lee/renderdoc
1
0
Fork 0
mirror of https://github.com/baldurk/renderdoc.git synced 2026-05-13 05:20:45 +00:00
Code Issues Packages Projects Releases Wiki Activity
Files
0fce39ff04401d1b21c54c4a17addde35dd2d722
renderdoc/renderdoc/driver/vulkan/vk_overlay.cpp
T
baldurk 0fce39ff04 Implement workaround for broken occlusion queries on Intel. Closes #2591 
* If a draw has no effects (color mask writes off, no depth attachment) then
  Intel will not return proper occlusion query results. Forcibly patching in a
  storage write works around the problem.
2022-06-16 15:53:50 +01:00

2905 lines
104 KiB
C++
Raw Blame History

/******************************************************************************
* The MIT License (MIT)
*
* Copyright (c) 2019-2022 Baldur Karlsson
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
* THE SOFTWARE.
******************************************************************************/
#include <float.h>
#include <math.h>
#include <algorithm>
#include "data/glsl_shaders.h"
#include "driver/shaders/spirv/spirv_common.h"
#include "driver/shaders/spirv/spirv_gen.h"
#include "maths/formatpacking.h"
#include "maths/matrix.h"
#include "strings/string_utils.h"
#include "vk_core.h"
#include "vk_debug.h"
#include "vk_replay.h"
#include "vk_shader_cache.h"
#define VULKAN 1
#include "data/glsl/glsl_ubos_cpp.h"
struct VulkanQuadOverdrawCallback : public VulkanActionCallback
{
VulkanQuadOverdrawCallback(WrappedVulkan *vk, VkDescriptorSetLayout descSetLayout,
VkDescriptorSet descSet, const rdcarray<uint32_t> &events)
: m_pDriver(vk), m_DescSetLayout(descSetLayout), m_DescSet(descSet), m_Events(events)
{
m_pDriver->SetActionCB(this);
}
~VulkanQuadOverdrawCallback()
{
m_pDriver->SetActionCB(NULL);
VkDevice dev = m_pDriver->GetDev();
for(auto it = m_PipelineCache.begin(); it != m_PipelineCache.end(); ++it)
{
m_pDriver->vkDestroyPipeline(dev, it->second.pipe, NULL);
m_pDriver->vkDestroyPipelineLayout(dev, it->second.pipeLayout, NULL);
}
}
void PreDraw(uint32_t eid, VkCommandBuffer cmd)
{
if(!m_Events.contains(eid))
return;
// we customise the pipeline to disable framebuffer writes, but perform normal testing
// and substitute our quad calculation fragment shader that writes to a storage image
// that is bound in a new descriptor set.
VkResult vkr = VK_SUCCESS;
m_PrevState = m_pDriver->GetCmdRenderState();
VulkanRenderState &pipestate = m_pDriver->GetCmdRenderState();
// check cache first
CachedPipeline pipe = m_PipelineCache[pipestate.graphics.pipeline];
// if we don't get a hit, create a modified pipeline
if(pipe.pipe == VK_NULL_HANDLE)
{
const VulkanCreationInfo::Pipeline &p =
m_pDriver->GetDebugManager()->GetPipelineInfo(pipestate.graphics.pipeline);
VkDescriptorSetLayout *descSetLayouts;
// descSet will be the index of our new descriptor set
uint32_t descSet = (uint32_t)p.descSetLayouts.size();
descSetLayouts = new VkDescriptorSetLayout[descSet + 1];
for(uint32_t i = 0; i < descSet; i++)
descSetLayouts[i] = m_pDriver->GetResourceManager()->GetCurrentHandle<VkDescriptorSetLayout>(
p.descSetLayouts[i]);
// this layout has storage image and
descSetLayouts[descSet] = m_DescSetLayout;
// don't have to handle separate vert/frag layouts as push constant ranges must be identical
const rdcarray<VkPushConstantRange> &push =
m_pDriver->GetDebugManager()->GetPipelineLayoutInfo(p.vertLayout).pushRanges;
VkPipelineLayoutCreateInfo pipeLayoutInfo = {
VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO,
NULL,
0,
descSet + 1,
descSetLayouts,
(uint32_t)push.size(),
push.empty() ? NULL : &push[0],
};
// create pipeline layout with same descriptor set layouts, plus our mesh output set
vkr = m_pDriver->vkCreatePipelineLayout(m_pDriver->GetDev(), &pipeLayoutInfo, NULL,
&pipe.pipeLayout);
m_pDriver->CheckVkResult(vkr);
SAFE_DELETE_ARRAY(descSetLayouts);
VkGraphicsPipelineCreateInfo pipeCreateInfo;
m_pDriver->GetShaderCache()->MakeGraphicsPipelineInfo(pipeCreateInfo,
pipestate.graphics.pipeline);
// repoint pipeline layout
pipeCreateInfo.layout = pipe.pipeLayout;
// disable colour writes/blends
VkPipelineColorBlendStateCreateInfo *cb =
(VkPipelineColorBlendStateCreateInfo *)pipeCreateInfo.pColorBlendState;
for(uint32_t i = 0; i < cb->attachmentCount; i++)
{
VkPipelineColorBlendAttachmentState *att =
(VkPipelineColorBlendAttachmentState *)&cb->pAttachments[i];
att->blendEnable = false;
att->colorWriteMask = 0x0;
}
// disable depth/stencil writes but keep any tests enabled
VkPipelineDepthStencilStateCreateInfo *ds =
(VkPipelineDepthStencilStateCreateInfo *)pipeCreateInfo.pDepthStencilState;
ds->depthWriteEnable = false;
ds->front.passOp = ds->front.failOp = ds->front.depthFailOp = VK_STENCIL_OP_KEEP;
ds->back.passOp = ds->back.failOp = ds->back.depthFailOp = VK_STENCIL_OP_KEEP;
// don't discard
VkPipelineRasterizationStateCreateInfo *rs =
(VkPipelineRasterizationStateCreateInfo *)pipeCreateInfo.pRasterizationState;
rs->rasterizerDiscardEnable = false;
rdcarray<uint32_t> spirv =
*m_pDriver->GetShaderCache()->GetBuiltinBlob(BuiltinShader::QuadWriteFS);
// patch spirv, change descriptor set to descSet value
size_t it = 5;
while(it < spirv.size())
{
uint16_t WordCount = spirv[it] >> rdcspv::WordCountShift;
rdcspv::Op opcode = rdcspv::Op(spirv[it] & rdcspv::OpCodeMask);
if(opcode == rdcspv::Op::Decorate &&
spirv[it + 2] == (uint32_t)rdcspv::Decoration::DescriptorSet)
{
spirv[it + 3] = descSet;
break;
}
it += WordCount;
}
VkShaderModuleCreateInfo modinfo = {
VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO,
NULL,
0,
spirv.size() * sizeof(uint32_t),
&spirv[0],
};
VkShaderModule module;
VkDevice dev = m_pDriver->GetDev();
vkr = m_pDriver->vkCreateShaderModule(dev, &modinfo, NULL, &module);
m_pDriver->CheckVkResult(vkr);
bool found = false;
for(uint32_t i = 0; i < pipeCreateInfo.stageCount; i++)
{
VkPipelineShaderStageCreateInfo &sh =
(VkPipelineShaderStageCreateInfo &)pipeCreateInfo.pStages[i];
if(sh.stage == VK_SHADER_STAGE_FRAGMENT_BIT)
{
sh.module = module;
sh.pName = "main";
found = true;
break;
}
}
if(!found)
{
// we know this is safe because it's pointing to a static array that's
// big enough for all shaders
VkPipelineShaderStageCreateInfo &sh =
(VkPipelineShaderStageCreateInfo &)pipeCreateInfo.pStages[pipeCreateInfo.stageCount++];
sh.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO;
sh.pNext = NULL;
sh.stage = VK_SHADER_STAGE_FRAGMENT_BIT;
sh.module = module;
sh.pName = "main";
sh.pSpecializationInfo = NULL;
}
vkr = m_pDriver->vkCreateGraphicsPipelines(dev, VK_NULL_HANDLE, 1, &pipeCreateInfo, NULL,
&pipe.pipe);
m_pDriver->CheckVkResult(vkr);
m_pDriver->vkDestroyShaderModule(dev, module, NULL);
pipe.descSet = descSet;
m_PipelineCache[pipestate.graphics.pipeline] = pipe;
}
// modify state for first draw call
pipestate.graphics.pipeline = GetResID(pipe.pipe);
RDCASSERT(pipestate.graphics.descSets.size() >= pipe.descSet);
pipestate.graphics.descSets.resize(pipe.descSet + 1);
pipestate.graphics.descSets[pipe.descSet].pipeLayout = GetResID(pipe.pipeLayout);
pipestate.graphics.descSets[pipe.descSet].descSet = GetResID(m_DescSet);
if(cmd)
pipestate.BindPipeline(m_pDriver, cmd, VulkanRenderState::BindGraphics, false);
}
bool PostDraw(uint32_t eid, VkCommandBuffer cmd)
{
if(!m_Events.contains(eid))
return false;
// restore the render state and go ahead with the real draw
m_pDriver->GetCmdRenderState() = m_PrevState;
RDCASSERT(cmd);
m_pDriver->GetCmdRenderState().BindPipeline(m_pDriver, cmd, VulkanRenderState::BindGraphics,
false);
return true;
}
void PostRedraw(uint32_t eid, VkCommandBuffer cmd)
{
// nothing to do
}
// Dispatches don't rasterize, so do nothing
void PreDispatch(uint32_t eid, VkCommandBuffer cmd) {}
bool PostDispatch(uint32_t eid, VkCommandBuffer cmd) { return false; }
void PostRedispatch(uint32_t eid, VkCommandBuffer cmd) {}
// Ditto copy/etc
void PreMisc(uint32_t eid, ActionFlags flags, VkCommandBuffer cmd) {}
bool PostMisc(uint32_t eid, ActionFlags flags, VkCommandBuffer cmd) { return false; }
void PostRemisc(uint32_t eid, ActionFlags flags, VkCommandBuffer cmd) {}
void PreEndCommandBuffer(VkCommandBuffer cmd) {}
void AliasEvent(uint32_t primary, uint32_t alias)
{
// don't care
}
bool SplitSecondary() { return false; }
bool ForceLoadRPs() { return false; }
void PreCmdExecute(uint32_t baseEid, uint32_t secondaryFirst, uint32_t secondaryLast,
VkCommandBuffer cmd)
{
}
void PostCmdExecute(uint32_t baseEid, uint32_t secondaryFirst, uint32_t secondaryLast,
VkCommandBuffer cmd)
{
}
WrappedVulkan *m_pDriver;
VkDescriptorSetLayout m_DescSetLayout;
VkDescriptorSet m_DescSet;
const rdcarray<uint32_t> &m_Events;
// cache modified pipelines
struct CachedPipeline
{
uint32_t descSet;
VkPipelineLayout pipeLayout;
VkPipeline pipe;
};
std::map<ResourceId, CachedPipeline> m_PipelineCache;
VulkanRenderState m_PrevState;
};
void VulkanDebugManager::PatchFixedColShader(VkShaderModule &mod, float col[4])
{
union
{
uint32_t *spirv;
float *data;
} alias;
rdcarray<uint32_t> spv = *m_pDriver->GetShaderCache()->GetBuiltinBlob(BuiltinShader::FixedColFS);
alias.spirv = &spv[0];
size_t spirvLength = spv.size();
int patched = 0;
size_t it = 5;
while(it < spirvLength)
{
uint16_t WordCount = alias.spirv[it] >> rdcspv::WordCountShift;
rdcspv::Op opcode = rdcspv::Op(alias.spirv[it] & rdcspv::OpCodeMask);
if(opcode == rdcspv::Op::Constant)
{
if(alias.data[it + 3] >= 1.0f && alias.data[it + 3] <= 1.5f)
alias.data[it + 3] = col[0];
else if(alias.data[it + 3] >= 2.0f && alias.data[it + 3] <= 2.5f)
alias.data[it + 3] = col[1];
else if(alias.data[it + 3] >= 3.0f && alias.data[it + 3] <= 3.5f)
alias.data[it + 3] = col[2];
else if(alias.data[it + 3] >= 4.0f && alias.data[it + 3] <= 4.5f)
alias.data[it + 3] = col[3];
else
RDCERR("Unexpected constant value");
patched++;
}
it += WordCount;
}
if(patched != 4)
RDCERR("Didn't patch all constants");
VkShaderModuleCreateInfo modinfo = {
VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO,
NULL,
0,
spv.size() * sizeof(uint32_t),
alias.spirv,
};
VkResult vkr = m_pDriver->vkCreateShaderModule(m_Device, &modinfo, NULL, &mod);
CheckVkResult(vkr);
}
void VulkanDebugManager::PatchLineStripIndexBuffer(const ActionDescription *action,
GPUBuffer &indexBuffer, uint32_t &indexCount)
{
VulkanRenderState &rs = m_pDriver->m_RenderState;
bytebuf indices;
uint8_t *idx8 = NULL;
uint16_t *idx16 = NULL;
uint32_t *idx32 = NULL;
if(action->flags & ActionFlags::Indexed)
{
GetBufferData(rs.ibuffer.buf,
rs.ibuffer.offs + uint64_t(action->indexOffset) * rs.ibuffer.bytewidth,
uint64_t(action->numIndices) * rs.ibuffer.bytewidth, indices);
if(rs.ibuffer.bytewidth == 4)
idx32 = (uint32_t *)indices.data();
else if(rs.ibuffer.bytewidth == 1)
idx8 = (uint8_t *)indices.data();
else
idx16 = (uint16_t *)indices.data();
}
// we just patch up to 32-bit since we'll be adding more indices and we might overflow 16-bit.
