Ryujinx/moltenvk
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Clear attachments using layered rendering only if framebuffer is layered.

Browse Source 
Consolidate testing of subpass layered rendering.
Add MVKCommandEncoder _isUsingLayeredRendering member variable.
Track layered rendering enabled in MVKRPSKeyClearAtt.
Rename MVKRPSKeyClearAtt::enable() and ::isEnabled() functions.
Add MVKRPSKeyClearAtt::reset() function.
Consolidate viewport, scissor & attachment limits.
MVKImage add validation for layered rendering when used as attachment.
This commit is contained in:
Bill Hollings 2019-05-15 22:40:47 -04:00
parent e80b535526
commit 2da855a6bc
14 changed files with 66 additions and 53 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
Docs/Whats_New.md
View File

@ -30,7 +30,7 @@ Released TBD
- Fix tessellated indirect draws using wrong kernels to map parameters.
- Work around potential Metal bug with stage-in indirect buffers.
- Fix zero local threadgroup size in indirect tessellated rendering.
- Fix crash with multisample layered rendering on older macOS devices.
- Fix crash when clearing attachments using layered rendering on older macOS devices.
- Fixes to Metal renderpass layered rendering settings.
- `MoltenVKShaderConverter` tool: Add MSL version and platform command-line options.
- Allow building external dependency libraries in `Debug` mode.

4
MoltenVK/MoltenVK/Commands/MVKCmdRenderPass.h
View File

@ -115,7 +115,7 @@ public:
private:
uint32_t _firstViewport;
MVKVectorInline<MTLViewport, kMVKCachedViewportCount> _mtlViewports;
MVKVectorInline<MTLViewport, kMVKCachedViewportScissorCount> _mtlViewports;
};
@ -134,7 +134,7 @@ public:
private:
uint32_t _firstScissor;
MVKVectorInline<MTLScissorRect, kMVKCachedScissorCount> _mtlScissors;
MVKVectorInline<MTLScissorRect, kMVKCachedViewportScissorCount> _mtlScissors;
};

4
MoltenVK/MoltenVK/Commands/MVKCmdTransfer.h
View File

@ -243,8 +243,8 @@ protected:
std::vector<VkClearRect> _clearRects;
std::vector<simd::float4> _vertices;
simd::float4 _clearColors[kMVKAttachmentFormatCount];
VkClearValue _vkClearValues[kMVKAttachmentFormatCount];
simd::float4 _clearColors[kMVKClearAttachmentCount];
VkClearValue _vkClearValues[kMVKClearAttachmentCount];
MVKRPSKeyClearAtt _rpsKey;
uint32_t _mtlStencilValue;
bool _isClearingDepth;

