задаче
Block a user