rdcarray<uint32_t> patchedIndices;
::PatchLineStripIndexBuffer(action, MakePrimitiveTopology(rs.primitiveTopology, 3), idx8, idx16,
idx32, patchedIndices);
indexBuffer.Create(m_pDriver, m_Device, patchedIndices.size() * sizeof(uint32_t), 1,
GPUBuffer::eGPUBufferIBuffer);
void *ptr = indexBuffer.Map(0, patchedIndices.size() * sizeof(uint32_t));
if(!ptr)
return;
memcpy(ptr, patchedIndices.data(), patchedIndices.size() * sizeof(uint32_t));
indexBuffer.Unmap();
rs.ibuffer.offs = 0;
rs.ibuffer.bytewidth = 4;
rs.ibuffer.buf = GetResID(indexBuffer.buf);
VkBufferMemoryBarrier uploadbarrier = {
VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER,
NULL,
VK_ACCESS_HOST_WRITE_BIT,
VK_ACCESS_INDEX_READ_BIT,
VK_QUEUE_FAMILY_IGNORED,
VK_QUEUE_FAMILY_IGNORED,
Unwrap(indexBuffer.buf),
0,
indexBuffer.totalsize,
};
VkCommandBuffer cmd = m_pDriver->GetNextCmd();
if(cmd == VK_NULL_HANDLE)
return;
VkCommandBufferBeginInfo beginInfo = {VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO, NULL,
VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT};
VkResult vkr = ObjDisp(m_Device)->BeginCommandBuffer(Unwrap(cmd), &beginInfo);
CheckVkResult(vkr);
// ensure host writes finish before using as index buffer
DoPipelineBarrier(cmd, 1, &uploadbarrier);
ObjDisp(m_Device)->EndCommandBuffer(Unwrap(cmd));
indexCount = (uint32_t)patchedIndices.size();
}
RenderOutputSubresource VulkanReplay::GetRenderOutputSubresource(ResourceId id)
{
const VulkanRenderState &state = m_pDriver->m_RenderState;
VulkanCreationInfo &c = m_pDriver->m_CreationInfo;
for(ResourceId viewid : state.GetFramebufferAttachments())
{
const VulkanCreationInfo::ImageView &viewInfo = c.m_ImageView[viewid];
if(viewid == id || viewInfo.image == id)
{
return RenderOutputSubresource(viewInfo.range.baseMipLevel,
c.m_ImageView[viewid].range.baseArrayLayer,
c.m_ImageView[viewid].range.layerCount);
}
}
return RenderOutputSubresource(~0U, ~0U, 0);
}
ResourceId VulkanReplay::RenderOverlay(ResourceId texid, FloatVector clearCol, DebugOverlay overlay,
uint32_t eventId, const rdcarray<uint32_t> &passEvents)
{
const VkDevDispatchTable *vt = ObjDisp(m_Device);
RenderOutputSubresource sub = GetRenderOutputSubresource(texid);
if(sub.slice == ~0U)
{
RDCERR("Rendering overlay for %s couldn't find output to get subresource.", ToStr(texid).c_str());
sub = RenderOutputSubresource(0, 0, 1);
}
VulkanShaderCache *shaderCache = m_pDriver->GetShaderCache();
VulkanCreationInfo::Image &iminfo = m_pDriver->m_CreationInfo.m_Image[texid];
VkCommandBuffer cmd = m_pDriver->GetNextCmd();
if(cmd == VK_NULL_HANDLE)
return ResourceId();
VkCommandBufferBeginInfo beginInfo = {VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO, NULL,
VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT};
VkResult vkr = vt->BeginCommandBuffer(Unwrap(cmd), &beginInfo);
CheckVkResult(vkr);
VkMarkerRegion::Begin(StringFormat::Fmt("RenderOverlay %d", overlay), cmd);
uint32_t multiviewMask = m_Overlay.MultiViewMask;
VulkanRenderState &state = m_pDriver->m_RenderState;
if(state.dynamicRendering.active)
{
multiviewMask = state.dynamicRendering.viewMask;
}
else if(state.GetRenderPass() != ResourceId())
{
const VulkanCreationInfo::RenderPass &rp =
m_pDriver->m_CreationInfo.m_RenderPass[state.GetRenderPass()];
multiviewMask = 0;
for(uint32_t v : rp.subpasses[state.subpass].multiviews)
multiviewMask |= 1U << v;
}
// if the overlay image is the wrong size, free it
if(m_Overlay.Image != VK_NULL_HANDLE &&
(iminfo.extent.width != m_Overlay.ImageDim.width ||
iminfo.extent.height != m_Overlay.ImageDim.height || iminfo.samples != m_Overlay.Samples ||
iminfo.mipLevels != m_Overlay.MipLevels || iminfo.arrayLayers != m_Overlay.ArrayLayers ||
multiviewMask != m_Overlay.MultiViewMask))
{
m_pDriver->vkDestroyRenderPass(m_Device, m_Overlay.NoDepthRP, NULL);
m_pDriver->vkDestroyFramebuffer(m_Device, m_Overlay.NoDepthFB, NULL);
m_pDriver->vkDestroyImageView(m_Device, m_Overlay.ImageView, NULL);
m_pDriver->vkDestroyImage(m_Device, m_Overlay.Image, NULL);
m_Overlay.Image = VK_NULL_HANDLE;
m_Overlay.ImageView = VK_NULL_HANDLE;
m_Overlay.NoDepthRP = VK_NULL_HANDLE;
m_Overlay.NoDepthFB = VK_NULL_HANDLE;
}
VkImageSubresourceRange subRange = {VK_IMAGE_ASPECT_COLOR_BIT, sub.mip, 1, sub.slice,
sub.numSlices};
const VkFormat overlayFormat = VK_FORMAT_R16G16B16A16_SFLOAT;
VkRenderPassMultiviewCreateInfo multiviewRP = {VK_STRUCTURE_TYPE_RENDER_PASS_MULTIVIEW_CREATE_INFO};
multiviewRP.correlationMaskCount = 1;
multiviewRP.pCorrelationMasks = &multiviewMask;
multiviewRP.subpassCount = 1;
multiviewRP.pViewMasks = &multiviewMask;
// create the overlay image if we don't have one already
// we go through the driver's creation functions so creation info
// is saved and the resources are registered as live resources for
// their IDs.
if(m_Overlay.Image == VK_NULL_HANDLE)
{
m_Overlay.ImageDim.width = iminfo.extent.width;
m_Overlay.ImageDim.height = iminfo.extent.height;
m_Overlay.MipLevels = iminfo.mipLevels;
m_Overlay.ArrayLayers = iminfo.arrayLayers;
m_Overlay.Samples = iminfo.samples;
m_Overlay.MultiViewMask = multiviewMask;
VkImageCreateInfo imInfo = {
VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
NULL,
0,
VK_IMAGE_TYPE_2D,
overlayFormat,
{m_Overlay.ImageDim.width, m_Overlay.ImageDim.height, 1},
(uint32_t)iminfo.mipLevels,
(uint32_t)iminfo.arrayLayers,
iminfo.samples,
VK_IMAGE_TILING_OPTIMAL,
VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT | VK_IMAGE_USAGE_SAMPLED_BIT |
VK_IMAGE_USAGE_TRANSFER_SRC_BIT | VK_IMAGE_USAGE_TRANSFER_DST_BIT,
VK_SHARING_MODE_EXCLUSIVE,
0,
NULL,
VK_IMAGE_LAYOUT_UNDEFINED,
};
vkr = m_pDriver->vkCreateImage(m_Device, &imInfo, NULL, &m_Overlay.Image);
CheckVkResult(vkr);
NameVulkanObject(m_Overlay.Image, "m_Overlay.Image");
VkMemoryRequirements mrq = {0};
m_pDriver->vkGetImageMemoryRequirements(m_Device, m_Overlay.Image, &mrq);
// if no memory is allocated, or it's not enough,
// then allocate
if(m_Overlay.ImageMem == VK_NULL_HANDLE || mrq.size > m_Overlay.ImageMemSize)
{
if(m_Overlay.ImageMem != VK_NULL_HANDLE)
{
m_pDriver->vkFreeMemory(m_Device, m_Overlay.ImageMem, NULL);
}
VkMemoryAllocateInfo allocInfo = {
VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO, NULL, mrq.size,
m_pDriver->GetGPULocalMemoryIndex(mrq.memoryTypeBits),
};
vkr = m_pDriver->vkAllocateMemory(m_Device, &allocInfo, NULL, &m_Overlay.ImageMem);
CheckVkResult(vkr);
if(vkr != VK_SUCCESS)
return ResourceId();
m_Overlay.ImageMemSize = mrq.size;
}
vkr = m_pDriver->vkBindImageMemory(m_Device, m_Overlay.Image, m_Overlay.ImageMem, 0);
CheckVkResult(vkr);
// need to update image layout into valid state
m_pDriver->FindImageState(GetResID(m_Overlay.Image))
->InlineTransition(
cmd, m_pDriver->m_QueueFamilyIdx, VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL, 0,
VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT, m_pDriver->GetImageTransitionInfo());
VkAttachmentDescription colDesc = {
0,
imInfo.format,
imInfo.samples,
VK_ATTACHMENT_LOAD_OP_LOAD,
VK_ATTACHMENT_STORE_OP_STORE,
VK_ATTACHMENT_LOAD_OP_DONT_CARE,
VK_ATTACHMENT_STORE_OP_DONT_CARE,
VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL,
VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL,
};
VkAttachmentReference colRef = {0, VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL};
VkSubpassDescription subp = {
0, VK_PIPELINE_BIND_POINT_GRAPHICS,
0, NULL, // inputs
1, &colRef, // color
NULL, // resolve
NULL, // depth-stencil
0, NULL, // preserve
};
VkRenderPassCreateInfo rpinfo = {
VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO,
NULL,
0,
1,
&colDesc,
1,
&subp,
0,
NULL, // dependencies
};
if(multiviewMask > 0)
rpinfo.pNext = &multiviewRP;
vkr = m_pDriver->vkCreateRenderPass(m_Device, &rpinfo, NULL, &m_Overlay.NoDepthRP);
CheckVkResult(vkr);
}
if(m_Overlay.ViewMip != sub.mip || m_Overlay.ViewSlice != sub.slice ||
m_Overlay.ViewNumSlices != sub.numSlices || m_Overlay.ImageView == VK_NULL_HANDLE)
{
m_pDriver->vkDestroyFramebuffer(m_Device, m_Overlay.NoDepthFB, NULL);
m_pDriver->vkDestroyImageView(m_Device, m_Overlay.ImageView, NULL);
m_Overlay.ViewMip = sub.mip;
m_Overlay.ViewSlice = sub.slice;
m_Overlay.ViewNumSlices = sub.numSlices;
VkImageViewCreateInfo viewInfo = {
VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
NULL,
0,
m_Overlay.Image,
VK_IMAGE_VIEW_TYPE_2D,
overlayFormat,
{VK_COMPONENT_SWIZZLE_IDENTITY, VK_COMPONENT_SWIZZLE_IDENTITY,
VK_COMPONENT_SWIZZLE_IDENTITY, VK_COMPONENT_SWIZZLE_IDENTITY},
subRange,
};
vkr = m_pDriver->vkCreateImageView(m_Device, &viewInfo, NULL, &m_Overlay.ImageView);
CheckVkResult(vkr);
// Create framebuffer rendering just to overlay image, no depth
VkFramebufferCreateInfo fbinfo = {
VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO,
NULL,
0,
m_Overlay.NoDepthRP,
1,
&m_Overlay.ImageView,
RDCMAX(1U, m_Overlay.ImageDim.width >> sub.mip),
RDCMAX(1U, m_Overlay.ImageDim.height >> sub.mip),
sub.numSlices,
};
vkr = m_pDriver->vkCreateFramebuffer(m_Device, &fbinfo, NULL, &m_Overlay.NoDepthFB);
CheckVkResult(vkr);
// can't create a framebuffer or renderpass for overlay image + depth as that
// needs to match the depth texture type wherever our draw is.
}
// bail out if the render area is outside our image.