15
MoltenVK/MoltenVK/Commands/MVKCmdTransfer.mm
View File

@ -816,7 +816,7 @@ void MVKCmdClearAttachments::setContent(uint32_t attachmentCount,
const VkClearAttachment* pAttachments,
uint32_t rectCount,
const VkClearRect* pRects) {
_rpsKey = kMVKRPSKeyClearAttDefault;
_rpsKey.reset();
_mtlStencilValue = 0;
_isClearingDepth = false;
_isClearingStencil = false;
@ -831,26 +831,26 @@ void MVKCmdClearAttachments::setContent(uint32_t attachmentCount,
if (mvkIsAnyFlagEnabled(clrAtt.aspectMask, VK_IMAGE_ASPECT_COLOR_BIT)) {
uint32_t caIdx = clrAtt.colorAttachment; // Might be VK_ATTACHMENT_UNUSED
if (caIdx != VK_ATTACHMENT_UNUSED) {
_rpsKey.enable(caIdx);
_rpsKey.enableAttachment(caIdx);
_vkClearValues[caIdx] = clrAtt.clearValue;
}
}
if (mvkIsAnyFlagEnabled(clrAtt.aspectMask, VK_IMAGE_ASPECT_DEPTH_BIT)) {
_isClearingDepth = true;
_rpsKey.enable(kMVKAttachmentFormatDepthStencilIndex);
_rpsKey.enableAttachment(kMVKClearAttachmentDepthStencilIndex);
mtlDepthVal = mvkMTLClearDepthFromVkClearValue(clrAtt.clearValue);
}
if (mvkIsAnyFlagEnabled(clrAtt.aspectMask, VK_IMAGE_ASPECT_STENCIL_BIT)) {
_isClearingStencil = true;
_rpsKey.enable(kMVKAttachmentFormatDepthStencilIndex);
_rpsKey.enableAttachment(kMVKClearAttachmentDepthStencilIndex);
_mtlStencilValue = mvkMTLClearStencilFromVkClearValue(clrAtt.clearValue);
}
}
// The depth value (including vertex position Z value) is held in the last index.
_clearColors[kMVKAttachmentFormatDepthStencilIndex] = { mtlDepthVal, mtlDepthVal, mtlDepthVal, mtlDepthVal };
_clearColors[kMVKClearAttachmentDepthStencilIndex] = { mtlDepthVal, mtlDepthVal, mtlDepthVal, mtlDepthVal };
_clearRects.clear();
_clearRects.reserve(rectCount);
@ -932,8 +932,9 @@ void MVKCmdClearAttachments::encode(MVKCommandEncoder* cmdEncoder) {
uint32_t vtxCnt = (uint32_t)_vertices.size();
uint32_t vtxBuffIdx = getDevice()->getMetalBufferIndexForVertexAttributeBinding(kMVKVertexContentBufferIndex);
// Populate the render pipeline state attachment key with attachment info from the subpass.
// Populate the render pipeline state attachment key with info from the subpass and framebuffer.
_rpsKey.mtlSampleCount = mvkSampleCountFromVkSampleCountFlagBits(subpass->getSampleCount());
if (cmdEncoder->_isUsingLayeredRendering) { _rpsKey.enableLayeredRendering(); }
uint32_t caCnt = subpass->getColorAttachmentCount();
for (uint32_t caIdx = 0; caIdx < caCnt; caIdx++) {
@ -945,7 +946,7 @@ void MVKCmdClearAttachments::encode(MVKCommandEncoder* cmdEncoder) {
VkFormat vkAttFmt = subpass->getDepthStencilFormat();
MTLPixelFormat mtlAttFmt = cmdPool->getMTLPixelFormatFromVkFormat(vkAttFmt);
_rpsKey.attachmentMTLPixelFormats[kMVKAttachmentFormatDepthStencilIndex] = mtlAttFmt;
_rpsKey.attachmentMTLPixelFormats[kMVKClearAttachmentDepthStencilIndex] = mtlAttFmt;
bool isClearingDepth = _isClearingDepth && mvkMTLPixelFormatIsDepthFormat(mtlAttFmt);
bool isClearingStencil = _isClearingStencil && mvkMTLPixelFormatIsStencilFormat(mtlAttFmt);

3
MoltenVK/MoltenVK/Commands/MVKCommandBuffer.h
View File

@ -420,6 +420,9 @@ public:
/** The size of the threadgroup for the compute shader. */
MTLSize _mtlThreadgroupSize;
/** Indicates whether the current render subpass is rendering to an array (layered) framebuffer. */
bool _isUsingLayeredRendering;
#pragma mark Construction

