void VertexArrayGL::computeStreamingAttributeSizes(const gl::AttributesMask &activeAttributesMask,
GLsizei instanceCount,
const gl::IndexRange &indexRange,
size_t *outStreamingDataSize,
size_t *outMaxAttributeDataSize) const
{
*outStreamingDataSize = 0;
*outMaxAttributeDataSize = 0;
ASSERT(mAttributesNeedStreaming.any());
const auto &attribs = mState.getVertexAttributes();
const auto &bindings = mState.getVertexBindings();
gl::AttributesMask attribsToStream = (mAttributesNeedStreaming & activeAttributesMask);
for (auto idx : attribsToStream)
{
const auto &attrib = attribs[idx];
const auto &binding = bindings[attrib.bindingIndex];
ASSERT(AttributeNeedsStreaming(attrib, binding));
// If streaming is going to be required, compute the size of the required buffer
// and how much slack space at the beginning of the buffer will be required by determining
// the attribute with the largest data size.
size_t typeSize = ComputeVertexAttributeTypeSize(attrib);
GLuint adjustedDivisor = GetAdjustedDivisor(mAppliedNumViews, binding.getDivisor());
*outStreamingDataSize +=
typeSize * ComputeVertexBindingElementCount(adjustedDivisor, indexRange.vertexCount(),
instanceCount);
*outMaxAttributeDataSize = std::max(*outMaxAttributeDataSize, typeSize);
}
}
void VertexArrayGL::computeStreamingAttributeSizes(const gl::AttributesMask &activeAttributesMask,
GLsizei instanceCount,
const gl::IndexRange &indexRange,
size_t *outStreamingDataSize,
size_t *outMaxAttributeDataSize) const
{
*outStreamingDataSize = 0;
*outMaxAttributeDataSize = 0;
ASSERT(mAttributesNeedStreaming.any());
const auto &attribs = mData.getVertexAttributes();
for (unsigned int idx : angle::IterateBitSet(mAttributesNeedStreaming & activeAttributesMask))
{
const auto &attrib = attribs[idx];
ASSERT(AttributeNeedsStreaming(attrib));
// If streaming is going to be required, compute the size of the required buffer
// and how much slack space at the beginning of the buffer will be required by determining
// the attribute with the largest data size.
size_t typeSize = ComputeVertexAttributeTypeSize(attrib);
*outStreamingDataSize += typeSize * ComputeVertexAttributeElementCount(
attrib, indexRange.vertexCount(), instanceCount);
*outMaxAttributeDataSize = std::max(*outMaxAttributeDataSize, typeSize);
}
}
gl::Error VertexArrayGL::streamAttributes(const gl::AttributesMask &activeAttributesMask,
GLsizei instanceCount,
const gl::IndexRange &indexRange) const
{
// Sync the vertex attribute state and track what data needs to be streamed
size_t streamingDataSize = 0;
size_t maxAttributeDataSize = 0;
computeStreamingAttributeSizes(activeAttributesMask, instanceCount, indexRange,
&streamingDataSize, &maxAttributeDataSize);
if (streamingDataSize == 0)
{
return gl::NoError();
}
if (mStreamingArrayBuffer == 0)
{
mFunctions->genBuffers(1, &mStreamingArrayBuffer);
mStreamingArrayBufferSize = 0;
}
// If first is greater than zero, a slack space needs to be left at the beginning of the buffer
// so that the same 'first' argument can be passed into the draw call.
const size_t bufferEmptySpace = maxAttributeDataSize * indexRange.start;
const size_t requiredBufferSize = streamingDataSize + bufferEmptySpace;
mStateManager->bindBuffer(gl::BufferBinding::Array, mStreamingArrayBuffer);
if (requiredBufferSize > mStreamingArrayBufferSize)
{
mFunctions->bufferData(GL_ARRAY_BUFFER, requiredBufferSize, nullptr, GL_DYNAMIC_DRAW);
mStreamingArrayBufferSize = requiredBufferSize;
}
// Unmapping a buffer can return GL_FALSE to indicate that the system has corrupted the data
// somehow (such as by a screen change), retry writing the data a few times and return
// OUT_OF_MEMORY if that fails.
