better storage buffer support

This commit is contained in:
Yun Hsiao Wu
2021-01-14 14:33:22 +08:00
parent f1d96c6d9c
commit ee05ce4fa0
7 changed files with 32 additions and 21 deletions
-9
View File
@@ -898,19 +898,10 @@ struct UniformSampler {
typedef cc::vector<UniformSampler> UniformSamplerList;
struct StorageBufferMember {
String name;
Type type = Type::UNKNOWN;
uint count = 0u;
};
typedef cc::vector<StorageBufferMember> StorageBufferMemberList;
struct StorageBuffer {
uint set = 0u;
uint binding = 0u;
String name;
StorageBufferMemberList members;
uint count = 0u;
};
@@ -1959,7 +1959,7 @@ void GLES2CmdFuncBindState(GLES2Device *device, GLES2GPUPipelineState *gpuPipeli
}
GL_CHECK(glTexParameteri(gpuTexture->glTarget, GL_TEXTURE_WRAP_S,
glWrapS));
glWrapS));
gpuTexture->glWrapS = glWrapS;
}
@@ -1969,7 +1969,7 @@ void GLES2CmdFuncBindState(GLES2Device *device, GLES2GPUPipelineState *gpuPipeli
cache->texUint = unit;
}
GL_CHECK(glTexParameteri(gpuTexture->glTarget, GL_TEXTURE_WRAP_T,
glWrapT));
glWrapT));
gpuTexture->glWrapT = glWrapT;
}
@@ -1979,7 +1979,7 @@ void GLES2CmdFuncBindState(GLES2Device *device, GLES2GPUPipelineState *gpuPipeli
cache->texUint = unit;
}
GL_CHECK(glTexParameteri(gpuTexture->glTarget, GL_TEXTURE_MIN_FILTER,
glMinFilter));
glMinFilter));
gpuTexture->glMinFilter = glMinFilter;
}
@@ -1989,7 +1989,7 @@ void GLES2CmdFuncBindState(GLES2Device *device, GLES2GPUPipelineState *gpuPipeli
cache->texUint = unit;
}
GL_CHECK(glTexParameteri(gpuTexture->glTarget, GL_TEXTURE_MAG_FILTER,
gpuDescriptor->gpuSampler->glMagFilter));
gpuDescriptor->gpuSampler->glMagFilter));
gpuTexture->glMagFilter = gpuDescriptor->gpuSampler->glMagFilter;
}
}
@@ -428,7 +428,7 @@ void GLES3CommandBuffer::pipelineBarrier(const GlobalBarrier& barrier) {
(_type == CommandBufferType::SECONDARY)) {
GLES3CmdBarrier *cmd = _cmdAllocator->barrierCmdPool.alloc();
MapGLBarriers(barrier, cmd->barriers, cmd->barriersByRegion);
MapGLBarriers(barrier.nextAccesses, cmd->barriers, cmd->barriersByRegion);
_curCmdPackage->barrierCmds.push(cmd);
_curCmdPackage->cmds.push(GLES3CmdType::BARRIER);
} else {
@@ -600,8 +600,7 @@ const GLenum GLES3_BLEND_FACTORS[] = {
};
} // namespace
void MapGLBarriers(const GlobalBarrier &barrier, GLbitfield &glBarriers, GLbitfield &glBarriersByRegion) {
const AccessTypeList &list = barrier.nextAccesses;
void MapGLBarriers(const AccessTypeList &list, GLbitfield &glBarriers, GLbitfield &glBarriersByRegion) {
for (size_t j = 0u; j < list.size(); ++j) {
switch (list[j]) {
case AccessType::INDIRECT_BUFFER:
@@ -638,8 +637,9 @@ void MapGLBarriers(const GlobalBarrier &barrier, GLbitfield &glBarriers, GLbitfi
case AccessType::TRANSFER_READ:
glBarriers |= GL_PIXEL_BUFFER_BARRIER_BIT;
break;
case AccessType::VERTEX_SHADER_WRITE:
case AccessType::FRAGMENT_SHADER_WRITE:
glBarriersByRegion |= GL_FRAMEBUFFER_BARRIER_BIT;
break;
case AccessType::COMPUTE_SHADER_WRITE:
glBarriersByRegion |= GL_SHADER_STORAGE_BARRIER_BIT;
break;
@@ -647,6 +647,7 @@ void MapGLBarriers(const GlobalBarrier &barrier, GLbitfield &glBarriers, GLbitfi
glBarriers |= GL_TEXTURE_UPDATE_BARRIER_BIT;
glBarriers |= GL_BUFFER_UPDATE_BARRIER_BIT;
break;
case AccessType::VERTEX_SHADER_WRITE:
