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

MVKPipeline: Reject non-multiple-of-4 vertex buffer strides.

Browse Source 
Metal unfortunately insists on this, or it will assert.
This commit is contained in:
Chip Davis 2018-12-12 22:48:40 -06:00
parent b4b2ca6b45
commit cb61882ed6
1 changed files with 7 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
MoltenVK/MoltenVK/GPUObjects/MVKPipeline.mm
View File

@ -384,6 +384,13 @@ MTLRenderPipelineDescriptor* MVKGraphicsPipeline::getMTLRenderPipelineDescriptor
uint32_t vbIdx = _device->getMetalBufferIndexForVertexAttributeBinding(pVKVB->binding);
if (shaderContext.isVertexBufferUsed(vbIdx)) {
MTLVertexBufferLayoutDescriptor* vbDesc = plDesc.vertexDescriptor.layouts[vbIdx];
// Vulkan allows any stride, but Metal only allows multiples of 4.
// TODO: We should try to expand the buffer to the required alignment
// in that case.
if ((pVKVB->stride % 4) != 0) {
setConfigurationResult(mvkNotifyErrorWithText(VK_ERROR_INITIALIZATION_FAILED, "Under Metal, vertex buffer strides must be aligned to four bytes."));
return nil;
}
vbDesc.stride = (pVKVB->stride == 0) ? sizeof(simd::float4) : pVKVB->stride; // Vulkan allows zero stride but Metal doesn't. Default to float4
vbDesc.stepFunction = mvkMTLVertexStepFunctionFromVkVertexInputRate(pVKVB->inputRate);
vbDesc.stepRate = 1;