// This is an order-of-operations problem, if the overlay is set when the event is changed it is
// refreshed before the UI layer can update the current texture.
if(state.renderArea.offset.x + state.renderArea.extent.width >
(m_Overlay.ImageDim.width >> sub.mip) ||
state.renderArea.offset.y + state.renderArea.extent.height >
(m_Overlay.ImageDim.height >> sub.mip))
{
return GetResID(m_Overlay.Image);
}
{
VkImageSubresourceRange fullSubRange = {VK_IMAGE_ASPECT_COLOR_BIT, 0, VK_REMAINING_MIP_LEVELS,
0, VK_REMAINING_ARRAY_LAYERS};
VkImageMemoryBarrier barrier = {
VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
NULL,
VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_TRANSFER_WRITE_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT,
VK_IMAGE_LAYOUT_UNDEFINED,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
VK_QUEUE_FAMILY_IGNORED,
VK_QUEUE_FAMILY_IGNORED,
Unwrap(m_Overlay.Image),
fullSubRange};
DoPipelineBarrier(cmd, 1, &barrier);
float black[4] = {};
vt->CmdClearColorImage(Unwrap(cmd), Unwrap(m_Overlay.Image), VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
(VkClearColorValue *)black, 1, &fullSubRange);
barrier.oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL;
barrier.newLayout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;
barrier.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
barrier.dstAccessMask =
VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT;
DoPipelineBarrier(cmd, 1, &barrier);
}
const ActionDescription *mainDraw = m_pDriver->GetAction(eventId);
const VulkanCreationInfo::Pipeline &pipeInfo =
m_pDriver->m_CreationInfo.m_Pipeline[state.graphics.pipeline];
// Secondary commands can't have render passes
if((mainDraw && !(mainDraw->flags & ActionFlags::Drawcall)) ||
!m_pDriver->m_Partial[WrappedVulkan::Primary].renderPassActive)
{
// don't do anything, no action capable of making overlays selected
float black[] = {0.0f, 0.0f, 0.0f, 0.0f};
VkImageMemoryBarrier barrier = {VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
NULL,
VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT,
VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
VK_QUEUE_FAMILY_IGNORED,
VK_QUEUE_FAMILY_IGNORED,
Unwrap(m_Overlay.Image),
subRange};
DoPipelineBarrier(cmd, 1, &barrier);
vt->CmdClearColorImage(Unwrap(cmd), Unwrap(m_Overlay.Image), VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
(VkClearColorValue *)black, 1, &subRange);
std::swap(barrier.oldLayout, barrier.newLayout);
std::swap(barrier.srcAccessMask, barrier.dstAccessMask);
barrier.dstAccessMask |= VK_ACCESS_COLOR_ATTACHMENT_READ_BIT;
DoPipelineBarrier(cmd, 1, &barrier);
}
else if(overlay == DebugOverlay::NaN || overlay == DebugOverlay::Clipping)
{
float black[] = {0.0f, 0.0f, 0.0f, 0.0f};
VkImageMemoryBarrier barrier = {VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
NULL,
VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT,
VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
VK_QUEUE_FAMILY_IGNORED,
VK_QUEUE_FAMILY_IGNORED,
Unwrap(m_Overlay.Image),
subRange};
DoPipelineBarrier(cmd, 1, &barrier);
vt->CmdClearColorImage(Unwrap(cmd), Unwrap(m_Overlay.Image), VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
(VkClearColorValue *)black, 1, &subRange);
std::swap(barrier.oldLayout, barrier.newLayout);
std::swap(barrier.srcAccessMask, barrier.dstAccessMask);
barrier.dstAccessMask |= VK_ACCESS_COLOR_ATTACHMENT_READ_BIT;
DoPipelineBarrier(cmd, 1, &barrier);
}
else if(overlay == DebugOverlay::Drawcall || overlay == DebugOverlay::Wireframe)
{
float highlightCol[] = {0.8f, 0.1f, 0.8f, 1.0f};
float bgclearCol[] = {0.0f, 0.0f, 0.0f, 0.5f};
if(overlay == DebugOverlay::Wireframe)
{
highlightCol[0] = 200 / 255.0f;
highlightCol[1] = 1.0f;
highlightCol[2] = 0.0f;
bgclearCol[0] = 200 / 255.0f;
bgclearCol[1] = 1.0f;
bgclearCol[2] = 0.0f;
bgclearCol[3] = 0.0f;
}
VkImageMemoryBarrier barrier = {VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
NULL,
VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT,
VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
VK_QUEUE_FAMILY_IGNORED,
VK_QUEUE_FAMILY_IGNORED,
Unwrap(m_Overlay.Image),
subRange};
DoPipelineBarrier(cmd, 1, &barrier);
vt->CmdClearColorImage(Unwrap(cmd), Unwrap(m_Overlay.Image), VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
(VkClearColorValue *)bgclearCol, 1, &subRange);
std::swap(barrier.oldLayout, barrier.newLayout);
std::swap(barrier.srcAccessMask, barrier.dstAccessMask);
barrier.dstAccessMask |= VK_ACCESS_COLOR_ATTACHMENT_READ_BIT;
DoPipelineBarrier(cmd, 1, &barrier);
if(!pipeInfo.rasterizerDiscardEnable)
{
vkr = vt->EndCommandBuffer(Unwrap(cmd));
CheckVkResult(vkr);
// backup state
VulkanRenderState prevstate = state;
// make patched shader
VkShaderModule mod = VK_NULL_HANDLE;
GetDebugManager()->PatchFixedColShader(mod, highlightCol);
// make patched pipeline
VkGraphicsPipelineCreateInfo pipeCreateInfo;
m_pDriver->GetShaderCache()->MakeGraphicsPipelineInfo(pipeCreateInfo,
prevstate.graphics.pipeline);
// disable all tests possible
VkPipelineDepthStencilStateCreateInfo *ds =
(VkPipelineDepthStencilStateCreateInfo *)pipeCreateInfo.pDepthStencilState;
ds->depthTestEnable = false;
ds->depthWriteEnable = false;
ds->stencilTestEnable = false;
ds->depthBoundsTestEnable = false;
VkPipelineRasterizationStateCreateInfo *rs =
(VkPipelineRasterizationStateCreateInfo *)pipeCreateInfo.pRasterizationState;
rs->cullMode = VK_CULL_MODE_NONE;
rs->rasterizerDiscardEnable = false;
// disable tests in dynamic state too
state.depthTestEnable = VK_FALSE;
state.depthWriteEnable = VK_FALSE;
state.depthCompareOp = VK_COMPARE_OP_ALWAYS;
state.stencilTestEnable = VK_FALSE;
state.depthBoundsTestEnable = VK_FALSE;
state.cullMode = VK_CULL_MODE_NONE;
// disable all discard rectangles
RemoveNextStruct(&pipeCreateInfo,
VK_STRUCTURE_TYPE_PIPELINE_DISCARD_RECTANGLE_STATE_CREATE_INFO_EXT);
if(m_pDriver->GetDeviceEnabledFeatures().depthClamp)
{
rs->depthClampEnable = true;
}
// disable line stipple
VkPipelineRasterizationLineStateCreateInfoEXT *lineRasterState =
(VkPipelineRasterizationLineStateCreateInfoEXT *)FindNextStruct(
rs, VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_LINE_STATE_CREATE_INFO_EXT);
if(lineRasterState)
{
lineRasterState->stippledLineEnable = VK_FALSE;
}
uint32_t patchedIndexCount = 0;
GPUBuffer patchedIB;
if(overlay == DebugOverlay::Wireframe)
{
rs->lineWidth = 1.0f;
if(mainDraw == NULL)
{
// do nothing
}
else if(m_pDriver->GetDeviceEnabledFeatures().fillModeNonSolid)
{
rs->polygonMode = VK_POLYGON_MODE_LINE;
}
else if(prevstate.primitiveTopology == VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST ||
prevstate.primitiveTopology == VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP ||
prevstate.primitiveTopology == VK_PRIMITIVE_TOPOLOGY_TRIANGLE_FAN ||
prevstate.primitiveTopology == VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST_WITH_ADJACENCY ||
prevstate.primitiveTopology == VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP_WITH_ADJACENCY)
{
// bad drivers (aka mobile) won't have non-solid fill mode, so we have to fall back to
// manually patching the index buffer and using a line list. This doesn't work with
// adjacency or patchlist topologies since those imply a vertex processing pipeline that
// requires a particular topology, or can't be implicitly converted to lines at input
// stage.
// It's unlikely those features will be used on said poor hw, so this should still catch
// most cases.
VkPipelineInputAssemblyStateCreateInfo *ia =
(VkPipelineInputAssemblyStateCreateInfo *)pipeCreateInfo.pInputAssemblyState;
ia->topology = VK_PRIMITIVE_TOPOLOGY_LINE_STRIP;
// thankfully, primitive restart is always supported! This makes the index buffer a bit
// more
// compact in the common cases where we don't need to repeat two indices for a triangle's
// three lines, instead we have a single restart index after each triangle.
ia->primitiveRestartEnable = true;
GetDebugManager()->PatchLineStripIndexBuffer(mainDraw, patchedIB, patchedIndexCount);
}
else
{
RDCWARN("Unable to draw wireframe overlay for %s topology draw via software patching",
ToStr(prevstate.primitiveTopology).c_str());
}
}
VkPipelineColorBlendStateCreateInfo *cb =
(VkPipelineColorBlendStateCreateInfo *)pipeCreateInfo.pColorBlendState;
cb->logicOpEnable = false;
cb->attachmentCount = 1; // only one colour attachment
for(uint32_t i = 0; i < cb->attachmentCount; i++)
{
VkPipelineColorBlendAttachmentState *att =
(VkPipelineColorBlendAttachmentState *)&cb->pAttachments[i];
att->blendEnable = false;
att->colorWriteMask = 0xf;
}
// set scissors to max for drawcall
if(overlay == DebugOverlay::Drawcall)
{
for(size_t i = 0; i < pipeCreateInfo.pViewportState->scissorCount; i++)
{
VkRect2D &sc = (VkRect2D &)pipeCreateInfo.pViewportState->pScissors[i];
sc.offset.x = 0;
sc.offset.y = 0;
sc.extent.width = 16384;
sc.extent.height = 16384;
}
}
// set our renderpass and shader
pipeCreateInfo.renderPass = m_Overlay.NoDepthRP;
pipeCreateInfo.subpass = 0;
// don't use dynamic rendering
RemoveNextStruct(&pipeCreateInfo, VK_STRUCTURE_TYPE_PIPELINE_RENDERING_CREATE_INFO);
bool found = false;
for(uint32_t i = 0; i < pipeCreateInfo.stageCount; i++)
{
VkPipelineShaderStageCreateInfo &sh =
(VkPipelineShaderStageCreateInfo &)pipeCreateInfo.pStages[i];
if(sh.stage == VK_SHADER_STAGE_FRAGMENT_BIT)
{
sh.module = mod;
sh.pName = "main";
found = true;
break;
}
}
if(!found)
{
// we know this is safe because it's pointing to a static array that's
// big enough for all shaders
VkPipelineShaderStageCreateInfo &sh =
(VkPipelineShaderStageCreateInfo &)pipeCreateInfo.pStages[pipeCreateInfo.stageCount++];
sh.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO;
sh.pNext = NULL;
sh.stage = VK_SHADER_STAGE_FRAGMENT_BIT;
sh.module = mod;
sh.pName = "main";
sh.pSpecializationInfo = NULL;
}
VkPipeline pipe = VK_NULL_HANDLE;
vkr = m_pDriver->vkCreateGraphicsPipelines(m_Device, VK_NULL_HANDLE, 1, &pipeCreateInfo, NULL,
&pipe);
CheckVkResult(vkr);
// modify state
state.SetRenderPass(GetResID(m_Overlay.NoDepthRP));
state.subpass = 0;
state.SetFramebuffer(m_pDriver, GetResID(m_Overlay.NoDepthFB));
state.subpassContents = VK_SUBPASS_CONTENTS_INLINE;
state.dynamicRendering.flags &= VK_RENDERING_CONTENTS_SECONDARY_COMMAND_BUFFERS_BIT;
state.graphics.pipeline = GetResID(pipe);
// set dynamic scissors in case pipeline was using them
if(overlay == DebugOverlay::Drawcall)
{
for(size_t i = 0; i < state.scissors.size(); i++)
{
state.scissors[i].offset.x = 0;
state.scissors[i].offset.y = 0;
state.scissors[i].extent.width = 16384;
state.scissors[i].extent.height = 16384;
}
}
if(overlay == DebugOverlay::Wireframe)
state.lineWidth = 1.0f;
if(overlay == DebugOverlay::Drawcall || overlay == DebugOverlay::Wireframe)
state.conditionalRendering.forceDisable = true;
if(patchedIndexCount == 0)
{
m_pDriver->ReplayLog(0, eventId, eReplay_OnlyDraw);
}
else
{
// if we patched the index buffer we need to manually play the draw with a higher index
// count
// and no index offset.
