Files
compute-runtime/opencl/source/sharings/gl/linux/gl_texture_linux.cpp
Sylvain Munaut e2c511bc00 feature(ocl): Improve Linux CL/GL sharing support
This commit is aimed at drastically improving the support for the CL/GL
sharing
extension on linux. The current support is not really usable as it only
supports a few texture format, and only on EGL contexts. It is also
pretty
buggy since it requires the texture to be bound when placing the CL call
to
share it which is just plain wrong and will not work in many
applications.
This new version makes used of the "official" interop extension from
MESA
which is available for GLX and EGL contexts, allows sharing of buffers
and
not just texture and supports many more formats.
This is still far from being a fully compliant / full featured version
of
the extension, but it's a big step forward in my opinion and allows to
run
some real applications.
I've tested gr-fosphor (SDR spectrum display) and Davinci Resolve as
examples.
Both of theses don't work without theses improvements.

Fixes: https://github.com/intel/compute-runtime/issues/659
Fixes: https://github.com/intel/compute-runtime/issues/667

https://github.com/intel/compute-runtime/pull/673

Related-To: NEO-3599

Signed-off-by: Sylvain Munaut <tnt@246tNt.com>
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
2024-03-01 13:38:56 +01:00

347 lines
13 KiB
C++

/*
* Copyright (C) 2023-2024 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
*/
#include "shared/source/execution_environment/root_device_environment.h"
#include "shared/source/gmm_helper/gmm.h"
#include "shared/source/gmm_helper/gmm_helper.h"
#include "shared/source/gmm_helper/resource_info.h"
#include "shared/source/helpers/aligned_memory.h"
#include "shared/source/helpers/get_info.h"
#include "shared/source/helpers/gfx_core_helper.h"
#include "shared/source/helpers/hw_info.h"
#include "shared/source/memory_manager/allocation_properties.h"
#include "shared/source/memory_manager/memory_manager.h"
#include "shared/source/os_interface/product_helper.h"
#include "opencl/extensions/public/cl_gl_private_intel.h"
#include "opencl/source/cl_device/cl_device.h"
#include "opencl/source/context/context.h"
#include "opencl/source/helpers/gmm_types_converter.h"
#include "opencl/source/mem_obj/image.h"
#include "opencl/source/sharings/gl/gl_texture.h"
#include "opencl/source/sharings/gl/linux/gl_sharing_linux.h"
#include "CL/cl_gl.h"
#include "config.h"
#include <GL/gl.h>
#include <poll.h>
#include <unistd.h>
namespace NEO {
Image *GlTexture::createSharedGlTexture(Context *context, cl_mem_flags flags, cl_GLenum target, cl_GLint miplevel, cl_GLuint texture,
cl_int *errcodeRet) {
ErrorCodeHelper errorCode(errcodeRet, CL_INVALID_GL_OBJECT);
auto memoryManager = context->getMemoryManager();
cl_image_desc imgDesc = {};
ImageInfo imgInfo = {};
cl_image_format imgFormat = {};
McsSurfaceInfo mcsSurfaceInfo = {};
/* Prepare export request */
struct mesa_glinterop_export_in texIn = {};
struct mesa_glinterop_export_out texOut = {};
texIn.version = 2;
texIn.target = getBaseTargetType(target);
texIn.obj = texture;
texIn.miplevel = miplevel;
switch (flags) {
case CL_MEM_READ_ONLY:
texIn.access = MESA_GLINTEROP_ACCESS_READ_ONLY;
break;
case CL_MEM_WRITE_ONLY:
texIn.access = MESA_GLINTEROP_ACCESS_WRITE_ONLY;