14
MoltenVK/MoltenVK/Commands/MVKCommandBuffer.mm
View File

@ -234,6 +234,7 @@ void MVKCommandBuffer::recordDraw(MVKLoadStoreOverrideMixin* mvkDraw) {
void MVKCommandEncoder::encode(id<MTLCommandBuffer> mtlCmdBuff) {
_subpassContents = VK_SUBPASS_CONTENTS_INLINE;
_renderSubpassIndex = 0;
_isUsingLayeredRendering = false;
_mtlCmdBuffer = mtlCmdBuff; // not retained
@ -280,7 +281,12 @@ void MVKCommandEncoder::setSubpass(VkSubpassContents subpassContents, uint32_t s
_subpassContents = subpassContents;
_renderSubpassIndex = subpassIndex;
beginMetalRenderPass(loadOverride, storeOverride);
_isUsingLayeredRendering = ((_framebuffer->getLayerCount() > 1) &&
_device->_pMetalFeatures->layeredRendering &&
(_device->_pMetalFeatures->multisampleLayeredRendering ||
(getSubpass()->getSampleCount() == VK_SAMPLE_COUNT_1_BIT)));
beginMetalRenderPass(loadOverride, storeOverride);
}
// Creates _mtlRenderEncoder and marks cached render state as dirty so it will be set into the _mtlRenderEncoder.
@ -293,11 +299,7 @@ void MVKCommandEncoder::beginMetalRenderPass(bool loadOverride, bool storeOverri
mtlRPDesc.visibilityResultBuffer = _occlusionQueryState.getVisibilityResultMTLBuffer();
// Only set the layered rendering properties if layered rendering is supported and the framebuffer really has multiple layers
if ((_framebuffer->getLayerCount() > 1) &&
_device->_pMetalFeatures->layeredRendering &&
(_device->_pMetalFeatures->multisampleLayeredRendering ||
(getSubpass()->getSampleCount() == VK_SAMPLE_COUNT_1_BIT))) {
if (_isUsingLayeredRendering) {
VkExtent2D fbExtent = _framebuffer->getExtent2D();
mtlRPDesc.renderTargetWidthMVK = min(_renderArea.offset.x + _renderArea.extent.width, fbExtent.width);
mtlRPDesc.renderTargetHeightMVK = min(_renderArea.offset.y + _renderArea.extent.height, fbExtent.height);

4
MoltenVK/MoltenVK/Commands/MVKCommandEncoderState.h
View File

@ -151,7 +151,7 @@ protected:
void encodeImpl(uint32_t stage) override;
void resetImpl() override;
MVKVectorInline<MTLViewport, kMVKCachedViewportCount> _mtlViewports, _mtlDynamicViewports;
MVKVectorInline<MTLViewport, kMVKCachedViewportScissorCount> _mtlViewports, _mtlDynamicViewports;
};
@ -180,7 +180,7 @@ protected:
void encodeImpl(uint32_t stage) override;
void resetImpl() override;
MVKVectorInline<MTLScissorRect, kMVKCachedScissorCount> _mtlScissors, _mtlDynamicScissors;
MVKVectorInline<MTLScissorRect, kMVKCachedViewportScissorCount> _mtlScissors, _mtlDynamicScissors;
};

1
MoltenVK/MoltenVK/Commands/MVKCommandPool.h
View File

@ -20,7 +20,6 @@
#include "MVKDevice.h"
#include "MVKCommandBuffer.h"
#include "MVKCommandResourceFactory.h"
#include "MVKCommandEncodingPool.h"
#include "MVKCommand.h"
#include "MVKCmdPipeline.h"

