@ -13,6 +13,7 @@
# include <Atom/RHI.Reflect/Bits.h> # include <Atom/RHI.Reflect/Bits.h>
# include <AzCore/Debug/EventTrace.h> # include <AzCore/Debug/EventTrace.h>
# include <AzCore/std/algorithm.h>
# include <RHI/ArgumentBuffer.h> # include <RHI/ArgumentBuffer.h>
# include <RHI/Buffer.h> # include <RHI/Buffer.h>
# include <RHI/BufferMemoryView.h> # include <RHI/BufferMemoryView.h>
@ -249,66 +250,191 @@ namespace AZ
ShaderResourceBindings & bindings = GetShaderResourceBindingsByPipelineType ( stateType ) ;
ShaderResourceBindings & bindings = GetShaderResourceBindingsByPipelineType ( stateType ) ;
const PipelineLayout & pipelineLayout = pipelineState - > GetPipelineLayout ( ) ;
const PipelineLayout & pipelineLayout = pipelineState - > GetPipelineLayout ( ) ;
for ( uint32_t srgIndex = 0 ; srgIndex < RHI : : Limits : : Pipeline : : ShaderResourceGroupCountMax ; + + srgIndex )
uint32_t bufferVertexRegisterIdMin = RHI : : Limits : : Pipeline : : ShaderResourceGroupCountMax ;
uint32_t bufferFragmentOrComputeRegisterIdMin = RHI : : Limits : : Pipeline : : ShaderResourceGroupCountMax ;
uint32_t bufferVertexRegisterIdMax = 0 ;
uint32_t bufferFragmentOrComputeRegisterIdMax = 0 ;
//Arrays to cache all the buffers and offsets in order to make batch calls
MetalArgumentBufferArray mtlVertexArgBuffers ;
MetalArgumentBufferArrayOffsets mtlVertexArgBufferOffsets ;
MetalArgumentBufferArray mtlFragmentOrComputeArgBuffers ;
MetalArgumentBufferArrayOffsets mtlFragmentOrComputeArgBufferOffsets ;
mtlVertexArgBuffers . fill ( nil ) ;
mtlFragmentOrComputeArgBuffers . fill ( nil ) ;
mtlVertexArgBufferOffsets . fill ( 0 ) ;
mtlFragmentOrComputeArgBufferOffsets . fill ( 0 ) ;
for ( uint32_t slot = 0 ; slot < RHI : : Limits : : Pipeline : : ShaderResourceGroupCountMax ; + + slot )
{
{
const ShaderResourceGroup * shaderResourceGroup = bindings . m_srgsBySlot [ srgIndex ] ;
const ShaderResourceGroup * shaderResourceGroup = bindings . m_srgsBySlot [ s lot ] ;
uint32_t slotIndex = pipelineLayout . GetSlotByIndex ( srgIndex ) ;
uint32_t slotIndex = pipelineLayout . Get IndexBySlot( slot ) ;
if ( ! shaderResourceGroup | | slotIndex = = RHI : : Limits : : Pipeline : : ShaderResourceGroupCountMax )
if ( ! shaderResourceGroup | | slotIndex = = RHI : : Limits : : Pipeline : : ShaderResourceGroupCountMax )
{
{
continue ;
continue ;
}
}
if ( bindings . m_srgsByIndex [ srgIndex ] ! = shaderResourceGroup )
uint32_t srgVisIndex = pipelineLayout . GetIndexBySlot ( shaderResourceGroup - > GetBindingSlot ( ) ) ;
const RHI : : ShaderStageMask & srgVisInfo = pipelineLayout . GetSrgVisibility ( srgVisIndex ) ;
bool isSrgUpdatd = bindings . m_srgsByIndex [ slot ] ! = shaderResourceGroup ;
if ( isSrgUpdatd )
{
{
bindings . m_srgsByIndex [ srgIndex ] = shaderResourceGroup ;
bindings . m_srgsByIndex [ s lot ] = shaderResourceGroup ;
auto & compiledArgBuffer = shaderResourceGroup - > GetCompiledArgumentBuffer ( ) ;
auto & compiledArgBuffer = shaderResourceGroup - > GetCompiledArgumentBuffer ( ) ;
id < MTLBuffer > argBuffer = compiledArgBuffer . GetArgEncoderBuffer ( ) ;
id < MTLBuffer > argBuffer = compiledArgBuffer . GetArgEncoderBuffer ( ) ;
size_t argBufferOffset = compiledArgBuffer . GetOffset ( ) ;
size_t argBufferOffset = compiledArgBuffer . GetOffset ( ) ;
uint32_t srgVisIndex = pipelineLayout . GetSlotByIndex ( shaderResourceGroup - > GetBindingSlot ( ) ) ;
const RHI : : ShaderStageMask & srgVisInfo = pipelineLayout . GetSrgVisibility ( srgVisIndex ) ;
if ( srgVisInfo ! = RHI : : ShaderStageMask : : None )
if ( srgVisInfo ! = RHI : : ShaderStageMask : : None )
{
{
const ShaderResourceGroupVisibility & srgResourcesVisInfo = pipelineLayout . GetSrgResourcesVisibility ( srgVisIndex ) ;
//For graphics and compute shader stages, cache all the argument buffers, offsets and track the min/max indices
//For graphics and compute encoder bind the argument buffer and
//make the resource resident for the duration of the work associated with the current scope
//and ensure that it's in a format compatible with the appropriate metal function.