cmd = m_pDriver->GetNextCmd();
if(cmd == VK_NULL_HANDLE)
return ResourceId();
vkr = ObjDisp(cmd)->BeginCommandBuffer(Unwrap(cmd), &beginInfo);
CheckVkResult(vkr);
// do single draw
state.BeginRenderPassAndApplyState(m_pDriver, cmd, VulkanRenderState::BindGraphics, false);
ActionDescription action = *mainDraw;
action.numIndices = patchedIndexCount;
action.baseVertex = 0;
action.indexOffset = 0;
m_pDriver->ReplayDraw(cmd, action);
state.EndRenderPass(cmd);
vkr = ObjDisp(cmd)->EndCommandBuffer(Unwrap(cmd));
CheckVkResult(vkr);
}
// submit & flush so that we don't have to keep pipeline around for a while
m_pDriver->SubmitCmds();
m_pDriver->FlushQ();
cmd = m_pDriver->GetNextCmd();
if(cmd == VK_NULL_HANDLE)
return ResourceId();
vkr = vt->BeginCommandBuffer(Unwrap(cmd), &beginInfo);
CheckVkResult(vkr);
// restore state
state = prevstate;
patchedIB.Destroy();
m_pDriver->vkDestroyPipeline(m_Device, pipe, NULL);
m_pDriver->vkDestroyShaderModule(m_Device, mod, NULL);
}
}
else if(overlay == DebugOverlay::ViewportScissor)
{
// clear the whole image to opaque black. We'll overwite the render area with transparent black
// before rendering the viewport/scissors
float black[] = {0.0f, 0.0f, 0.0f, 0.0f};
VkImageMemoryBarrier barrier = {VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
NULL,
VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT,
VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
VK_QUEUE_FAMILY_IGNORED,
VK_QUEUE_FAMILY_IGNORED,
Unwrap(m_Overlay.Image),
subRange};
DoPipelineBarrier(cmd, 1, &barrier);
vt->CmdClearColorImage(Unwrap(cmd), Unwrap(m_Overlay.Image), VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
(VkClearColorValue *)black, 1, &subRange);
std::swap(barrier.oldLayout, barrier.newLayout);
std::swap(barrier.srcAccessMask, barrier.dstAccessMask);
barrier.dstAccessMask |= VK_ACCESS_COLOR_ATTACHMENT_READ_BIT;
DoPipelineBarrier(cmd, 1, &barrier);
black[3] = 0.0f;
if(!pipeInfo.rasterizerDiscardEnable)
{
vkr = vt->EndCommandBuffer(Unwrap(cmd));
CheckVkResult(vkr);
float highlightCol[] = {1.0f, 0.0f, 0.0f, 1.0f};
// backup state
VulkanRenderState prevstate = state;
// make patched shader
VkShaderModule mod[2] = {0};
VkPipeline pipe[2] = {0};
// first shader, no culling, writes red
GetDebugManager()->PatchFixedColShader(mod[0], highlightCol);
highlightCol[0] = 0.0f;
highlightCol[1] = 1.0f;
// second shader, normal culling, writes green
GetDebugManager()->PatchFixedColShader(mod[1], highlightCol);
// make patched pipeline
VkGraphicsPipelineCreateInfo pipeCreateInfo;
m_pDriver->GetShaderCache()->MakeGraphicsPipelineInfo(pipeCreateInfo,
prevstate.graphics.pipeline);
// disable all tests possible
VkPipelineDepthStencilStateCreateInfo *ds =
(VkPipelineDepthStencilStateCreateInfo *)pipeCreateInfo.pDepthStencilState;
ds->depthTestEnable = false;
ds->depthWriteEnable = false;
ds->stencilTestEnable = false;
ds->depthBoundsTestEnable = false;
VkPipelineRasterizationStateCreateInfo *rs =
(VkPipelineRasterizationStateCreateInfo *)pipeCreateInfo.pRasterizationState;
rs->cullMode = VK_CULL_MODE_NONE; // first render without any culling
rs->rasterizerDiscardEnable = false;
// disable tests in dynamic state too
state.depthTestEnable = VK_FALSE;
state.depthWriteEnable = VK_FALSE;
state.depthCompareOp = VK_COMPARE_OP_ALWAYS;
state.stencilTestEnable = VK_FALSE;
state.depthBoundsTestEnable = VK_FALSE;
state.cullMode = VK_CULL_MODE_NONE;
if(m_pDriver->GetDeviceEnabledFeatures().depthClamp)
rs->depthClampEnable = true;
VkPipelineColorBlendStateCreateInfo *cb =
(VkPipelineColorBlendStateCreateInfo *)pipeCreateInfo.pColorBlendState;
cb->logicOpEnable = false;
cb->attachmentCount = 1; // only one colour attachment
for(uint32_t i = 0; i < cb->attachmentCount; i++)
{
VkPipelineColorBlendAttachmentState *att =
(VkPipelineColorBlendAttachmentState *)&cb->pAttachments[i];
att->blendEnable = false;
att->colorWriteMask = 0xf;
}
// set our renderpass and shader
pipeCreateInfo.renderPass = m_Overlay.NoDepthRP;
pipeCreateInfo.subpass = 0;
// don't use dynamic rendering
RemoveNextStruct(&pipeCreateInfo, VK_STRUCTURE_TYPE_PIPELINE_RENDERING_CREATE_INFO);
VkPipelineShaderStageCreateInfo *fragShader = NULL;
for(uint32_t i = 0; i < pipeCreateInfo.stageCount; i++)
{
VkPipelineShaderStageCreateInfo &sh =
(VkPipelineShaderStageCreateInfo &)pipeCreateInfo.pStages[i];
if(sh.stage == VK_SHADER_STAGE_FRAGMENT_BIT)
{
sh.module = mod[0];
sh.pName = "main";
fragShader = &sh;
break;
}
}
if(fragShader == NULL)
{
// we know this is safe because it's pointing to a static array that's
// big enough for all shaders
VkPipelineShaderStageCreateInfo &sh =
(VkPipelineShaderStageCreateInfo &)pipeCreateInfo.pStages[pipeCreateInfo.stageCount++];
sh.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO;
sh.pNext = NULL;
sh.stage = VK_SHADER_STAGE_FRAGMENT_BIT;
sh.module = mod[0];
sh.pName = "main";
sh.pSpecializationInfo = NULL;
fragShader = &sh;
}
vkr = m_pDriver->vkCreateGraphicsPipelines(m_Device, VK_NULL_HANDLE, 1, &pipeCreateInfo, NULL,
&pipe[0]);
CheckVkResult(vkr);
fragShader->module = mod[1];
vkr = m_pDriver->vkCreateGraphicsPipelines(m_Device, VK_NULL_HANDLE, 1, &pipeCreateInfo, NULL,
&pipe[1]);
CheckVkResult(vkr);
// modify state
state.SetRenderPass(GetResID(m_Overlay.NoDepthRP));
state.subpass = 0;
state.SetFramebuffer(m_pDriver, GetResID(m_Overlay.NoDepthFB));
state.graphics.pipeline = GetResID(pipe[0]);
state.scissors = prevstate.scissors;
for(VkRect2D &sc : state.scissors)
{
sc.offset.x = 0;
sc.offset.y = 0;
sc.extent.width = 16384;
sc.extent.height = 16384;
}
m_pDriver->ReplayLog(0, eventId, eReplay_OnlyDraw);
state.graphics.pipeline = GetResID(pipe[1]);
state.scissors = prevstate.scissors;
m_pDriver->ReplayLog(0, eventId, eReplay_OnlyDraw);
// restore state
state = prevstate;
cmd = m_pDriver->GetNextCmd();
if(cmd == VK_NULL_HANDLE)
return ResourceId();
vkr = vt->BeginCommandBuffer(Unwrap(cmd), &beginInfo);
CheckVkResult(vkr);
{
VkClearValue clearval = {};
VkRenderPassBeginInfo rpbegin = {
VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
NULL,
Unwrap(m_Overlay.NoDepthRP),
Unwrap(m_Overlay.NoDepthFB),
state.renderArea,
1,
&clearval,
};
vt->CmdBeginRenderPass(Unwrap(cmd), &rpbegin, VK_SUBPASS_CONTENTS_INLINE);
VkViewport viewport = state.views[0];
vt->CmdSetViewport(Unwrap(cmd), 0, 1, &viewport);
uint32_t uboOffs = 0;
CheckerboardUBOData *ubo = (CheckerboardUBOData *)m_Overlay.m_CheckerUBO.Map(&uboOffs);
if(!ubo)
return ResourceId();
ubo->BorderWidth = 3;
ubo->CheckerSquareDimension = 16.0f;
// set primary/secondary to the same to 'disable' checkerboard
ubo->PrimaryColor = ubo->SecondaryColor = Vec4f(0.1f, 0.1f, 0.1f, 1.0f);
ubo->InnerColor = Vec4f(0.2f, 0.2f, 0.9f, 0.4f);
// set viewport rect
ubo->RectPosition = Vec2f(viewport.x, viewport.y);
ubo->RectSize = Vec2f(viewport.width, viewport.height);
if(m_pDriver->GetExtensions(GetRecord(m_Device)).ext_AMD_negative_viewport_height ||
m_pDriver->GetExtensions(GetRecord(m_Device)).ext_KHR_maintenance1)
{
ubo->RectSize.y = fabsf(viewport.height);
// VK_KHR_maintenance1 requires the position to be adjusted as well
if(m_pDriver->GetExtensions(GetRecord(m_Device)).ext_KHR_maintenance1 &&
viewport.height < 0.0f)
ubo->RectPosition.y += viewport.height;
}
m_Overlay.m_CheckerUBO.Unmap();
vt->CmdBindPipeline(Unwrap(cmd), VK_PIPELINE_BIND_POINT_GRAPHICS,
Unwrap(m_Overlay.m_CheckerF16Pipeline[SampleIndex(iminfo.samples)]));
vt->CmdBindDescriptorSets(Unwrap(cmd), VK_PIPELINE_BIND_POINT_GRAPHICS,
Unwrap(m_Overlay.m_CheckerPipeLayout), 0, 1,
UnwrapPtr(m_Overlay.m_CheckerDescSet), 1, &uboOffs);
vt->CmdDraw(Unwrap(cmd), 4, 1, 0, 0);
if(!state.scissors.empty())
{
Vec4f scissor((float)state.scissors[0].offset.x, (float)state.scissors[0].offset.y,
(float)state.scissors[0].extent.width,
(float)state.scissors[0].extent.height);
ubo = (CheckerboardUBOData *)m_Overlay.m_CheckerUBO.Map(&uboOffs);
if(!ubo)
return ResourceId();
ubo->BorderWidth = 3;
ubo->CheckerSquareDimension = 16.0f;
// black/white checkered border
ubo->PrimaryColor = Vec4f(1.0f, 1.0f, 1.0f, 1.0f);
ubo->SecondaryColor = Vec4f(0.0f, 0.0f, 0.0f, 1.0f);
// nothing at all inside
ubo->InnerColor = Vec4f(0.0f, 0.0f, 0.0f, 0.0f);
ubo->RectPosition = Vec2f(scissor.x, scissor.y);
ubo->RectSize = Vec2f(scissor.z, scissor.w);
m_Overlay.m_CheckerUBO.Unmap();
viewport.x = scissor.x;
viewport.y = scissor.y;
viewport.width = scissor.z;
viewport.height = scissor.w;
vt->CmdSetViewport(Unwrap(cmd), 0, 1, &viewport);
vt->CmdBindDescriptorSets(Unwrap(cmd), VK_PIPELINE_BIND_POINT_GRAPHICS,
Unwrap(m_Overlay.m_CheckerPipeLayout), 0, 1,
UnwrapPtr(m_Overlay.m_CheckerDescSet), 1, &uboOffs);
vt->CmdDraw(Unwrap(cmd), 4, 1, 0, 0);
}
vt->CmdEndRenderPass(Unwrap(cmd));
}
vkr = vt->EndCommandBuffer(Unwrap(cmd));
CheckVkResult(vkr);
// submit & flush so that we don't have to keep pipeline around for a while
m_pDriver->SubmitCmds();
m_pDriver->FlushQ();
cmd = m_pDriver->GetNextCmd();
if(cmd == VK_NULL_HANDLE)
return ResourceId();
vkr = vt->BeginCommandBuffer(Unwrap(cmd), &beginInfo);
CheckVkResult(vkr);
for(int i = 0; i < 2; i++)
{
m_pDriver->vkDestroyPipeline(m_Device, pipe[i], NULL);
m_pDriver->vkDestroyShaderModule(m_Device, mod[i], NULL);
}
}
}
else if(overlay == DebugOverlay::BackfaceCull)
{
float highlightCol[] = {0.0f, 0.0f, 0.0f, 0.0f};
VkImageMemoryBarrier barrier = {VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
NULL,
VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT,
VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
VK_QUEUE_FAMILY_IGNORED,
VK_QUEUE_FAMILY_IGNORED,
Unwrap(m_Overlay.Image),
subRange};
DoPipelineBarrier(cmd, 1, &barrier);
vt->CmdClearColorImage(Unwrap(cmd), Unwrap(m_Overlay.Image), VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
(VkClearColorValue *)highlightCol, 1, &subRange);
std::swap(barrier.oldLayout, barrier.newLayout);
std::swap(barrier.srcAccessMask, barrier.dstAccessMask);
barrier.dstAccessMask |= VK_ACCESS_COLOR_ATTACHMENT_READ_BIT;
DoPipelineBarrier(cmd, 1, &barrier);
highlightCol[0] = 1.0f;
highlightCol[1] = 0.0f;
highlightCol[3] = 1.0f;
if(!pipeInfo.rasterizerDiscardEnable)
{
vkr = vt->EndCommandBuffer(Unwrap(cmd));
CheckVkResult(vkr);
// backup state
VulkanRenderState prevstate = state;
// make patched shader
VkShaderModule mod[2] = {0};
VkPipeline pipe[2] = {0};
// first shader, no culling, writes red
GetDebugManager()->PatchFixedColShader(mod[0], highlightCol);
highlightCol[0] = 0.0f;
highlightCol[1] = 1.0f;
// second shader, normal culling, writes green
GetDebugManager()->PatchFixedColShader(mod[1], highlightCol);
// make patched pipeline
VkGraphicsPipelineCreateInfo pipeCreateInfo;
m_pDriver->GetShaderCache()->MakeGraphicsPipelineInfo(pipeCreateInfo,
prevstate.graphics.pipeline);
// disable all tests possible
VkPipelineDepthStencilStateCreateInfo *ds =
(VkPipelineDepthStencilStateCreateInfo *)pipeCreateInfo.pDepthStencilState;
ds->depthTestEnable = false;
ds->depthWriteEnable = false;
ds->stencilTestEnable = false;
ds->depthBoundsTestEnable = false;
VkPipelineRasterizationStateCreateInfo *rs =
(VkPipelineRasterizationStateCreateInfo *)pipeCreateInfo.pRasterizationState;
VkCullModeFlags origCullMode = prevstate.cullMode;
rs->cullMode = VK_CULL_MODE_NONE; // first render without any culling
rs->rasterizerDiscardEnable = false;
if(m_pDriver->GetDeviceEnabledFeatures().depthClamp)
rs->depthClampEnable = true;
// disable tests in dynamic state too
state.depthTestEnable = VK_FALSE;
state.depthWriteEnable = VK_FALSE;
state.depthCompareOp = VK_COMPARE_OP_ALWAYS;
state.stencilTestEnable = VK_FALSE;
state.depthBoundsTestEnable = VK_FALSE;
state.cullMode = VK_CULL_MODE_NONE;
VkPipelineColorBlendStateCreateInfo *cb =
(VkPipelineColorBlendStateCreateInfo *)pipeCreateInfo.pColorBlendState;
cb->logicOpEnable = false;
cb->attachmentCount = 1; // only one colour attachment
for(uint32_t i = 0; i < cb->attachmentCount; i++)
{
VkPipelineColorBlendAttachmentState *att =
(VkPipelineColorBlendAttachmentState *)&cb->pAttachments[i];
att->blendEnable = false;
att->colorWriteMask = 0xf;
}
// set our renderpass and shader
pipeCreateInfo.renderPass = m_Overlay.NoDepthRP;
pipeCreateInfo.subpass = 0;
// don't use dynamic rendering
RemoveNextStruct(&pipeCreateInfo, VK_STRUCTURE_TYPE_PIPELINE_RENDERING_CREATE_INFO);
VkPipelineShaderStageCreateInfo *fragShader = NULL;
for(uint32_t i = 0; i < pipeCreateInfo.stageCount; i++)
{
VkPipelineShaderStageCreateInfo &sh =
(VkPipelineShaderStageCreateInfo &)pipeCreateInfo.pStages[i];