30
MoltenVK/MoltenVK/Commands/MVKCommandResourceFactory.h
View File

@ -74,8 +74,9 @@ namespace std {
#pragma mark -
#pragma mark MVKRPSKeyClearAtt
#define kMVKAttachmentFormatCount 9
#define kMVKAttachmentFormatDepthStencilIndex (kMVKAttachmentFormatCount - 1)
#define kMVKClearAttachmentCount (kMVKCachedColorAttachmentCount + 1)
#define kMVKClearAttachmentDepthStencilIndex (kMVKClearAttachmentCount - 1)
#define kMVKClearAttachmentLayeredRenderingBitIndex kMVKClearAttachmentCount
/**
* Key to use for looking up cached MTLRenderPipelineState instances.
@ -85,37 +86,40 @@ namespace std {
* This structure can be used as a key in a std::map and std::unordered_map.
*/
typedef struct MVKRPSKeyClearAtt_t {
uint16_t attachmentMTLPixelFormats[kMVKAttachmentFormatCount];
uint16_t attachmentMTLPixelFormats[kMVKClearAttachmentCount];
uint16_t mtlSampleCount;
uint32_t enabledFlags;
uint16_t flags; // bitcount > kMVKClearAttachmentLayeredRenderingBitIndex
const static uint32_t bitFlag = 1;
void enable(uint32_t attIdx) { mvkEnableFlag(enabledFlags, bitFlag << attIdx); }
void enableAttachment(uint32_t attIdx) { mvkEnableFlag(flags, bitFlag << attIdx); }
bool isEnabled(uint32_t attIdx) { return mvkIsAnyFlagEnabled(enabledFlags, bitFlag << attIdx); }
bool isAttachmentEnabled(uint32_t attIdx) { return mvkIsAnyFlagEnabled(flags, bitFlag << attIdx); }
void enableLayeredRendering() { mvkEnableFlag(flags, bitFlag << kMVKClearAttachmentLayeredRenderingBitIndex); }
bool isLayeredRenderingEnabled() { return mvkIsAnyFlagEnabled(flags, bitFlag << kMVKClearAttachmentLayeredRenderingBitIndex); }
bool operator==(const MVKRPSKeyClearAtt_t& rhs) const {
return ((enabledFlags == rhs.enabledFlags) &&
return ((flags == rhs.flags) &&
(mtlSampleCount == rhs.mtlSampleCount) &&
(memcmp(attachmentMTLPixelFormats, rhs.attachmentMTLPixelFormats, sizeof(attachmentMTLPixelFormats)) == 0));
}
std::size_t hash() const {
std::size_t hash = mvkHash(&enabledFlags);
std::size_t hash = mvkHash(&flags);
hash = mvkHash(&mtlSampleCount, 1, hash);
return mvkHash(attachmentMTLPixelFormats, kMVKAttachmentFormatCount, hash);
return mvkHash(attachmentMTLPixelFormats, kMVKClearAttachmentCount, hash);
}
MVKRPSKeyClearAtt_t() {
void reset() {
memset(this, 0, sizeof(*this));
mtlSampleCount = mvkSampleCountFromVkSampleCountFlagBits(VK_SAMPLE_COUNT_1_BIT);
}
} MVKRPSKeyClearAtt;
MVKRPSKeyClearAtt_t() { reset(); }
/** An instance populated with default values, for use in resetting other instances to default state. */
const MVKRPSKeyClearAtt kMVKRPSKeyClearAttDefault;
} MVKRPSKeyClearAtt;
/**
* Hash structure implementation for MVKRPSKeyClearAtt in std namespace,