if ( m_commandEncoderType = = CommandEncoderType : : Render )
if ( m_commandEncoderType = = CommandEncoderType : : Render )
{
{
id < MTLRenderCommandEncoder > renderEncoder = GetEncoder < id < MTLRenderCommandEncoder > > ( ) ;
id < MTLRenderCommandEncoder > renderEncoder = GetEncoder < id < MTLRenderCommandEncoder > > ( ) ;
uint8_t numBitsSet = RHI : : CountBitsSet ( static_cast < uint64_t > ( srgVisInfo ) ) ;
uint8_t numBitsSet = RHI : : CountBitsSet ( static_cast < uint64_t > ( srgVisInfo ) ) ;
if ( numBitsSet > 1 | | srgVisInfo = = RHI : : ShaderStageMask : : Vertex )
if ( numBitsSet > 1 | | srgVisInfo = = RHI : : ShaderStageMask : : Vertex )
{
{
[ renderEncoder setVertexBuffer : argBuffer
mtlVertexArgBuffers [ slotIndex ] = argBuffer ;
offset : argBufferOffset
mtlVertexArgBufferOffsets [ slotIndex ] = argBufferOffset ;
atIndex : slotIndex ] ;
bufferVertexRegisterIdMin = AZStd : : min ( slotIndex , bufferVertexRegisterIdMin ) ;
bufferVertexRegisterIdMax = AZStd : : max ( slotIndex , bufferVertexRegisterIdMax ) ;
}
}
if ( numBitsSet > 1 | | srgVisInfo = = RHI : : ShaderStageMask : : Fragment )
if ( numBitsSet > 1 | | srgVisInfo = = RHI : : ShaderStageMask : : Fragment )
{
{
[ renderEncoder setFragmentBuffer : argBuffer
mtlFragmentOrComputeArgBuffers [ slotIndex ] = argBuffer ;
offset : argBufferOffset
mtlFragmentOrComputeArgBufferOffsets [ slotIndex ] = argBufferOffset ;
atIndex : slotIndex ] ;
bufferFragmentOrComputeRegisterIdMin = AZStd : : min ( slotIndex , bufferFragmentOrComputeRegisterIdMin ) ;
bufferFragmentOrComputeRegisterIdMax = AZStd : : max ( slotIndex , bufferFragmentOrComputeRegisterIdMax ) ;
}
}
}
else if ( m_commandEncoderType = = CommandEncoderType : : Compute )
{
mtlFragmentOrComputeArgBuffers [ slotIndex ] = argBuffer ;
mtlFragmentOrComputeArgBufferOffsets [ slotIndex ] = argBufferOffset ;
bufferFragmentOrComputeRegisterIdMin = AZStd : : min ( slotIndex , bufferFragmentOrComputeRegisterIdMin ) ;
bufferFragmentOrComputeRegisterIdMax = AZStd : : max ( slotIndex , bufferFragmentOrComputeRegisterIdMax ) ;
}
}
}
//Check if the srg has been updated or if the srg resources visibility hash has been updated
//as it is possible for draw items to have different PSOs in the same pass.
const AZ : : HashValue64 srgResourcesVisHash = pipelineLayout . GetSrgResourcesVisibilityHash ( srgVisIndex ) ;
if ( bindings . m_srgVisHashByIndex [ slot ] ! = srgResourcesVisHash | | isSrgUpdatd )
{
bindings . m_srgVisHashByIndex [ slot ] = srgResourcesVisHash ;
if ( srgVisInfo ! = RHI : : ShaderStageMask : : None )
{
const ShaderResourceGroupVisibility & srgResourcesVisInfo = pipelineLayout . GetSrgResourcesVisibility ( srgVisIndex ) ;
//For graphics and compute encoder make the resource resident (call UseResource) for the duration
//of the work associated with the current scope and ensure that it's in a
//format compatible with the appropriate metal function.