break;
case CL_MEM_READ_WRITE:
texIn.access = MESA_GLINTEROP_ACCESS_READ_WRITE;
break;
default:
errorCode.set(CL_INVALID_VALUE);
return nullptr;
}
if (texIn.target != GL_TEXTURE_2D) {
printf("target %x not supported\n", target);
errorCode.set(CL_INVALID_GL_OBJECT);
return nullptr;
}
texOut.version = 2;
/* Call MESA interop */
GLSharingFunctionsLinux *sharingFunctions = context->getSharing<GLSharingFunctionsLinux>();
int retValue = sharingFunctions->exportObject(&texIn, &texOut);
if ((retValue != MESA_GLINTEROP_SUCCESS) || (texOut.version != 2)) {
switch (retValue) {
case MESA_GLINTEROP_INVALID_DISPLAY:
case MESA_GLINTEROP_INVALID_CONTEXT:
errorCode.set(CL_INVALID_CONTEXT);
break;
case MESA_GLINTEROP_INVALID_TARGET:
errorCode.set(CL_INVALID_VALUE);
break;
case MESA_GLINTEROP_INVALID_MIP_LEVEL:
errorCode.set(CL_INVALID_MIP_LEVEL);
break;
case MESA_GLINTEROP_INVALID_OBJECT:
errorCode.set(CL_INVALID_GL_OBJECT);
break;
case MESA_GLINTEROP_UNSUPPORTED:
errorCode.set(CL_INVALID_IMAGE_FORMAT_DESCRIPTOR);
break;
case MESA_GLINTEROP_INVALID_OPERATION:
errorCode.set(CL_INVALID_OPERATION);
break;
case MESA_GLINTEROP_OUT_OF_HOST_MEMORY:
errorCode.set(CL_OUT_OF_HOST_MEMORY);
break;
case MESA_GLINTEROP_OUT_OF_RESOURCES:
default:
errorCode.set(CL_OUT_OF_RESOURCES);
break;
}
}
/* Map result for rest of the function */
CL_GL_RESOURCE_INFO texInfo = {};
texInfo.name = texIn.obj,
texInfo.globalShareHandle = static_cast<unsigned int>(texOut.dmabuf_fd);
texInfo.glInternalFormat = static_cast<GLint>(texOut.internal_format);
texInfo.textureBufferSize = static_cast<GLint>(texOut.buf_size);
texInfo.textureBufferOffset = static_cast<GLint>(texOut.buf_offset);
imgDesc.image_width = texOut.width;
imgDesc.image_height = texOut.height;
imgDesc.image_depth = texOut.depth;
imgDesc.image_row_pitch = texOut.stride;
imgInfo.imgDesc.imageType = ImageType::image2D;
imgInfo.imgDesc.imageWidth = imgDesc.image_width;
imgInfo.imgDesc.imageHeight = imgDesc.image_height;
imgInfo.imgDesc.imageDepth = imgDesc.image_depth;
imgInfo.imgDesc.imageRowPitch = imgDesc.image_row_pitch;
imgInfo.linearStorage = (texOut.modifier == 0);
errorCode.set(CL_SUCCESS);
if (setClImageFormat(texInfo.glInternalFormat, imgFormat) == false) {
errorCode.set(CL_INVALID_GL_OBJECT);
return nullptr;
}
auto surfaceFormatInfoAddress = Image::getSurfaceFormatFromTable(flags, &imgFormat, context->getDevice(0)->getHardwareInfo().capabilityTable.supportsOcl21Features);
if (!surfaceFormatInfoAddress) {
errorCode.set(CL_INVALID_GL_OBJECT);
return nullptr;
}
auto surfaceFormatInfo = *surfaceFormatInfoAddress;
imgInfo.surfaceFormat = &surfaceFormatInfo.surfaceFormat;
AllocationProperties allocProperties(context->getDevice(0)->getRootDeviceIndex(),
false, // allocateMemory
&imgInfo,
AllocationType::sharedImage,
context->getDeviceBitfieldForAllocation(context->getDevice(0)->getRootDeviceIndex()));
auto alloc = memoryManager->createGraphicsAllocationFromSharedHandle(texInfo.globalShareHandle, allocProperties, false, false, false, nullptr);
if (alloc == nullptr) {
errorCode.set(CL_INVALID_GL_OBJECT);
return nullptr;
}
memoryManager->closeSharedHandle(alloc);
auto gmm = alloc->getDefaultGmm();
imgDesc.image_type = getClMemObjectType(target);
if (target == GL_TEXTURE_BUFFER) {
imgDesc.image_row_pitch = texInfo.textureBufferSize;
