Emulation/ppsspp
1
0
Fork 0
mirror of https://github.com/hrydgard/ppsspp.git synced 2025-04-02 11:01:50 -04:00
Code Issues Projects Releases Packages Wiki Activity

Make the barrier a bit more focused. Up some limits.

Browse source
This commit is contained in:
Henrik Rydgård 2017-12-11 20:23:10 +01:00 • committed by Unknown W. Brackets
parent 080f34acfe
commit eeec18c018
3 changed files with 15 additions and 8 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

9
GPU/Vulkan/TextureCacheVulkan.cpp
View file

@ -836,11 +836,16 @@ void TextureCacheVulkan::BuildTexture(TexCacheEntry *const entry) {
vkCmdDispatch(cmdInit, (mipWidth + 15) / 16, (mipHeight + 15) / 16, 1);
// After the compute, before the copy, we need a memory barrier.
VkMemoryBarrier barrier{ VK_STRUCTURE_TYPE_MEMORY_BARRIER };
VkBufferMemoryBarrier barrier{ VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER };
barrier.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT;
barrier.dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT;
barrier.buffer = localBuf;
barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barrier.offset = localOffset;
barrier.size = localSize;
vkCmdPipelineBarrier(cmdInit, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
0, 1, &barrier, 0, nullptr, 0, nullptr);
0, 0, nullptr, 1, &barrier, 0, nullptr);
entry->vkTex->UploadMip(cmdInit, i, mipWidth, mipHeight, localBuf, localOffset, stride / bpp);
} else {

13
GPU/Vulkan/VulkanUtil.cpp
View file

@ -432,14 +432,14 @@ void VulkanComputeShaderManager::InitDeviceObjects() {
assert(VK_SUCCESS == res);
VkDescriptorPoolSize dpTypes[2];
dpTypes[0].descriptorCount = 2048;
dpTypes[0].descriptorCount = 8192;
dpTypes[0].type = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
dpTypes[1].descriptorCount = 1024;
dpTypes[1].descriptorCount = 4096;
dpTypes[1].type = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE;
VkDescriptorPoolCreateInfo dp = { VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO };
dp.flags = 0; // Don't want to mess around with individually freeing these, let's go fixed each frame and zap the whole array. Might try the dynamic approach later.
dp.maxSets = 1024;
dp.maxSets = 4096; // GTA can end up creating more than 1000 textures in the first frame!
dp.pPoolSizes = dpTypes;
dp.poolSizeCount = ARRAY_SIZE(dpTypes);
for (int i = 0; i < ARRAY_SIZE(frameData_); i++) {
@ -484,12 +484,12 @@ void VulkanComputeShaderManager::DestroyDeviceObjects() {
VkDescriptorSet VulkanComputeShaderManager::GetDescriptorSet(VkImageView image, VkBuffer buffer, VkDeviceSize offset, VkDeviceSize range, VkBuffer buffer2, VkDeviceSize offset2, VkDeviceSize range2) {
int curFrame = vulkan_->GetCurFrame();
FrameData *frame = &frameData_[curFrame];
FrameData &frameData = frameData_[curFrame];
frameData_[curFrame].numDescriptors++;
VkDescriptorSet desc;
VkDescriptorSetAllocateInfo descAlloc = { VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO };
descAlloc.pSetLayouts = &descriptorSetLayout_;
descAlloc.descriptorPool = frame->descPool;
descAlloc.descriptorPool = frameData.descPool;
descAlloc.descriptorSetCount = 1;
VkResult result = vkAllocateDescriptorSets(vulkan_->GetDevice(), &descAlloc, &desc);
assert(result == VK_SUCCESS);
@ -559,6 +559,7 @@ VkPipeline VulkanComputeShaderManager::GetPipeline(VkShaderModule cs) {
void VulkanComputeShaderManager::BeginFrame() {
int curFrame = vulkan_->GetCurFrame();
FrameData &frame = frameData_[curFrame];
frameData_[curFrame].numDescriptors = 0;
vkResetDescriptorPool(vulkan_->GetDevice(), frame.descPool, 0);
}

1
GPU/Vulkan/VulkanUtil.h
View file

@ -161,6 +161,7 @@ private:
struct FrameData {
VkDescriptorPool descPool;
int numDescriptors;
};
FrameData frameData_[VulkanContext::MAX_INFLIGHT_FRAMES];