if ( m_commandEncoderType = = CommandEncoderType : : Render )
{
shaderResourceGroup - > AddUntrackedResourcesToEncoder ( m_encoder , srgResourcesVisInfo ) ;
shaderResourceGroup - > AddUntrackedResourcesToEncoder ( m_encoder , srgResourcesVisInfo ) ;
}
}
else if ( m_commandEncoderType = = CommandEncoderType : : Compute )
else if ( m_commandEncoderType = = CommandEncoderType : : Compute )
{
{
id < MTLComputeCommandEncoder > computeEncoder = GetEncoder < id < MTLComputeCommandEncoder > > ( ) ;
[ computeEncoder setBuffer : argBuffer
offset : argBufferOffset
atIndex : pipelineLayout . GetSlotByIndex ( srgIndex ) ] ;
shaderResourceGroup - > AddUntrackedResourcesToEncoder ( m_encoder , srgResourcesVisInfo ) ;
shaderResourceGroup - > AddUntrackedResourcesToEncoder ( m_encoder , srgResourcesVisInfo ) ;
}
}
}
}
}
}
}
}
//For graphics and compute encoder bind all the argument buffers
if ( m_commandEncoderType = = CommandEncoderType : : Render )
{
BindArgumentBuffers ( RHI : : ShaderStage : : Vertex ,
bufferVertexRegisterIdMin ,
bufferVertexRegisterIdMax ,
mtlVertexArgBuffers ,
mtlVertexArgBufferOffsets ) ;
BindArgumentBuffers ( RHI : : ShaderStage : : Fragment ,
bufferFragmentOrComputeRegisterIdMin ,
bufferFragmentOrComputeRegisterIdMax ,
mtlFragmentOrComputeArgBuffers ,
mtlFragmentOrComputeArgBufferOffsets ) ;
}
else if ( m_commandEncoderType = = CommandEncoderType : : Compute )
{
BindArgumentBuffers ( RHI : : ShaderStage : : Compute ,
bufferFragmentOrComputeRegisterIdMin ,
bufferFragmentOrComputeRegisterIdMax ,
mtlFragmentOrComputeArgBuffers ,
mtlFragmentOrComputeArgBufferOffsets ) ;
}
return true ;
return true ;
}
}
void CommandList : : BindArgumentBuffers ( RHI : : ShaderStage shaderStage ,
uint16_t registerIdMin ,
uint16_t registerIdMax ,
MetalArgumentBufferArray & mtlArgBuffers ,
MetalArgumentBufferArrayOffsets mtlArgBufferOffsets )
{
//Metal Api only lets you bind multiple argument buffers in an array as long as there are no gaps in the array
//In order to accomodate that we break up the calls when a gap is noticed in the array and reconfigure the NSRange.
uint16_t startingIndex = registerIdMin ;
bool trackingRange = true ;
for ( int i = registerIdMin ; i < = registerIdMax + 1 ; i + + )
{
if ( trackingRange )
{
if ( mtlArgBuffers [ i ] = = nil )
{
NSRange range = { startingIndex , i - startingIndex } ;
switch ( shaderStage )
{
case RHI : : ShaderStage : : Vertex :
{
id < MTLRenderCommandEncoder > renderEncoder = GetEncoder < id < MTLRenderCommandEncoder > > ( ) ;
[ renderEncoder setVertexBuffers : & mtlArgBuffers [ startingIndex ]
offsets : & mtlArgBufferOffsets [ startingIndex ]
withRange : range ] ;
break ;
}
case RHI : : ShaderStage : : Fragment :
{
id < MTLRenderCommandEncoder > renderEncoder = GetEncoder < id < MTLRenderCommandEncoder > > ( ) ;
[ renderEncoder setFragmentBuffers : & mtlArgBuffers [ startingIndex ]
offsets : & mtlArgBufferOffsets [ startingIndex ]
withRange : range ] ;
break ;
}
case RHI : : ShaderStage : : Compute :
{
id < MTLComputeCommandEncoder > computeEncoder = GetEncoder < id < MTLComputeCommandEncoder > > ( ) ;
[ computeEncoder setBuffers : & mtlArgBuffers [ startingIndex ]
offsets : & mtlArgBufferOffsets [ startingIndex ]
withRange : range ] ;
break ;
}
default :
{
AZ_Assert ( false , " Not supported " ) ;
}
}
trackingRange = false ;
}
}
else