} else if (imgDesc.image_row_pitch == 0) {
imgDesc.image_row_pitch = gmm->gmmResourceInfo->getRenderPitch();
if (imgDesc.image_row_pitch == 0) {
size_t alignedWidth = alignUp(imgDesc.image_width, gmm->gmmResourceInfo->getHAlign());
size_t bpp = gmm->gmmResourceInfo->getBitsPerPixel() >> 3;
imgDesc.image_row_pitch = alignedWidth * bpp;
}
}
uint32_t numSamples = static_cast<uint32_t>(gmm->gmmResourceInfo->getNumSamples());
imgDesc.num_samples = getValidParam(numSamples, 0u, 1u);
imgDesc.image_height = gmm->gmmResourceInfo->getBaseHeight();
imgDesc.image_array_size = gmm->gmmResourceInfo->getArraySize();
if (target == GL_TEXTURE_3D) {
imgDesc.image_depth = gmm->gmmResourceInfo->getBaseDepth();
}
if (imgDesc.image_array_size > 1 || imgDesc.image_depth > 1) {
GMM_REQ_OFFSET_INFO gmmReqInfo = {};
gmmReqInfo.ArrayIndex = imgDesc.image_array_size > 1 ? 1 : 0;
gmmReqInfo.Slice = imgDesc.image_depth > 1 ? 1 : 0;
gmmReqInfo.ReqLock = 1;
gmm->gmmResourceInfo->getOffset(gmmReqInfo);
imgDesc.image_slice_pitch = gmmReqInfo.Lock.Offset;
} else {
imgDesc.image_slice_pitch = alloc->getUnderlyingBufferSize();
}
uint32_t cubeFaceIndex = GmmTypesConverter::getCubeFaceIndex(target);
uint32_t qPitch = gmm->queryQPitch(gmm->gmmResourceInfo->getResourceType());
GraphicsAllocation *mcsAlloc = nullptr;
imgDesc.image_type = CL_MEM_OBJECT_IMAGE2D;
imgInfo.imgDesc = Image::convertDescriptor(imgDesc);
imgInfo.surfaceFormat = &surfaceFormatInfo.surfaceFormat;
imgInfo.qPitch = qPitch;
auto glTexture = new GlTexture(sharingFunctions, getClGlObjectType(target), texture, texInfo, target, std::max(miplevel, 0));
if (texInfo.isAuxEnabled && alloc->getDefaultGmm()->unifiedAuxTranslationCapable()) {
const auto &hwInfo = context->getDevice(0)->getHardwareInfo();
const auto &productHelper = context->getDevice(0)->getRootDeviceEnvironment().getHelper<ProductHelper>();
alloc->getDefaultGmm()->isCompressionEnabled = productHelper.isPageTableManagerSupported(hwInfo) ? memoryManager->mapAuxGpuVA(alloc)
: true;
}
auto multiGraphicsAllocation = MultiGraphicsAllocation(context->getDevice(0)->getRootDeviceIndex());
multiGraphicsAllocation.addAllocation(alloc);
return Image::createSharedImage(context, glTexture, mcsSurfaceInfo, std::move(multiGraphicsAllocation), mcsAlloc, flags, 0, &surfaceFormatInfo, imgInfo, cubeFaceIndex,
std::max(miplevel, 0), imgInfo.imgDesc.numMipLevels);
}
void GlTexture::synchronizeObject(UpdateData &updateData) {
auto sharingFunctions = static_cast<GLSharingFunctionsLinux *>(this->sharingFunctions);
/* Prepare flush request */
struct mesa_glinterop_export_in texIn = {};
struct mesa_glinterop_flush_out syncOut = {};
int fenceFd = -1;
texIn.version = 2;
texIn.target = this->target;
texIn.obj = this->clGlObjectId;
texIn.miplevel = this->miplevel;
syncOut.version = 1;
syncOut.fence_fd = &fenceFd;
/* Call MESA interop */
int retValue = sharingFunctions->flushObjects(1, &texIn, &syncOut);
if (retValue != MESA_GLINTEROP_SUCCESS) {
updateData.synchronizationStatus = SynchronizeStatus::SYNCHRONIZE_ERROR;
return;
}
/* Wait on the fence fd */
struct pollfd fp = {
.fd = fenceFd,
.events = POLLIN,
.revents = 0,
};
poll(&fp, 1, 1000);
close(fenceFd);
/* Done */
updateData.synchronizationStatus = SynchronizeStatus::ACQUIRE_SUCCESFUL;
}