17
MoltenVK/MoltenVK/Commands/MVKCommandResourceFactory.mm
View File

@ -95,12 +95,12 @@ id<MTLRenderPipelineState> MVKCommandResourceFactory::newCmdClearMTLRenderPipeli
plDesc.sampleCount = attKey.mtlSampleCount;
plDesc.inputPrimitiveTopologyMVK = MTLPrimitiveTopologyClassTriangle;
for (uint32_t caIdx = 0; caIdx < kMVKAttachmentFormatDepthStencilIndex; caIdx++) {
for (uint32_t caIdx = 0; caIdx < kMVKClearAttachmentDepthStencilIndex; caIdx++) {
MTLRenderPipelineColorAttachmentDescriptor* colorDesc = plDesc.colorAttachments[caIdx];
colorDesc.pixelFormat = (MTLPixelFormat)attKey.attachmentMTLPixelFormats[caIdx];
colorDesc.writeMask = attKey.isEnabled(caIdx) ? MTLColorWriteMaskAll : MTLColorWriteMaskNone;
colorDesc.writeMask = attKey.isAttachmentEnabled(caIdx) ? MTLColorWriteMaskAll : MTLColorWriteMaskNone;
}
MTLPixelFormat mtlDSFormat = (MTLPixelFormat)attKey.attachmentMTLPixelFormats[kMVKAttachmentFormatDepthStencilIndex];
MTLPixelFormat mtlDSFormat = (MTLPixelFormat)attKey.attachmentMTLPixelFormats[kMVKClearAttachmentDepthStencilIndex];
if (mvkMTLPixelFormatIsDepthFormat(mtlDSFormat)) { plDesc.depthAttachmentPixelFormat = mtlDSFormat; }
if (mvkMTLPixelFormatIsStencilFormat(mtlDSFormat)) { plDesc.stencilAttachmentPixelFormat = mtlDSFormat; }
@ -181,7 +181,6 @@ id<MTLFunction> MVKCommandResourceFactory::getBlitFragFunction(MVKRPSKeyBlitImg&
id<MTLFunction> MVKCommandResourceFactory::getClearVertFunction(MVKRPSKeyClearAtt& attKey) {
id<MTLFunction> mtlFunc = nil;
bool allowLayers = _device->_pMetalFeatures->layeredRendering && (attKey.mtlSampleCount == 1 || _device->_pMetalFeatures->multisampleLayeredRendering);
@autoreleasepool {
NSMutableString* msl = [NSMutableString stringWithCapacity: (2 * KIBI) ];
[msl appendLineMVK: @"#include <metal_stdlib>"];
@ -197,7 +196,7 @@ id<MTLFunction> MVKCommandResourceFactory::getClearVertFunction(MVKRPSKeyClearAt
[msl appendLineMVK];
[msl appendLineMVK: @"typedef struct {"];
[msl appendLineMVK: @" float4 v_position [[position]];"];
[msl appendFormat: @" uint layer%s;", allowLayers ? " [[render_target_array_index]]" : ""];
[msl appendFormat: @" uint layer%s;", attKey.isLayeredRenderingEnabled() ? " [[render_target_array_index]]" : ""];
[msl appendLineMVK: @"} VaryingsPos;"];
[msl appendLineMVK];
@ -232,8 +231,8 @@ id<MTLFunction> MVKCommandResourceFactory::getClearFragFunction(MVKRPSKeyClearAt
[msl appendLineMVK: @"} ClearColorsIn;"];
[msl appendLineMVK];
[msl appendLineMVK: @"typedef struct {"];
for (uint32_t caIdx = 0; caIdx < kMVKAttachmentFormatDepthStencilIndex; caIdx++) {
if (attKey.isEnabled(caIdx)) {
for (uint32_t caIdx = 0; caIdx < kMVKClearAttachmentDepthStencilIndex; caIdx++) {
if (attKey.isAttachmentEnabled(caIdx)) {
NSString* typeStr = getMTLFormatTypeString((MTLPixelFormat)attKey.attachmentMTLPixelFormats[caIdx]);
[msl appendFormat: @" %@4 color%u [[color(%u)]];", typeStr, caIdx, caIdx];
[msl appendLineMVK];
@ -246,8 +245,8 @@ id<MTLFunction> MVKCommandResourceFactory::getClearFragFunction(MVKRPSKeyClearAt
[msl appendFormat: @"fragment ClearColorsOut %@(VaryingsPos varyings [[stage_in]], constant ClearColorsIn& ccIn [[buffer(0)]]) {", funcName];
[msl appendLineMVK];
[msl appendLineMVK: @" ClearColorsOut ccOut;"];
for (uint32_t caIdx = 0; caIdx < kMVKAttachmentFormatDepthStencilIndex; caIdx++) {
if (attKey.isEnabled(caIdx)) {
for (uint32_t caIdx = 0; caIdx < kMVKClearAttachmentDepthStencilIndex; caIdx++) {
if (attKey.isAttachmentEnabled(caIdx)) {
NSString* typeStr = getMTLFormatTypeString((MTLPixelFormat)attKey.attachmentMTLPixelFormats[caIdx]);
[msl appendFormat: @" ccOut.color%u = %@4(ccIn.colors[%u]);", caIdx, typeStr, caIdx];
[msl appendLineMVK];

4
MoltenVK/MoltenVK/GPUObjects/MVKDevice.h
View File

@ -67,8 +67,8 @@ class MVKCommandResourceFactory;
const static uint32_t kMVKVertexContentBufferIndex = 0;
// Parameters to define the sizing of inline collections
const static uint32_t kMVKCachedViewportCount = 16;
const static uint32_t kMVKCachedScissorCount = 16;
const static uint32_t kMVKCachedViewportScissorCount = 16;
const static uint32_t kMVKCachedColorAttachmentCount = 8;
#pragma mark -