if(sh.stage == VK_SHADER_STAGE_FRAGMENT_BIT)
{
sh.module = mod[0];
sh.pName = "main";
fragShader = &sh;
break;
}
}
if(fragShader == NULL)
{
// we know this is safe because it's pointing to a static array that's
// big enough for all shaders
VkPipelineShaderStageCreateInfo &sh =
(VkPipelineShaderStageCreateInfo &)pipeCreateInfo.pStages[pipeCreateInfo.stageCount++];
sh.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO;
sh.pNext = NULL;
sh.stage = VK_SHADER_STAGE_FRAGMENT_BIT;
sh.module = mod[0];
sh.pName = "main";
sh.pSpecializationInfo = NULL;
fragShader = &sh;
}
vkr = m_pDriver->vkCreateGraphicsPipelines(m_Device, VK_NULL_HANDLE, 1, &pipeCreateInfo, NULL,
&pipe[0]);
CheckVkResult(vkr);
fragShader->module = mod[1];
rs->cullMode = origCullMode;
vkr = m_pDriver->vkCreateGraphicsPipelines(m_Device, VK_NULL_HANDLE, 1, &pipeCreateInfo, NULL,
&pipe[1]);
CheckVkResult(vkr);
// modify state
state.SetRenderPass(GetResID(m_Overlay.NoDepthRP));
state.subpass = 0;
state.SetFramebuffer(m_pDriver, GetResID(m_Overlay.NoDepthFB));
state.graphics.pipeline = GetResID(pipe[0]);
m_pDriver->ReplayLog(0, eventId, eReplay_OnlyDraw);
state.graphics.pipeline = GetResID(pipe[1]);
state.cullMode = origCullMode;
m_pDriver->ReplayLog(0, eventId, eReplay_OnlyDraw);
// submit & flush so that we don't have to keep pipeline around for a while
m_pDriver->SubmitCmds();
m_pDriver->FlushQ();
cmd = m_pDriver->GetNextCmd();
if(cmd == VK_NULL_HANDLE)
return ResourceId();
vkr = vt->BeginCommandBuffer(Unwrap(cmd), &beginInfo);
CheckVkResult(vkr);
// restore state
state = prevstate;
for(int i = 0; i < 2; i++)
{
m_pDriver->vkDestroyPipeline(m_Device, pipe[i], NULL);
m_pDriver->vkDestroyShaderModule(m_Device, mod[i], NULL);
}
}
}
else if(overlay == DebugOverlay::Depth || overlay == DebugOverlay::Stencil)
{
float highlightCol[] = {0.0f, 0.0f, 0.0f, 0.0f};
VkImageMemoryBarrier barrier = {VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
NULL,
VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT,
VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
VK_QUEUE_FAMILY_IGNORED,
VK_QUEUE_FAMILY_IGNORED,
Unwrap(m_Overlay.Image),
subRange};
DoPipelineBarrier(cmd, 1, &barrier);
vt->CmdClearColorImage(Unwrap(cmd), Unwrap(m_Overlay.Image), VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
(VkClearColorValue *)highlightCol, 1, &subRange);
std::swap(barrier.oldLayout, barrier.newLayout);
std::swap(barrier.srcAccessMask, barrier.dstAccessMask);
barrier.dstAccessMask |= VK_ACCESS_COLOR_ATTACHMENT_READ_BIT;
DoPipelineBarrier(cmd, 1, &barrier);
if(!pipeInfo.rasterizerDiscardEnable)
{
vkr = vt->EndCommandBuffer(Unwrap(cmd));
CheckVkResult(vkr);
VkFramebuffer depthFB = VK_NULL_HANDLE;
VkRenderPass depthRP = VK_NULL_HANDLE;
VulkanCreationInfo &createinfo = m_pDriver->m_CreationInfo;
ResourceId depthStencilView;
if(state.dynamicRendering.active)
{
depthStencilView = GetResID(state.dynamicRendering.depth.imageView);
if(depthStencilView == ResourceId())
depthStencilView = GetResID(state.dynamicRendering.stencil.imageView);
}
else
{
RDCASSERT(state.subpass < createinfo.m_RenderPass[state.GetRenderPass()].subpasses.size());
int32_t dsIdx =
createinfo.m_RenderPass[state.GetRenderPass()].subpasses[state.subpass].depthstencilAttachment;
// make a renderpass and framebuffer for rendering to overlay color and using
// depth buffer from the orignial render
if(dsIdx >= 0 &&
dsIdx < (int32_t)createinfo.m_Framebuffer[state.GetFramebuffer()].attachments.size())
{
depthStencilView = state.GetFramebufferAttachments()[dsIdx];
}
}
if(depthStencilView != ResourceId())
{
VkAttachmentDescription attDescs[] = {
{0, overlayFormat, VK_SAMPLE_COUNT_1_BIT, VK_ATTACHMENT_LOAD_OP_LOAD,
VK_ATTACHMENT_STORE_OP_STORE, VK_ATTACHMENT_LOAD_OP_DONT_CARE,
VK_ATTACHMENT_STORE_OP_DONT_CARE, VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL,
VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL},
{0, VK_FORMAT_UNDEFINED, VK_SAMPLE_COUNT_1_BIT, // will patch this just below
VK_ATTACHMENT_LOAD_OP_LOAD, VK_ATTACHMENT_STORE_OP_STORE, VK_ATTACHMENT_LOAD_OP_LOAD,
VK_ATTACHMENT_STORE_OP_STORE, VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL,
VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL},
};
VulkanCreationInfo::ImageView &depthViewInfo = createinfo.m_ImageView[depthStencilView];
ResourceId depthIm = depthViewInfo.image;
VulkanCreationInfo::Image &depthImageInfo = createinfo.m_Image[depthIm];
attDescs[1].format = depthImageInfo.format;
attDescs[0].samples = attDescs[1].samples = iminfo.samples;
{
LockedConstImageStateRef imState = m_pDriver->FindConstImageState(depthIm);
if(imState)
{
// find the state that overlaps the view's subresource range start. We assume all
// subresources are correctly in the same state (as they should be) so we just need to
// find the first match.
auto it = imState->subresourceStates.RangeBegin(depthViewInfo.range);
if(it != imState->subresourceStates.end())
attDescs[1].initialLayout = attDescs[1].finalLayout = it->state().newLayout;
}
}
VkAttachmentReference colRef = {0, VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL};
VkAttachmentReference dsRef = {1, attDescs[1].initialLayout};
VkSubpassDescription subp = {
0, VK_PIPELINE_BIND_POINT_GRAPHICS,
0, NULL, // inputs
1, &colRef, // color
NULL, // resolve
&dsRef, // depth-stencil
0, NULL, // preserve
};
VkRenderPassCreateInfo rpinfo = {
VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO,
NULL,
0,
2,
attDescs,
1,
&subp,
0,
NULL, // dependencies
};
if(multiviewMask > 0)
rpinfo.pNext = &multiviewRP;
vkr = m_pDriver->vkCreateRenderPass(m_Device, &rpinfo, NULL, &depthRP);
CheckVkResult(vkr);
VkImageView views[] = {
m_Overlay.ImageView,
m_pDriver->GetResourceManager()->GetCurrentHandle<VkImageView>(depthStencilView),
};
// Create framebuffer rendering just to overlay image, no depth
VkFramebufferCreateInfo fbinfo = {
VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO,
NULL,
0,
depthRP,
2,
views,
(uint32_t)m_Overlay.ImageDim.width,
(uint32_t)m_Overlay.ImageDim.height,
1,
};
vkr = m_pDriver->vkCreateFramebuffer(m_Device, &fbinfo, NULL, &depthFB);
CheckVkResult(vkr);
}
// if depthRP is NULL, so is depthFB, and it means no depth buffer was
// bound, so we just render green.
highlightCol[0] = 1.0f;
highlightCol[1] = 0.0f;
highlightCol[3] = 1.0f;
// backup state
VulkanRenderState prevstate = state;
// make patched shader
VkShaderModule failmod = {}, passmod = {};
VkPipeline failpipe = {}, passpipe = {};
// first shader, no depth/stencil testing, writes red
GetDebugManager()->PatchFixedColShader(failmod, highlightCol);
highlightCol[0] = 0.0f;
highlightCol[1] = 1.0f;
// second shader, enabled depth/stencil testing, writes green
GetDebugManager()->PatchFixedColShader(passmod, highlightCol);
// make patched pipeline
VkGraphicsPipelineCreateInfo pipeCreateInfo;
m_pDriver->GetShaderCache()->MakeGraphicsPipelineInfo(pipeCreateInfo,
prevstate.graphics.pipeline);
// disable all tests possible
VkPipelineDepthStencilStateCreateInfo *ds =
(VkPipelineDepthStencilStateCreateInfo *)pipeCreateInfo.pDepthStencilState;
VkBool32 origDepthTest = prevstate.depthTestEnable;
ds->depthTestEnable = false;
ds->depthWriteEnable = false;
VkBool32 origStencilTest = prevstate.stencilTestEnable;
ds->stencilTestEnable = false;
ds->depthBoundsTestEnable = false;
VkPipelineColorBlendStateCreateInfo *cb =
(VkPipelineColorBlendStateCreateInfo *)pipeCreateInfo.pColorBlendState;
cb->logicOpEnable = false;
cb->attachmentCount = 1; // only one colour attachment
for(uint32_t i = 0; i < cb->attachmentCount; i++)
{
VkPipelineColorBlendAttachmentState *att =
(VkPipelineColorBlendAttachmentState *)&cb->pAttachments[i];
att->blendEnable = false;
att->colorWriteMask = 0xf;
}
// subpass 0 in either render pass
pipeCreateInfo.subpass = 0;
VkPipelineShaderStageCreateInfo *fragShader = NULL;
for(uint32_t i = 0; i < pipeCreateInfo.stageCount; i++)
{
VkPipelineShaderStageCreateInfo &sh =
(VkPipelineShaderStageCreateInfo &)pipeCreateInfo.pStages[i];
if(sh.stage == VK_SHADER_STAGE_FRAGMENT_BIT)
{
sh.pName = "main";
fragShader = &sh;
break;
}
}
if(fragShader == NULL)
{
// we know this is safe because it's pointing to a static array that's
// big enough for all shaders
VkPipelineShaderStageCreateInfo &sh =
(VkPipelineShaderStageCreateInfo &)pipeCreateInfo.pStages[pipeCreateInfo.stageCount++];
sh.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO;
sh.pNext = NULL;
sh.stage = VK_SHADER_STAGE_FRAGMENT_BIT;
sh.pName = "main";
sh.pSpecializationInfo = NULL;
fragShader = &sh;
}
fragShader->module = passmod;
if(depthRP != VK_NULL_HANDLE)
{
if(overlay == DebugOverlay::Depth)
ds->depthTestEnable = origDepthTest;
else
ds->stencilTestEnable = origStencilTest;
pipeCreateInfo.renderPass = depthRP;
}
else
{
pipeCreateInfo.renderPass = m_Overlay.NoDepthRP;
}
// don't use dynamic rendering
RemoveNextStruct(&pipeCreateInfo, VK_STRUCTURE_TYPE_PIPELINE_RENDERING_CREATE_INFO);
vkr = m_pDriver->vkCreateGraphicsPipelines(m_Device, VK_NULL_HANDLE, 1, &pipeCreateInfo, NULL,
&passpipe);
CheckVkResult(vkr);
fragShader->module = failmod;
// set our renderpass and shader
pipeCreateInfo.renderPass = m_Overlay.NoDepthRP;
// disable culling/discard and enable depth clamp. That way we show any failures due to these
VkPipelineRasterizationStateCreateInfo *rs =
(VkPipelineRasterizationStateCreateInfo *)pipeCreateInfo.pRasterizationState;
rs->cullMode = VK_CULL_MODE_NONE;
rs->rasterizerDiscardEnable = false;
if(m_pDriver->GetDeviceEnabledFeatures().depthClamp)
rs->depthClampEnable = true;
vkr = m_pDriver->vkCreateGraphicsPipelines(m_Device, VK_NULL_HANDLE, 1, &pipeCreateInfo, NULL,
&failpipe);
CheckVkResult(vkr);
// modify state
state.SetRenderPass(GetResID(m_Overlay.NoDepthRP));
state.subpass = 0;
state.SetFramebuffer(m_pDriver, GetResID(m_Overlay.NoDepthFB));
state.graphics.pipeline = GetResID(failpipe);
// disable tests in dynamic state too
state.depthTestEnable = VK_FALSE;
state.depthWriteEnable = VK_FALSE;
state.stencilTestEnable = VK_FALSE;
state.depthBoundsTestEnable = VK_FALSE;
state.cullMode = VK_CULL_MODE_NONE;
m_pDriver->ReplayLog(0, eventId, eReplay_OnlyDraw);
state.graphics.pipeline = GetResID(passpipe);
if(depthRP != VK_NULL_HANDLE)
{
state.SetRenderPass(GetResID(depthRP));
state.SetFramebuffer(m_pDriver, GetResID(depthFB));
}
if(overlay == DebugOverlay::Depth)
state.depthTestEnable = origDepthTest;
else
state.stencilTestEnable = origStencilTest;
m_pDriver->ReplayLog(0, eventId, eReplay_OnlyDraw);
// submit & flush so that we don't have to keep pipeline around for a while
m_pDriver->SubmitCmds();
m_pDriver->FlushQ();
cmd = m_pDriver->GetNextCmd();
if(cmd == VK_NULL_HANDLE)
return ResourceId();
vkr = vt->BeginCommandBuffer(Unwrap(cmd), &beginInfo);
CheckVkResult(vkr);
// restore state
state = prevstate;
m_pDriver->vkDestroyPipeline(m_Device, failpipe, NULL);
m_pDriver->vkDestroyShaderModule(m_Device, failmod, NULL);
m_pDriver->vkDestroyPipeline(m_Device, passpipe, NULL);
m_pDriver->vkDestroyShaderModule(m_Device, passmod, NULL);
if(depthRP != VK_NULL_HANDLE)
{
m_pDriver->vkDestroyRenderPass(m_Device, depthRP, NULL);
m_pDriver->vkDestroyFramebuffer(m_Device, depthFB, NULL);
}
}
}
else if(overlay == DebugOverlay::ClearBeforeDraw || overlay == DebugOverlay::ClearBeforePass)
{
// clear the overlay image itself
float black[] = {0.0f, 0.0f, 0.0f, 0.0f};
VkImageMemoryBarrier barrier = {VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
NULL,
VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT,
VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
VK_QUEUE_FAMILY_IGNORED,
VK_QUEUE_FAMILY_IGNORED,
Unwrap(m_Overlay.Image),
subRange};
DoPipelineBarrier(cmd, 1, &barrier);
vt->CmdClearColorImage(Unwrap(cmd), Unwrap(m_Overlay.Image), VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
(VkClearColorValue *)black, 1, &subRange);
std::swap(barrier.oldLayout, barrier.newLayout);
std::swap(barrier.srcAccessMask, barrier.dstAccessMask);
barrier.dstAccessMask |= VK_ACCESS_COLOR_ATTACHMENT_READ_BIT;
DoPipelineBarrier(cmd, 1, &barrier);
rdcarray<uint32_t> events = passEvents;
if(overlay == DebugOverlay::ClearBeforeDraw)
events.clear();
events.push_back(eventId);
{
vkr = vt->EndCommandBuffer(Unwrap(cmd));
CheckVkResult(vkr);
#if ENABLED(SINGLE_FLUSH_VALIDATE)
m_pDriver->SubmitCmds();
#endif
size_t startEvent = 0;
// this is a bit of a hack, but it's the simplest fix without refactoring the ClearBefore..