cl_int GlTexture::getGlTextureInfo(cl_gl_texture_info paramName, size_t paramValueSize, void *paramValue, size_t *paramValueSizeRet) const {
GetInfoHelper info(paramValue, paramValueSize, paramValueSizeRet);
if (paramName == CL_GL_TEXTURE_TARGET) {
info.set<GLenum>(target);
} else if (paramName == CL_GL_MIPMAP_LEVEL) {
info.set<GLenum>(miplevel);
} else if (paramName == CL_GL_NUM_SAMPLES) {
info.set<GLsizei>(textureInfo.numberOfSamples > 1 ? textureInfo.numberOfSamples : 0);
} else {
return CL_INVALID_VALUE;
}
return CL_SUCCESS;
}
cl_mem_object_type GlTexture::getClMemObjectType(cl_GLenum glType) {
return static_cast<cl_mem_object_type>(getClObjectType(glType, false));
}
cl_gl_object_type GlTexture::getClGlObjectType(cl_GLenum glType) {
return static_cast<cl_gl_object_type>(getClObjectType(glType, true));
}
uint32_t GlTexture::getClObjectType(cl_GLenum glType, bool returnClGlObjectType) {
// return cl_gl_object_type if returnClGlObjectType is ture, otherwise cl_mem_object_type
uint32_t retValue = 0;
switch (glType) {
case GL_TEXTURE_1D:
retValue = returnClGlObjectType ? CL_GL_OBJECT_TEXTURE1D : CL_MEM_OBJECT_IMAGE1D;
break;
case GL_TEXTURE_1D_ARRAY:
retValue = returnClGlObjectType ? CL_GL_OBJECT_TEXTURE1D_ARRAY : CL_MEM_OBJECT_IMAGE1D_ARRAY;
break;
case GL_TEXTURE_2D:
case GL_TEXTURE_RECTANGLE:
case GL_TEXTURE_CUBE_MAP_NEGATIVE_X:
case GL_TEXTURE_CUBE_MAP_POSITIVE_X:
case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y:
case GL_TEXTURE_CUBE_MAP_POSITIVE_Y:
case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z:
case GL_TEXTURE_CUBE_MAP_POSITIVE_Z:
case GL_TEXTURE_2D_MULTISAMPLE:
retValue = returnClGlObjectType ? CL_GL_OBJECT_TEXTURE2D : CL_MEM_OBJECT_IMAGE2D;
break;
case GL_TEXTURE_2D_ARRAY:
case GL_TEXTURE_2D_MULTISAMPLE_ARRAY:
retValue = returnClGlObjectType ? CL_GL_OBJECT_TEXTURE2D_ARRAY : CL_MEM_OBJECT_IMAGE2D_ARRAY;
break;
case GL_TEXTURE_3D:
retValue = returnClGlObjectType ? CL_GL_OBJECT_TEXTURE3D : CL_MEM_OBJECT_IMAGE3D;
break;
case GL_TEXTURE_BUFFER:
retValue = returnClGlObjectType ? CL_GL_OBJECT_TEXTURE_BUFFER : CL_MEM_OBJECT_IMAGE1D_BUFFER;
break;
case GL_RENDERBUFFER_EXT:
retValue = returnClGlObjectType ? CL_GL_OBJECT_RENDERBUFFER : CL_MEM_OBJECT_IMAGE2D;
break;
default:
retValue = 0;
break;
}
return retValue;
}
cl_GLenum GlTexture::getBaseTargetType(cl_GLenum target) {
cl_GLenum returnTarget = 0;
switch (target) {
case GL_TEXTURE_CUBE_MAP_NEGATIVE_X:
case GL_TEXTURE_CUBE_MAP_POSITIVE_X:
case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y:
case GL_TEXTURE_CUBE_MAP_POSITIVE_Y:
case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z:
case GL_TEXTURE_CUBE_MAP_POSITIVE_Z:
returnTarget = GL_TEXTURE_CUBE_MAP_ARB;
break;
default:
returnTarget = target;
break;
}
return returnTarget;
}
void GlTexture::releaseResource(MemObj *memObject, uint32_t rootDeviceIndex) {
auto memoryManager = memObject->getMemoryManager();
memoryManager->closeSharedHandle(memObject->getGraphicsAllocation(rootDeviceIndex));
}
void GlTexture::resolveGraphicsAllocationChange(osHandle currentSharedHandle, UpdateData *updateData) {
const auto memObject = updateData->memObject;
auto graphicsAllocation = memObject->getGraphicsAllocation(updateData->rootDeviceIndex);
graphicsAllocation->setSharedHandle(updateData->sharedHandle);
}
} // namespace NEO