8
MoltenVK/MoltenVK/GPUObjects/MVKDevice.mm
View File

@ -1002,13 +1002,13 @@ void MVKPhysicalDevice::initProperties() {
// Limits
#if MVK_IOS
if ( [_mtlDevice supportsFeatureSet: MTLFeatureSet_iOS_GPUFamily2_v1] ) {
_properties.limits.maxColorAttachments = 8;
_properties.limits.maxColorAttachments = kMVKCachedColorAttachmentCount;
} else {
_properties.limits.maxColorAttachments = 4;
_properties.limits.maxColorAttachments = 4; // < kMVKCachedColorAttachmentCount
}
#endif
#if MVK_MACOS
_properties.limits.maxColorAttachments = 8;
_properties.limits.maxColorAttachments = kMVKCachedColorAttachmentCount;
#endif
_properties.limits.maxFragmentOutputAttachments = _properties.limits.maxColorAttachments;
@ -1038,7 +1038,7 @@ void MVKPhysicalDevice::initProperties() {
float maxVPDim = max(_properties.limits.maxViewportDimensions[0], _properties.limits.maxViewportDimensions[1]);
_properties.limits.viewportBoundsRange[0] = (-2.0 * maxVPDim);
_properties.limits.viewportBoundsRange[1] = (2.0 * maxVPDim) - 1;
_properties.limits.maxViewports = _features.multiViewport ? 16 : 1;
_properties.limits.maxViewports = _features.multiViewport ? kMVKCachedViewportScissorCount : 1;
_properties.limits.maxImageDimension3D = (2 * KIBI);
_properties.limits.maxImageArrayLayers = (2 * KIBI);

9
MoltenVK/MoltenVK/GPUObjects/MVKImage.mm
View File

@ -608,6 +608,11 @@ void MVKImage::validateConfig(const VkImageCreateInfo* pCreateInfo) {
setConfigurationResult(reportError(VK_ERROR_FEATURE_NOT_PRESENT, "vkCreateImage() : Under Metal, depth/stencil formats may only be used with 2D images."));
}
bool isAttachment = mvkIsAnyFlagEnabled(pCreateInfo->usage, (VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT | VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT));
if (isAttachment && (pCreateInfo->arrayLayers > 1) && !_device->_pMetalFeatures->layeredRendering) {
setConfigurationResult(reportError(VK_ERROR_FEATURE_NOT_PRESENT, "vkCreateImage() : This device does not support rendering to array (layered) attachments."));
}
if (mvkIsAnyFlagEnabled(pCreateInfo->flags, VK_IMAGE_CREATE_BLOCK_TEXEL_VIEW_COMPATIBLE_BIT)) {
setConfigurationResult(reportError(VK_ERROR_FEATURE_NOT_PRESENT, "vkCreateImage() : Metal does not allow uncompressed views of compressed images."));
}
@ -635,8 +640,8 @@ VkSampleCountFlagBits MVKImage::validateSamples(const VkImageCreateInfo* pCreate
validSamples = VK_SAMPLE_COUNT_1_BIT;
}
if ( !_device->_pMetalFeatures->multisampleLayeredRendering && mvkIsAnyFlagEnabled(pCreateInfo->usage, (VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT |
VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT))) {
bool isAttachment = mvkIsAnyFlagEnabled(pCreateInfo->usage, (VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT | VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT));
if (isAttachment && !_device->_pMetalFeatures->multisampleLayeredRendering) {
setConfigurationResult(reportError(VK_ERROR_FEATURE_NOT_PRESENT, "vkCreateImage() : This device does not support rendering to multisampled array (layered) attachments. Setting sample count to 1."));
validSamples = VK_SAMPLE_COUNT_1_BIT;
}

4
MoltenVK/MoltenVK/GPUObjects/MVKPipeline.h
View File

@ -236,8 +236,8 @@ protected:
VkPipelineRasterizationStateCreateInfo _rasterInfo;
VkPipelineDepthStencilStateCreateInfo _depthStencilInfo;
MVKVectorInline<MTLViewport, kMVKCachedViewportCount> _mtlViewports;
MVKVectorInline<MTLScissorRect, kMVKCachedScissorCount> _mtlScissors;
MVKVectorInline<MTLViewport, kMVKCachedViewportScissorCount> _mtlViewports;
MVKVectorInline<MTLScissorRect, kMVKCachedViewportScissorCount> _mtlScissors;
MTLComputePipelineDescriptor* _mtlTessControlStageDesc = nil;