GLboolean unmapResult = GL_FALSE;
size_t unmapRetryAttempts = 5;
while (unmapResult != GL_TRUE && --unmapRetryAttempts > 0)
{
uint8_t *bufferPointer = MapBufferRangeWithFallback(mFunctions, GL_ARRAY_BUFFER, 0,
requiredBufferSize, GL_MAP_WRITE_BIT);
size_t curBufferOffset = bufferEmptySpace;
const auto &attribs = mState.getVertexAttributes();
const auto &bindings = mState.getVertexBindings();
gl::AttributesMask attribsToStream = (mAttributesNeedStreaming & activeAttributesMask);
for (auto idx : attribsToStream)
{
const auto &attrib = attribs[idx];
ASSERT(IsVertexAttribPointerSupported(idx, attrib));
const auto &binding = bindings[attrib.bindingIndex];
ASSERT(AttributeNeedsStreaming(attrib, binding));
GLuint adjustedDivisor = GetAdjustedDivisor(mAppliedNumViews, binding.getDivisor());
const size_t streamedVertexCount = ComputeVertexBindingElementCount(
adjustedDivisor, indexRange.vertexCount(), instanceCount);
const size_t sourceStride = ComputeVertexAttributeStride(attrib, binding);
const size_t destStride = ComputeVertexAttributeTypeSize(attrib);
// Vertices do not apply the 'start' offset when the divisor is non-zero even when doing
// a non-instanced draw call
const size_t firstIndex = adjustedDivisor == 0 ? indexRange.start : 0;
// Attributes using client memory ignore the VERTEX_ATTRIB_BINDING state.
// https://www.opengl.org/registry/specs/ARB/vertex_attrib_binding.txt
const uint8_t *inputPointer = reinterpret_cast<const uint8_t *>(attrib.pointer);
// Pack the data when copying it, user could have supplied a very large stride that
// would cause the buffer to be much larger than needed.
if (destStride == sourceStride)
{
// Can copy in one go, the data is packed
memcpy(bufferPointer + curBufferOffset, inputPointer + (sourceStride * firstIndex),
destStride * streamedVertexCount);
}
else
{
// Copy each vertex individually
for (size_t vertexIdx = 0; vertexIdx < streamedVertexCount; vertexIdx++)
{
uint8_t *out = bufferPointer + curBufferOffset + (destStride * vertexIdx);
const uint8_t *in = inputPointer + sourceStride * (vertexIdx + firstIndex);
memcpy(out, in, destStride);
}
}
// Compute where the 0-index vertex would be.
const size_t vertexStartOffset = curBufferOffset - (firstIndex * destStride);
callVertexAttribPointer(static_cast<GLuint>(idx), attrib,
static_cast<GLsizei>(destStride),
static_cast<GLintptr>(vertexStartOffset));
curBufferOffset += destStride * streamedVertexCount;
}
unmapResult = mFunctions->unmapBuffer(GL_ARRAY_BUFFER);
}
if (unmapResult != GL_TRUE)
{
return gl::OutOfMemory() << "Failed to unmap the client data streaming buffer.";
}
return gl::NoError();
}
gl::Error VertexArrayGL::streamAttributes(const gl::AttributesMask &activeAttributesMask,
GLsizei instanceCount,
const gl::IndexRange &indexRange) const
{
// Sync the vertex attribute state and track what data needs to be streamed
size_t streamingDataSize = 0;
size_t maxAttributeDataSize = 0;
computeStreamingAttributeSizes(activeAttributesMask, instanceCount, indexRange,
&streamingDataSize, &maxAttributeDataSize);
if (streamingDataSize == 0)
{
return gl::Error(GL_NO_ERROR);
}
if (mStreamingArrayBuffer == 0)
{
mFunctions->genBuffers(1, &mStreamingArrayBuffer);
mStreamingArrayBufferSize = 0;
}
// If first is greater than zero, a slack space needs to be left at the beginning of the buffer so that
// the same 'first' argument can be passed into the draw call.