// overlay to copy the result off into the overlay texture (which has its own issues with
// alpha channels since the overlay expects to be blended). If we are selecting a new event
// with ClearBeforeDraw enabled then we do this here, and then later the pipeline is fetched
// and that triggers the bindless feedback - which will reset the state and undo our clear. If
// we force it to be cached here then it won't mess things up later.
FetchShaderFeedback(eventId);
// if we're ClearBeforePass the first event will be a vkBeginRenderPass.
// if there are any other events, we need to play up to right before them
// so that we have all the render state set up to do
// BeginRenderPassAndApplyState and a clear. If it's just the begin, we
// just play including it, do the clear, then we won't replay anything
// in the loop below
if(overlay == DebugOverlay::ClearBeforePass)
{
const ActionDescription *action = m_pDriver->GetAction(events[0]);
if(action && action->flags & ActionFlags::BeginPass)
{
if(events.size() == 1)
{
m_pDriver->ReplayLog(0, events[0], eReplay_Full);
}
else
{
startEvent = 1;
m_pDriver->ReplayLog(0, events[1], eReplay_WithoutDraw);
}
}
}
else
{
m_pDriver->ReplayLog(0, events[0], eReplay_WithoutDraw);
}
cmd = m_pDriver->GetNextCmd();
if(cmd == VK_NULL_HANDLE)
return ResourceId();
vkr = vt->BeginCommandBuffer(Unwrap(cmd), &beginInfo);
CheckVkResult(vkr);
state.BeginRenderPassAndApplyState(m_pDriver, cmd, VulkanRenderState::BindGraphics, false);
VkClearAttachment clearatt = {VK_IMAGE_ASPECT_COLOR_BIT, 0, {}};
memcpy(clearatt.clearValue.color.float32, &clearCol, sizeof(clearatt.clearValue.color.float32));
rdcarray<VkClearAttachment> atts;
if(state.dynamicRendering.active)
{
for(size_t i = 0; i < state.dynamicRendering.color.size(); i++)
{
if(state.dynamicRendering.color[i].imageView == VK_NULL_HANDLE)
continue;
clearatt.colorAttachment = (uint32_t)i;
atts.push_back(clearatt);
}
}
else
{
VulkanCreationInfo::RenderPass &rp =
m_pDriver->m_CreationInfo.m_RenderPass[state.GetRenderPass()];
for(size_t i = 0; i < rp.subpasses[state.subpass].colorAttachments.size(); i++)
{
clearatt.colorAttachment = (uint32_t)i;
atts.push_back(clearatt);
}
}
// Try to clear depth as well, to help debug shadow rendering
if(state.graphics.pipeline != ResourceId() && IsDepthOrStencilFormat(iminfo.format))
{
VkCompareOp depthCompareOp = state.depthCompareOp;
// If the depth func is equal or not equal, don't clear at all since the output would be
// altered in an way that would cause replay to produce mostly incorrect results.
// Similarly, skip if the depth func is always, as we'd have a 50% chance of guessing the
// wrong clear value.
if(depthCompareOp != VK_COMPARE_OP_EQUAL && depthCompareOp != VK_COMPARE_OP_NOT_EQUAL &&
depthCompareOp != VK_COMPARE_OP_ALWAYS)
{
// If the depth func is less or less equal, clear to 1 instead of 0
bool depthFuncLess =
depthCompareOp == VK_COMPARE_OP_LESS || depthCompareOp == VK_COMPARE_OP_LESS_OR_EQUAL;
float depthClear = depthFuncLess ? 1.0f : 0.0f;
VkClearAttachment clearDepthAtt = {
VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT, 0, {}};
clearDepthAtt.clearValue.depthStencil.depth = depthClear;
clearDepthAtt.clearValue.depthStencil.stencil = 0;
atts.push_back(clearDepthAtt);
}
}
VkClearRect rect = {
state.renderArea, 0, 1,
};
vt->CmdClearAttachments(Unwrap(cmd), (uint32_t)atts.size(), &atts[0], 1, &rect);
state.EndRenderPass(cmd);
vkr = vt->EndCommandBuffer(Unwrap(cmd));
CheckVkResult(vkr);
for(size_t i = startEvent; i < events.size(); i++)
{
m_pDriver->ReplayLog(events[i], events[i], eReplay_OnlyDraw);
if(overlay == DebugOverlay::ClearBeforePass && i + 1 < events.size())
m_pDriver->ReplayLog(events[i] + 1, events[i + 1], eReplay_WithoutDraw);
}
cmd = m_pDriver->GetNextCmd();
if(cmd == VK_NULL_HANDLE)
return ResourceId();
vkr = vt->BeginCommandBuffer(Unwrap(cmd), &beginInfo);
CheckVkResult(vkr);
}
}
else if(overlay == DebugOverlay::QuadOverdrawPass || overlay == DebugOverlay::QuadOverdrawDraw)
{
if(m_Overlay.m_QuadResolvePipeline[0] != VK_NULL_HANDLE && !pipeInfo.rasterizerDiscardEnable)
{
VulkanRenderState prevstate = state;
SCOPED_TIMER("Quad Overdraw");
float black[] = {0.0f, 0.0f, 0.0f, 0.0f};
VkImageMemoryBarrier barrier = {VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
NULL,
VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT,
VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
VK_QUEUE_FAMILY_IGNORED,
VK_QUEUE_FAMILY_IGNORED,
Unwrap(m_Overlay.Image),
subRange};
DoPipelineBarrier(cmd, 1, &barrier);
vt->CmdClearColorImage(Unwrap(cmd), Unwrap(m_Overlay.Image),
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, (VkClearColorValue *)black, 1,
&subRange);
std::swap(barrier.oldLayout, barrier.newLayout);
std::swap(barrier.srcAccessMask, barrier.dstAccessMask);
barrier.dstAccessMask |= VK_ACCESS_COLOR_ATTACHMENT_READ_BIT;
DoPipelineBarrier(cmd, 1, &barrier);
rdcarray<uint32_t> events = passEvents;
if(overlay == DebugOverlay::QuadOverdrawDraw)
events.clear();
events.push_back(eventId);
// if we're rendering the whole pass, and the first action is a BeginRenderPass, don't include
// it in the list. We want to start by replaying into the renderpass so that we have the
// correct state being applied.
if(overlay == DebugOverlay::QuadOverdrawPass)
{
const ActionDescription *action = m_pDriver->GetAction(events[0]);
if(action->flags & ActionFlags::BeginPass)
events.erase(0);
}
VkImage quadImg;
VkDeviceMemory quadImgMem;
VkImageView quadImgView;
VkImageCreateInfo imInfo = {
VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
NULL,
0,
VK_IMAGE_TYPE_2D,
VK_FORMAT_R32_UINT,
{RDCMAX(1U, m_Overlay.ImageDim.width >> 1), RDCMAX(1U, m_Overlay.ImageDim.height >> 1), 1},
1,
4,
VK_SAMPLE_COUNT_1_BIT,
VK_IMAGE_TILING_OPTIMAL,
VK_IMAGE_USAGE_TRANSFER_DST_BIT | VK_IMAGE_USAGE_STORAGE_BIT | VK_IMAGE_USAGE_SAMPLED_BIT,
VK_SHARING_MODE_EXCLUSIVE,
0,
NULL,
VK_IMAGE_LAYOUT_UNDEFINED,
};
vkr = m_pDriver->vkCreateImage(m_Device, &imInfo, NULL, &quadImg);
CheckVkResult(vkr);
NameVulkanObject(quadImg, "m_Overlay.quadImg");
VkMemoryRequirements mrq = {0};
m_pDriver->vkGetImageMemoryRequirements(m_Device, quadImg, &mrq);
VkMemoryAllocateInfo allocInfo = {
VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO, NULL, mrq.size,
m_pDriver->GetGPULocalMemoryIndex(mrq.memoryTypeBits),
};
vkr = m_pDriver->vkAllocateMemory(m_Device, &allocInfo, NULL, &quadImgMem);
CheckVkResult(vkr);
if(vkr != VK_SUCCESS)
return ResourceId();
vkr = m_pDriver->vkBindImageMemory(m_Device, quadImg, quadImgMem, 0);
CheckVkResult(vkr);
VkImageViewCreateInfo viewinfo = {
VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
NULL,
0,
quadImg,
VK_IMAGE_VIEW_TYPE_2D_ARRAY,
VK_FORMAT_R32_UINT,
{VK_COMPONENT_SWIZZLE_IDENTITY, VK_COMPONENT_SWIZZLE_IDENTITY,
VK_COMPONENT_SWIZZLE_IDENTITY, VK_COMPONENT_SWIZZLE_IDENTITY},
{VK_IMAGE_ASPECT_COLOR_BIT, 0, 1, 0, 4},
};
vkr = m_pDriver->vkCreateImageView(m_Device, &viewinfo, NULL, &quadImgView);
CheckVkResult(vkr);
// update descriptor to point to our R32 result image
VkDescriptorImageInfo imdesc = {0};
imdesc.imageLayout = VK_IMAGE_LAYOUT_GENERAL;
imdesc.sampler = VK_NULL_HANDLE;
imdesc.imageView = Unwrap(quadImgView);
VkWriteDescriptorSet write = {VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
NULL,
Unwrap(m_Overlay.m_QuadDescSet),
0,
0,
1,
VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
&imdesc,
NULL,
NULL};
vt->UpdateDescriptorSets(Unwrap(m_Device), 1, &write, 0, NULL);
VkImageMemoryBarrier quadImBarrier = {
VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
NULL,
0,
VK_ACCESS_TRANSFER_WRITE_BIT,
VK_IMAGE_LAYOUT_UNDEFINED,
VK_IMAGE_LAYOUT_GENERAL,
VK_QUEUE_FAMILY_IGNORED,
VK_QUEUE_FAMILY_IGNORED,
Unwrap(quadImg),
{VK_IMAGE_ASPECT_COLOR_BIT, 0, 1, 0, 4},
};
// clear all to black
DoPipelineBarrier(cmd, 1, &quadImBarrier);
vt->CmdClearColorImage(Unwrap(cmd), Unwrap(quadImg), VK_IMAGE_LAYOUT_GENERAL,
(VkClearColorValue *)&black, 1, &quadImBarrier.subresourceRange);
quadImBarrier.srcAccessMask = quadImBarrier.dstAccessMask;
quadImBarrier.oldLayout = quadImBarrier.newLayout;
quadImBarrier.dstAccessMask = VK_ACCESS_SHADER_WRITE_BIT;
// set to general layout, for load/store operations
DoPipelineBarrier(cmd, 1, &quadImBarrier);
VkMemoryBarrier memBarrier = {
VK_STRUCTURE_TYPE_MEMORY_BARRIER, NULL, VK_ACCESS_ALL_WRITE_BITS, VK_ACCESS_ALL_READ_BITS,
};
DoPipelineBarrier(cmd, 1, &memBarrier);
// end this cmd buffer so the image is in the right state for the next part
vkr = vt->EndCommandBuffer(Unwrap(cmd));
CheckVkResult(vkr);
#if ENABLED(SINGLE_FLUSH_VALIDATE)
m_pDriver->SubmitCmds();
#endif
m_pDriver->ReplayLog(0, events[0], eReplay_WithoutDraw);
{
// declare callback struct here
VulkanQuadOverdrawCallback cb(m_pDriver, m_Overlay.m_QuadDescSetLayout,
m_Overlay.m_QuadDescSet, events);
m_pDriver->ReplayLog(events.front(), events.back(), eReplay_Full);
// resolve pass
{
cmd = m_pDriver->GetNextCmd();
if(cmd == VK_NULL_HANDLE)
return ResourceId();
vkr = vt->BeginCommandBuffer(Unwrap(cmd), &beginInfo);
CheckVkResult(vkr);
quadImBarrier.srcAccessMask = quadImBarrier.dstAccessMask;
quadImBarrier.oldLayout = quadImBarrier.newLayout;
quadImBarrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT;
// wait for writing to finish
DoPipelineBarrier(cmd, 1, &quadImBarrier);
VkClearValue clearval = {};
VkRenderPassBeginInfo rpbegin = {
VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
NULL,
Unwrap(m_Overlay.NoDepthRP),
Unwrap(m_Overlay.NoDepthFB),
state.renderArea,
1,
&clearval,
};
vt->CmdBeginRenderPass(Unwrap(cmd), &rpbegin, VK_SUBPASS_CONTENTS_INLINE);
vt->CmdBindPipeline(Unwrap(cmd), VK_PIPELINE_BIND_POINT_GRAPHICS,
Unwrap(m_Overlay.m_QuadResolvePipeline[SampleIndex(iminfo.samples)]));