{
if ( mtlArgBuffers [ i ] ! = nil )
{
startingIndex = i ;
trackingRange = true ;
}
}
}
}
void CommandList : : Submit ( const RHI : : DrawItem & drawItem )
void CommandList : : Submit ( const RHI : : DrawItem & drawItem )
{
{
@ -447,6 +573,7 @@ namespace AZ
for ( size_t i = 0 ; i < bindings . m_srgsByIndex . size ( ) ; + + i )
for ( size_t i = 0 ; i < bindings . m_srgsByIndex . size ( ) ; + + i )
{
{
bindings . m_srgsByIndex [ i ] = nullptr ;
bindings . m_srgsByIndex [ i ] = nullptr ;
bindings . m_srgVisHashByIndex [ i ] = AZ : : HashValue64 { 0 } ;
}
}
const PipelineLayout & pipelineLayout = pipelineState - > GetPipelineLayout ( ) ;
const PipelineLayout & pipelineLayout = pipelineState - > GetPipelineLayout ( ) ;
@ -469,6 +596,10 @@ namespace AZ
void CommandList : : SetStreamBuffers ( const RHI : : StreamBufferView * streams , uint32_t count )
void CommandList : : SetStreamBuffers ( const RHI : : StreamBufferView * streams , uint32_t count )
{
{
uint16_t bufferArrayLen = 0 ;
AZStd : : array < id < MTLBuffer > , METAL_MAX_ENTRIES_BUFFER_ARG_TABLE > mtlStreamBuffers ;
AZStd : : array < NSUInteger , METAL_MAX_ENTRIES_BUFFER_ARG_TABLE > mtlStreamBufferOffsets ;
AZ : : HashValue64 streamsHash = AZ : : HashValue64 { 0 } ;
AZ : : HashValue64 streamsHash = AZ : : HashValue64 { 0 } ;
for ( uint32_t i = 0 ; i < count ; + + i )
for ( uint32_t i = 0 ; i < count ; + + i )
{
{
@ -479,18 +610,25 @@ namespace AZ
{
{
m_state . m_streamsHash = streamsHash ;
m_state . m_streamsHash = streamsHash ;
AZ_Assert ( count < = METAL_MAX_ENTRIES_BUFFER_ARG_TABLE , " Slots needed cannot exceed METAL_MAX_ENTRIES_BUFFER_ARG_TABLE " ) ;
AZ_Assert ( count < = METAL_MAX_ENTRIES_BUFFER_ARG_TABLE , " Slots needed cannot exceed METAL_MAX_ENTRIES_BUFFER_ARG_TABLE " ) ;
for ( uint32_t i = 0 ; i < count ; + + i )
NSRange range = { METAL_MAX_ENTRIES_BUFFER_ARG_TABLE - count , count } ;
//The stream buffers are populated from bottom to top as the top slots are taken by argument buffers
for ( int i = count - 1 ; i > = 0 ; - - i )
{
{
if ( streams [ i ] . GetBuffer ( ) )
if ( streams [ i ] . GetBuffer ( ) )
{
{
const Buffer * buff = static_cast < const Buffer * > ( streams [ i ] . GetBuffer ( ) ) ;
const Buffer * buff = static_cast < const Buffer * > ( streams [ i ] . GetBuffer ( ) ) ;
id < MTLBuffer > mtlBuff = buff - > GetMemoryView ( ) . GetGpuAddress < id < MTLBuffer > > ( ) ;
id < MTLBuffer > mtlBuff = buff - > GetMemoryView ( ) . GetGpuAddress < id < MTLBuffer > > ( ) ;
uint32_t VBIndex = ( METAL_MAX_ENTRIES_BUFFER_ARG_TABLE - 1 ) - i ;
uint32_t offset = streams [ i ] . GetByteOffset ( ) + buff - > GetMemoryView ( ) . GetOffset ( ) ;
uint32_t offset = streams [ i ] . GetByteOffset ( ) + buff - > GetMemoryView ( ) . GetOffset ( ) ;
id < MTLRenderCommandEncoder > renderEncoder = GetEncoder < id < MTLRenderCommandEncoder > > ( ) ;
mtlStreamBuffers [ bufferArrayLen ] = mtlBuff ;
[ renderEncoder setVertexBuffer : mtlBuff offset : offset atIndex : VBIndex ] ;
mtlStreamBufferOffsets [ bufferArrayLen ] = offset ;
bufferArrayLen + + ;
}
}
}
}
id < MTLRenderCommandEncoder > renderEncoder = GetEncoder < id < MTLRenderCommandEncoder > > ( ) ;
[ renderEncoder setVertexBuffers : mtlStreamBuffers . data ( )
offsets : mtlStreamBufferOffsets . data ( )
withRange : range ] ;
}
}
}
}
moudgils
5 years ago
committed by
GitHub