const size_t bufferEmptySpace = maxAttributeDataSize * indexRange.start;
const size_t requiredBufferSize = streamingDataSize + bufferEmptySpace;
mStateManager->bindBuffer(GL_ARRAY_BUFFER, mStreamingArrayBuffer);
if (requiredBufferSize > mStreamingArrayBufferSize)
{
mFunctions->bufferData(GL_ARRAY_BUFFER, requiredBufferSize, nullptr, GL_DYNAMIC_DRAW);
mStreamingArrayBufferSize = requiredBufferSize;
}
// Unmapping a buffer can return GL_FALSE to indicate that the system has corrupted the data
// somehow (such as by a screen change), retry writing the data a few times and return OUT_OF_MEMORY
// if that fails.
GLboolean unmapResult = GL_FALSE;
size_t unmapRetryAttempts = 5;
while (unmapResult != GL_TRUE && --unmapRetryAttempts > 0)
{
uint8_t *bufferPointer = reinterpret_cast<uint8_t*>(mFunctions->mapBuffer(GL_ARRAY_BUFFER, GL_WRITE_ONLY));
size_t curBufferOffset = bufferEmptySpace;
const auto &attribs = mData.getVertexAttributes();
for (unsigned int idx :
angle::IterateBitSet(mAttributesNeedStreaming & activeAttributesMask))
{
const auto &attrib = attribs[idx];
ASSERT(AttributeNeedsStreaming(attrib));
const size_t streamedVertexCount =
ComputeVertexAttributeElementCount(attrib, indexRange.vertexCount(), instanceCount);
const size_t sourceStride = ComputeVertexAttributeStride(attrib);
const size_t destStride = ComputeVertexAttributeTypeSize(attrib);
const uint8_t *inputPointer = reinterpret_cast<const uint8_t *>(attrib.pointer);
// Pack the data when copying it, user could have supplied a very large stride that
// would cause the buffer to be much larger than needed.
if (destStride == sourceStride)
{
// Can copy in one go, the data is packed
memcpy(bufferPointer + curBufferOffset,
inputPointer + (sourceStride * indexRange.start),
destStride * streamedVertexCount);
}
else
{
// Copy each vertex individually
for (size_t vertexIdx = 0; vertexIdx < streamedVertexCount; vertexIdx++)
{
uint8_t *out = bufferPointer + curBufferOffset + (destStride * vertexIdx);
const uint8_t *in =
inputPointer + sourceStride * (vertexIdx + indexRange.start);
memcpy(out, in, destStride);
}
}
// Compute where the 0-index vertex would be.
const size_t vertexStartOffset = curBufferOffset - (indexRange.start * destStride);
if (attrib.pureInteger)
{
ASSERT(!attrib.normalized);
mFunctions->vertexAttribIPointer(
idx, attrib.size, attrib.type, static_cast<GLsizei>(destStride),
reinterpret_cast<const GLvoid *>(vertexStartOffset));
}
else
{
mFunctions->vertexAttribPointer(
idx, attrib.size, attrib.type, attrib.normalized,
static_cast<GLsizei>(destStride),
reinterpret_cast<const GLvoid *>(vertexStartOffset));
}
curBufferOffset += destStride * streamedVertexCount;
// Mark the applied attribute as dirty by setting an invalid size so that if it doesn't
// need to be streamed later, there is no chance that the caching will skip it.
mAppliedAttributes[idx].size = static_cast<GLuint>(-1);
}
unmapResult = mFunctions->unmapBuffer(GL_ARRAY_BUFFER);
}
if (unmapResult != GL_TRUE)
{
return Error(GL_OUT_OF_MEMORY, "Failed to unmap the client data streaming buffer.");
}
return Error(GL_NO_ERROR);
}