vt->CmdBindDescriptorSets(Unwrap(cmd), VK_PIPELINE_BIND_POINT_GRAPHICS,
Unwrap(m_Overlay.m_QuadResolvePipeLayout), 0, 1,
UnwrapPtr(m_Overlay.m_QuadDescSet), 0, NULL);
VkViewport viewport = {
0.0f, 0.0f, (float)m_Overlay.ImageDim.width, (float)m_Overlay.ImageDim.height,
0.0f, 1.0f};
vt->CmdSetViewport(Unwrap(cmd), 0, 1, &viewport);
vt->CmdDraw(Unwrap(cmd), 4, 1, 0, 0);
vt->CmdEndRenderPass(Unwrap(cmd));
vkr = vt->EndCommandBuffer(Unwrap(cmd));
CheckVkResult(vkr);
}
m_pDriver->SubmitCmds();
m_pDriver->FlushQ();
m_pDriver->vkDestroyImageView(m_Device, quadImgView, NULL);
m_pDriver->vkDestroyImage(m_Device, quadImg, NULL);
m_pDriver->vkFreeMemory(m_Device, quadImgMem, NULL);
}
// restore back to normal
m_pDriver->ReplayLog(0, eventId, eReplay_WithoutDraw);
cmd = m_pDriver->GetNextCmd();
if(cmd == VK_NULL_HANDLE)
return ResourceId();
vkr = vt->BeginCommandBuffer(Unwrap(cmd), &beginInfo);
CheckVkResult(vkr);
}
}
else if(overlay == DebugOverlay::TriangleSizePass || overlay == DebugOverlay::TriangleSizeDraw)
{
if(!pipeInfo.rasterizerDiscardEnable)
{
VulkanRenderState prevstate = state;
VkPipelineShaderStageCreateInfo stages[3] = {
{VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, NULL, 0, VK_SHADER_STAGE_VERTEX_BIT,
shaderCache->GetBuiltinModule(BuiltinShader::MeshVS), "main", NULL},
{VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, NULL, 0, VK_SHADER_STAGE_FRAGMENT_BIT,
shaderCache->GetBuiltinModule(BuiltinShader::TrisizeFS), "main", NULL},
{VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, NULL, 0, VK_SHADER_STAGE_GEOMETRY_BIT,
shaderCache->GetBuiltinModule(BuiltinShader::TrisizeGS), "main", NULL},
};
if(stages[0].module != VK_NULL_HANDLE && stages[1].module != VK_NULL_HANDLE &&
stages[2].module != VK_NULL_HANDLE)
{
SCOPED_TIMER("Triangle Size");
float black[] = {0.0f, 0.0f, 0.0f, 0.0f};
VkImageMemoryBarrier barrier = {VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
NULL,
VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT,
VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
VK_QUEUE_FAMILY_IGNORED,
VK_QUEUE_FAMILY_IGNORED,
Unwrap(m_Overlay.Image),
subRange};
DoPipelineBarrier(cmd, 1, &barrier);
vt->CmdClearColorImage(Unwrap(cmd), Unwrap(m_Overlay.Image),
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, (VkClearColorValue *)black, 1,
&subRange);
std::swap(barrier.oldLayout, barrier.newLayout);
std::swap(barrier.srcAccessMask, barrier.dstAccessMask);
barrier.dstAccessMask |= VK_ACCESS_COLOR_ATTACHMENT_READ_BIT;
DoPipelineBarrier(cmd, 1, &barrier);
// end this cmd buffer so the image is in the right state for the next part
vkr = vt->EndCommandBuffer(Unwrap(cmd));
CheckVkResult(vkr);
#if ENABLED(SINGLE_FLUSH_VALIDATE)
m_pDriver->SubmitCmds();
#endif
rdcarray<uint32_t> events = passEvents;
if(overlay == DebugOverlay::TriangleSizeDraw)
events.clear();
while(!events.empty())
{
const ActionDescription *action = m_pDriver->GetAction(events[0]);
// remove any non-drawcalls, like the pass boundary.
if(!action || !(action->flags & ActionFlags::Drawcall))
events.erase(0);
else
break;
}
events.push_back(eventId);
m_pDriver->ReplayLog(0, events[0], eReplay_WithoutDraw);
uint32_t meshOffs = 0;
MeshUBOData *data = (MeshUBOData *)m_MeshRender.UBO.Map(&meshOffs);
if(!data)
return ResourceId();
data->mvp = Matrix4f::Identity();
data->invProj = Matrix4f::Identity();
data->color = Vec4f();
data->homogenousInput = 1;
data->pointSpriteSize = Vec2f(0.0f, 0.0f);
data->displayFormat = 0;
data->rawoutput = 1;
data->flipY = 0;
data->padding = Vec2f();
m_MeshRender.UBO.Unmap();
uint32_t viewOffs = 0;
Vec4f *ubo = (Vec4f *)m_Overlay.m_TriSizeUBO.Map(&viewOffs);
if(!ubo)
return ResourceId();
*ubo = Vec4f(state.views[0].width, state.views[0].height);
m_Overlay.m_TriSizeUBO.Unmap();
uint32_t offsets[2] = {meshOffs, viewOffs};
VkDescriptorBufferInfo bufdesc;
m_MeshRender.UBO.FillDescriptor(bufdesc);
VkWriteDescriptorSet write = {VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
NULL,
Unwrap(m_Overlay.m_TriSizeDescSet),
0,
0,
1,
VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC,
NULL,
&bufdesc,
NULL};
vt->UpdateDescriptorSets(Unwrap(m_Device), 1, &write, 0, NULL);
m_Overlay.m_TriSizeUBO.FillDescriptor(bufdesc);
write.dstBinding = 2;
vt->UpdateDescriptorSets(Unwrap(m_Device), 1, &write, 0, NULL);
VkRenderPass RP = m_Overlay.NoDepthRP;
VkFramebuffer FB = m_Overlay.NoDepthFB;
VulkanCreationInfo &createinfo = m_pDriver->m_CreationInfo;
ResourceId depthStencilView;
if(state.dynamicRendering.active)
{
depthStencilView = GetResID(state.dynamicRendering.depth.imageView);
if(depthStencilView == ResourceId())
depthStencilView = GetResID(state.dynamicRendering.stencil.imageView);
}
else
{
RDCASSERT(state.subpass < createinfo.m_RenderPass[state.GetRenderPass()].subpasses.size());
int32_t dsIdx =
createinfo.m_RenderPass[state.GetRenderPass()].subpasses[state.subpass].depthstencilAttachment;
// make a renderpass and framebuffer for rendering to overlay color and using
// depth buffer from the orignial render
if(dsIdx >= 0 &&
dsIdx < (int32_t)createinfo.m_Framebuffer[state.GetFramebuffer()].attachments.size())
{
depthStencilView = state.GetFramebufferAttachments()[dsIdx];
}
}
if(depthStencilView != ResourceId())
{
VkAttachmentDescription attDescs[] = {
{0, overlayFormat, VK_SAMPLE_COUNT_1_BIT, VK_ATTACHMENT_LOAD_OP_LOAD,
VK_ATTACHMENT_STORE_OP_STORE, VK_ATTACHMENT_LOAD_OP_DONT_CARE,
VK_ATTACHMENT_STORE_OP_DONT_CARE, VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL,
VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL},
{0, VK_FORMAT_UNDEFINED, VK_SAMPLE_COUNT_1_BIT, // will patch this just below
VK_ATTACHMENT_LOAD_OP_LOAD, VK_ATTACHMENT_STORE_OP_STORE, VK_ATTACHMENT_LOAD_OP_LOAD,
VK_ATTACHMENT_STORE_OP_STORE, VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL,
VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL},
};
VulkanCreationInfo::ImageView &depthViewInfo = createinfo.m_ImageView[depthStencilView];
ResourceId depthIm = depthViewInfo.image;
VulkanCreationInfo::Image &depthImageInfo = createinfo.m_Image[depthIm];
attDescs[1].format = depthImageInfo.format;
attDescs[0].samples = attDescs[1].samples = iminfo.samples;
{
LockedConstImageStateRef imState = m_pDriver->FindConstImageState(depthIm);
if(imState)
{
// find the state that overlaps the view's subresource range start. We assume all
// subresources are correctly in the same state (as they should be) so we just need to
// find the first match.
auto it = imState->subresourceStates.RangeBegin(depthViewInfo.range);
if(it != imState->subresourceStates.end())
attDescs[1].initialLayout = attDescs[1].finalLayout = it->state().newLayout;
}
}
VkAttachmentReference colRef = {0, VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL};
VkAttachmentReference dsRef = {1, attDescs[1].initialLayout};
VkSubpassDescription subp = {
0, VK_PIPELINE_BIND_POINT_GRAPHICS,
0, NULL, // inputs
1, &colRef, // color
NULL, // resolve
&dsRef, // depth-stencil
0, NULL, // preserve
};
VkRenderPassCreateInfo rpinfo = {
VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO,
NULL,
0,
2,
attDescs,
1,
&subp,
0,
NULL, // dependencies
};
if(multiviewMask > 0)
rpinfo.pNext = &multiviewRP;
vkr = m_pDriver->vkCreateRenderPass(m_Device, &rpinfo, NULL, &RP);
CheckVkResult(vkr);
VkImageView views[] = {
m_Overlay.ImageView,
m_pDriver->GetResourceManager()->GetCurrentHandle<VkImageView>(depthStencilView),
};
// Create framebuffer rendering just to overlay image, no depth
VkFramebufferCreateInfo fbinfo = {
VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO,
NULL,
0,
RP,
2,
views,
RDCMAX(1U, m_Overlay.ImageDim.width >> sub.mip),
RDCMAX(1U, m_Overlay.ImageDim.height >> sub.mip),
sub.numSlices,
};
vkr = m_pDriver->vkCreateFramebuffer(m_Device, &fbinfo, NULL, &FB);
CheckVkResult(vkr);
}
VkGraphicsPipelineCreateInfo pipeCreateInfo;
m_pDriver->GetShaderCache()->MakeGraphicsPipelineInfo(pipeCreateInfo,
state.graphics.pipeline);
VkPipelineInputAssemblyStateCreateInfo ia = {
VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO};
// ia.topology will be set below on a per-draw basis
VkVertexInputBindingDescription binds[] = {
// primary
{0, 0, VK_VERTEX_INPUT_RATE_VERTEX},
// secondary
{1, 0, VK_VERTEX_INPUT_RATE_VERTEX},
};
VkVertexInputAttributeDescription vertAttrs[] = {
{0, 0, VK_FORMAT_R32G32B32A32_SFLOAT, 0}, {1, 0, VK_FORMAT_R32G32B32A32_SFLOAT, 0},
};
VkPipelineVertexInputStateCreateInfo vi = {
VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO,
NULL,
0,
1,
binds,
2,
vertAttrs,
};
VkPipelineColorBlendAttachmentState attState = {
false,
VK_BLEND_FACTOR_ONE,
VK_BLEND_FACTOR_ZERO,
VK_BLEND_OP_ADD,
VK_BLEND_FACTOR_ONE,
VK_BLEND_FACTOR_ZERO,
VK_BLEND_OP_ADD,
0xf,
};
VkPipelineColorBlendStateCreateInfo cb = {
VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO,
NULL,
0,
false,
VK_LOGIC_OP_NO_OP,
1,
&attState,
{1.0f, 1.0f, 1.0f, 1.0f},
};
pipeCreateInfo.stageCount = 3;
pipeCreateInfo.pStages = stages;
pipeCreateInfo.pTessellationState = NULL;
pipeCreateInfo.renderPass = RP;
pipeCreateInfo.subpass = 0;
pipeCreateInfo.layout = m_Overlay.m_TriSizePipeLayout;
pipeCreateInfo.basePipelineHandle = VK_NULL_HANDLE;
pipeCreateInfo.basePipelineIndex = 0;
pipeCreateInfo.pInputAssemblyState = &ia;
pipeCreateInfo.pVertexInputState = &vi;
pipeCreateInfo.pColorBlendState = &cb;
// don't use dynamic rendering
RemoveNextStruct(&pipeCreateInfo, VK_STRUCTURE_TYPE_PIPELINE_RENDERING_CREATE_INFO);
uint32_t &dynamicStateCount = (uint32_t &)pipeCreateInfo.pDynamicState->dynamicStateCount;
VkDynamicState *dynamicStateList =
(VkDynamicState *)pipeCreateInfo.pDynamicState->pDynamicStates;
// remove any dynamic states we don't want
for(uint32_t i = 0; i < dynamicStateCount;)
{
// we are controlling the vertex binding so we don't need the stride or input to be
// dynamic.