case AccessType::FRAGMENT_SHADER_READ_COLOR_INPUT_ATTACHMENT:
case AccessType::FRAGMENT_SHADER_READ_DEPTH_STENCIL_INPUT_ATTACHMENT:
case AccessType::HOST_PREINITIALIZED:
@@ -714,6 +715,18 @@ void GLES3CmdFuncCreateBuffer(GLES3Device *device, GLES3GPUBuffer *gpuBuffer) {
GL_CHECK(glBindBuffer(GL_UNIFORM_BUFFER, 0));
device->stateCache()->glUniformBuffer = 0;
}
} else if (gpuBuffer->usage & BufferUsageBit::STORAGE) {
gpuBuffer->glTarget = GL_SHADER_STORAGE_BUFFER;
GL_CHECK(glGenBuffers(1, &gpuBuffer->glBuffer));
if (gpuBuffer->size) {
if (device->stateCache()->glShaderStorageBuffer != gpuBuffer->glBuffer) {
GL_CHECK(glBindBuffer(GL_SHADER_STORAGE_BUFFER, gpuBuffer->glBuffer));
}
GL_CHECK(glBufferData(GL_SHADER_STORAGE_BUFFER, gpuBuffer->size, nullptr, glUsage));
GL_CHECK(glBindBuffer(GL_SHADER_STORAGE_BUFFER, 0));
device->stateCache()->glShaderStorageBuffer = 0;
}
} else if (gpuBuffer->usage & BufferUsageBit::INDIRECT) {
gpuBuffer->glTarget = GL_NONE;
} else if ((gpuBuffer->usage & BufferUsageBit::TRANSFER_DST) ||
@@ -1270,7 +1283,6 @@ void GLES3CmdFuncCreateShader(GLES3Device *device, GLES3GPUShader *gpuShader) {
gpuBlock.binding = buffer.binding;
gpuBlock.glBinding = buffer.binding + device->bindingMappingInfo().bufferOffsets[buffer.set];
gpuBlock.isStorage = true;
break;
}
// create uniform samplers
@@ -2374,6 +2386,14 @@ void GLES3CmdFuncUpdateBuffer(GLES3Device *device, GLES3GPUBuffer *gpuBuffer, co
GL_CHECK(glBufferSubData(GL_UNIFORM_BUFFER, offset, size, buffer));
break;
}
case GL_SHADER_STORAGE_BUFFER: {
if (device->stateCache()->glShaderStorageBuffer != gpuBuffer->glBuffer) {
GL_CHECK(glBindBuffer(GL_SHADER_STORAGE_BUFFER, gpuBuffer->glBuffer));
device->stateCache()->glShaderStorageBuffer = gpuBuffer->glBuffer;
}
GL_CHECK(glBufferSubData(GL_SHADER_STORAGE_BUFFER, offset, size, buffer));
break;
}
default:
CCASSERT(false, "Unsupported BufferType, update buffer failed.");
break;
@@ -270,7 +270,7 @@ public:
}
};
CC_GLES3_API void MapGLBarriers(const GlobalBarrier &barrier, GLbitfield &glBarriers, GLbitfield &glBarriersByRegion);
CC_GLES3_API void MapGLBarriers(const AccessTypeList &barrier, GLbitfield &glBarriers, GLbitfield &glBarriersByRegion);
CC_GLES3_API void GLES3CmdFuncCreateBuffer(GLES3Device *device, GLES3GPUBuffer *gpuBuffer);
CC_GLES3_API void GLES3CmdFuncDestroyBuffer(GLES3Device *device, GLES3GPUBuffer *gpuBuffer);
CC_GLES3_API void GLES3CmdFuncResizeBuffer(GLES3Device *device, GLES3GPUBuffer *gpuBuffer);
@@ -179,7 +179,7 @@ void GLES3PrimaryCommandBuffer::pipelineBarrier(const GlobalBarrier& barrier) {
GLbitfield glBarriers = 0u;
GLbitfield glBarriersByRegion = 0u;
MapGLBarriers(barrier, glBarriers, glBarriersByRegion);
MapGLBarriers(barrier.nextAccesses, glBarriers, glBarriersByRegion);
GLES3CmdFuncMemoryBarrier((GLES3Device *)_device, glBarriers, glBarriersByRegion);
} else {
CC_LOG_ERROR("Command 'pipelineBarrier' must be recorded outside a render pass.");
@@ -472,7 +472,7 @@ void CCVKDevice::destroy() {
CC_SAFE_DESTROY(_context);
}
// op-op since we maintain surface size internally
// no-op since we maintain surface size internally
void CCVKDevice::resize(uint width, uint height) {}
void CCVKDevice::acquire() {