// Similarly we're controlling the topology so that doesn't need to be dynamic
if(dynamicStateList[i] == VK_DYNAMIC_STATE_VERTEX_INPUT_BINDING_STRIDE ||
dynamicStateList[i] == VK_DYNAMIC_STATE_VERTEX_INPUT_EXT ||
dynamicStateList[i] == VK_DYNAMIC_STATE_PRIMITIVE_TOPOLOGY)
{
// swap with the last item if this isn't the last one
if(i != dynamicStateCount - 1)
std::swap(dynamicStateList[i], dynamicStateList[dynamicStateCount - 1]);
// then pop the last item.
dynamicStateCount--;
// process this item again. If we swapped we'll then consider that dynamic state, and if
// we didn't then this was the last item and i will be past dynamicStateCount now
continue;
}
i++;
}
typedef rdcpair<uint32_t, Topology> PipeKey;
std::map<PipeKey, VkPipeline> pipes;
for(size_t i = 0; i < events.size(); i++)
{
cmd = m_pDriver->GetNextCmd();
if(cmd == VK_NULL_HANDLE)
return ResourceId();
vkr = vt->BeginCommandBuffer(Unwrap(cmd), &beginInfo);
CheckVkResult(vkr);
VkClearValue clearval = {};
VkRenderPassBeginInfo rpbegin = {
VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
NULL,
Unwrap(RP),
Unwrap(FB),
{
{0, 0},
{
RDCMAX(1U, m_Overlay.ImageDim.width >> sub.mip),
RDCMAX(1U, m_Overlay.ImageDim.height >> sub.mip),
},
},
1,
&clearval,
};
vt->CmdBeginRenderPass(Unwrap(cmd), &rpbegin, VK_SUBPASS_CONTENTS_INLINE);
const ActionDescription *action = m_pDriver->GetAction(events[i]);
for(uint32_t inst = 0; action && inst < RDCMAX(1U, action->numInstances); inst++)
{
MeshFormat fmt = GetPostVSBuffers(events[i], inst, 0, MeshDataStage::GSOut);
if(fmt.vertexResourceId == ResourceId())
fmt = GetPostVSBuffers(events[i], inst, 0, MeshDataStage::VSOut);
if(fmt.vertexResourceId != ResourceId())
{
ia.topology = MakeVkPrimitiveTopology(fmt.topology);
binds[0].stride = binds[1].stride = fmt.vertexByteStride;
PipeKey key = make_rdcpair(fmt.vertexByteStride, fmt.topology);
VkPipeline pipe = pipes[key];
if(pipe == VK_NULL_HANDLE)
{
vkr = m_pDriver->vkCreateGraphicsPipelines(m_Device, VK_NULL_HANDLE, 1,
&pipeCreateInfo, NULL, &pipe);
CheckVkResult(vkr);
}
VkBuffer vb =
m_pDriver->GetResourceManager()->GetCurrentHandle<VkBuffer>(fmt.vertexResourceId);
VkDeviceSize offs = fmt.vertexByteOffset;
vt->CmdBindVertexBuffers(Unwrap(cmd), 0, 1, UnwrapPtr(vb), &offs);
pipes[key] = pipe;
vt->CmdBindDescriptorSets(Unwrap(cmd), VK_PIPELINE_BIND_POINT_GRAPHICS,
Unwrap(m_Overlay.m_TriSizePipeLayout), 0, 1,
UnwrapPtr(m_Overlay.m_TriSizeDescSet), 2, offsets);
vt->CmdBindPipeline(Unwrap(cmd), VK_PIPELINE_BIND_POINT_GRAPHICS, Unwrap(pipe));
const VkPipelineDynamicStateCreateInfo *dyn = pipeCreateInfo.pDynamicState;
for(uint32_t dynState = 0; dyn && dynState < dyn->dynamicStateCount; dynState++)
{
VkDynamicState d = dyn->pDynamicStates[dynState];
if(!state.views.empty() && d == VK_DYNAMIC_STATE_VIEWPORT)
{
vt->CmdSetViewport(Unwrap(cmd), 0, (uint32_t)state.views.size(), &state.views[0]);
}
else if(!state.scissors.empty() && d == VK_DYNAMIC_STATE_SCISSOR)
{
vt->CmdSetScissor(Unwrap(cmd), 0, (uint32_t)state.scissors.size(),
&state.scissors[0]);
}
else if(d == VK_DYNAMIC_STATE_LINE_WIDTH)
{
vt->CmdSetLineWidth(Unwrap(cmd), state.lineWidth);
}
else if(d == VK_DYNAMIC_STATE_DEPTH_BIAS)
{
vt->CmdSetDepthBias(Unwrap(cmd), state.bias.depth, state.bias.biasclamp,
state.bias.slope);
}
else if(d == VK_DYNAMIC_STATE_BLEND_CONSTANTS)
{
vt->CmdSetBlendConstants(Unwrap(cmd), state.blendConst);
}
else if(d == VK_DYNAMIC_STATE_DEPTH_BOUNDS)
{
vt->CmdSetDepthBounds(Unwrap(cmd), state.mindepth, state.maxdepth);
}
else if(d == VK_DYNAMIC_STATE_STENCIL_COMPARE_MASK)
{
vt->CmdSetStencilCompareMask(Unwrap(cmd), VK_STENCIL_FACE_BACK_BIT,
state.back.compare);
vt->CmdSetStencilCompareMask(Unwrap(cmd), VK_STENCIL_FACE_FRONT_BIT,
state.front.compare);
}
else if(d == VK_DYNAMIC_STATE_STENCIL_WRITE_MASK)
{
vt->CmdSetStencilWriteMask(Unwrap(cmd), VK_STENCIL_FACE_BACK_BIT, state.back.write);
vt->CmdSetStencilWriteMask(Unwrap(cmd), VK_STENCIL_FACE_FRONT_BIT,
state.front.write);
}
else if(d == VK_DYNAMIC_STATE_STENCIL_REFERENCE)
{
vt->CmdSetStencilReference(Unwrap(cmd), VK_STENCIL_FACE_BACK_BIT, state.back.ref);
vt->CmdSetStencilReference(Unwrap(cmd), VK_STENCIL_FACE_FRONT_BIT, state.front.ref);
}
else if(d == VK_DYNAMIC_STATE_VIEWPORT_WITH_COUNT)
{
vt->CmdSetViewportWithCountEXT(Unwrap(cmd), (uint32_t)state.views.size(),
state.views.data());
}
else if(d == VK_DYNAMIC_STATE_SCISSOR_WITH_COUNT)
{
vt->CmdSetScissorWithCountEXT(Unwrap(cmd), (uint32_t)state.scissors.size(),
state.scissors.data());
}
else if(d == VK_DYNAMIC_STATE_CULL_MODE)
{
vt->CmdSetCullModeEXT(Unwrap(cmd), state.cullMode);
}
else if(d == VK_DYNAMIC_STATE_FRONT_FACE)
{
vt->CmdSetFrontFaceEXT(Unwrap(cmd), state.frontFace);
}
else if(d == VK_DYNAMIC_STATE_PRIMITIVE_TOPOLOGY)
{
RDCERR("Primitive topology dynamic state found, should have been stripped");
}
else if(d == VK_DYNAMIC_STATE_VERTEX_INPUT_BINDING_STRIDE)
{
RDCERR(
"Vertex input binding stride dynamic state found, should have been stripped");
}
else if(d == VK_DYNAMIC_STATE_DEPTH_TEST_ENABLE)
{
vt->CmdSetDepthTestEnableEXT(Unwrap(cmd), state.depthTestEnable);
}
else if(d == VK_DYNAMIC_STATE_DEPTH_WRITE_ENABLE)
{
vt->CmdSetDepthWriteEnableEXT(Unwrap(cmd), state.depthWriteEnable);
}
else if(d == VK_DYNAMIC_STATE_DEPTH_COMPARE_OP)
{
vt->CmdSetDepthCompareOpEXT(Unwrap(cmd), state.depthCompareOp);
}
else if(d == VK_DYNAMIC_STATE_DEPTH_BOUNDS_TEST_ENABLE)
{
vt->CmdSetDepthBoundsTestEnableEXT(Unwrap(cmd), state.depthBoundsTestEnable);
}
else if(d == VK_DYNAMIC_STATE_STENCIL_TEST_ENABLE)
{
vt->CmdSetStencilTestEnableEXT(Unwrap(cmd), state.stencilTestEnable);
}
else if(d == VK_DYNAMIC_STATE_STENCIL_OP)
{
vt->CmdSetStencilOpEXT(Unwrap(cmd), VK_STENCIL_FACE_FRONT_BIT, state.front.failOp,
state.front.passOp, state.front.depthFailOp,
state.front.compareOp);
vt->CmdSetStencilOpEXT(Unwrap(cmd), VK_STENCIL_FACE_BACK_BIT, state.front.failOp,
state.front.passOp, state.front.depthFailOp,
state.front.compareOp);
}
else if(d == VK_DYNAMIC_STATE_COLOR_WRITE_ENABLE_EXT)
{
vt->CmdSetColorWriteEnableEXT(Unwrap(cmd), (uint32_t)state.colorWriteEnable.size(),
state.colorWriteEnable.data());
}
else if(d == VK_DYNAMIC_STATE_DEPTH_BIAS_ENABLE)
{
vt->CmdSetDepthBiasEnableEXT(Unwrap(cmd), state.depthBiasEnable);
}
else if(d == VK_DYNAMIC_STATE_LOGIC_OP_EXT)
{
vt->CmdSetLogicOpEXT(Unwrap(cmd), state.logicOp);
}
else if(d == VK_DYNAMIC_STATE_PATCH_CONTROL_POINTS_EXT)
{
vt->CmdSetPatchControlPointsEXT(Unwrap(cmd), state.patchControlPoints);
}
else if(d == VK_DYNAMIC_STATE_PRIMITIVE_RESTART_ENABLE)
{
vt->CmdSetPrimitiveRestartEnableEXT(Unwrap(cmd), state.primRestartEnable);
}
else if(d == VK_DYNAMIC_STATE_RASTERIZER_DISCARD_ENABLE)
{
vt->CmdSetRasterizerDiscardEnableEXT(Unwrap(cmd), state.rastDiscardEnable);
}
else if(d == VK_DYNAMIC_STATE_VERTEX_INPUT_EXT)
{
RDCERR("Vertex input dynamic state found, should have been stripped");
}
}
if(fmt.indexByteStride)
{
VkIndexType idxtype = VK_INDEX_TYPE_UINT16;
if(fmt.indexByteStride == 4)
idxtype = VK_INDEX_TYPE_UINT32;
else if(fmt.indexByteStride == 1)
idxtype = VK_INDEX_TYPE_UINT8_EXT;
if(fmt.indexResourceId != ResourceId())
{
VkBuffer ib =
m_pDriver->GetResourceManager()->GetLiveHandle<VkBuffer>(fmt.indexResourceId);
vt->CmdBindIndexBuffer(Unwrap(cmd), Unwrap(ib), fmt.indexByteOffset, idxtype);
vt->CmdDrawIndexed(Unwrap(cmd), fmt.numIndices, 1, 0, fmt.baseVertex, 0);
}
}
else
{
vt->CmdDraw(Unwrap(cmd), fmt.numIndices, 1, 0, 0);
}
}
}
vt->CmdEndRenderPass(Unwrap(cmd));
vkr = vt->EndCommandBuffer(Unwrap(cmd));
CheckVkResult(vkr);
if(overlay == DebugOverlay::TriangleSizePass)
{
m_pDriver->ReplayLog(events[i], events[i], eReplay_OnlyDraw);
if(i + 1 < events.size())
m_pDriver->ReplayLog(events[i], events[i + 1], eReplay_WithoutDraw);
}
}
m_pDriver->SubmitCmds();
m_pDriver->FlushQ();
if(depthStencilView != ResourceId())
{
m_pDriver->vkDestroyFramebuffer(m_Device, FB, NULL);
m_pDriver->vkDestroyRenderPass(m_Device, RP, NULL);
}
for(auto it = pipes.begin(); it != pipes.end(); ++it)
m_pDriver->vkDestroyPipeline(m_Device, it->second, NULL);
}
// restore back to normal
m_pDriver->ReplayLog(0, eventId, eReplay_WithoutDraw);
// restore state
state = prevstate;
cmd = m_pDriver->GetNextCmd();
if(cmd == VK_NULL_HANDLE)
return ResourceId();
vkr = vt->BeginCommandBuffer(Unwrap(cmd), &beginInfo);
CheckVkResult(vkr);
}
}
VkMarkerRegion::End(cmd);
vkr = vt->EndCommandBuffer(Unwrap(cmd));
CheckVkResult(vkr);
#if ENABLED(SINGLE_FLUSH_VALIDATE)
m_pDriver->SubmitCmds();
#endif
return GetResID(m_Overlay.Image);
}
Reference in New Issue View Git Blame Copy Permalink