Mantle and Nitrous - Combining Efficient Engine Design with a modern API - AMD at GDC14

NITROUS AND MANTLE:
Combining efficient engine design with a modern API
Dan Baker, Partner, Oxide Games

2 | Nitrous and Mantle | 19 March 2014
PRE-REQUISITE MOTIVATIONAL SLIDE
MODERN APIS ARE
STARTING TO FEEL
RATHER DATED
BUT HOW MUCH
BETTER CAN WE
BE?

PRE-REQUISITE MOTIVATIONAL SLIDE
TURNS OUT… A
WHOLE LOT
FASTER

HONEY, DOES THIS DRESS MAKE ME LOOK FAT?
…

STATE OF THE ART TODAY: WHAT’S GOING ON?
Lots of little things add up
2 major problems require rearchitecture
–Functional threading model throws a wrench into task
based systems
–Implicit Hazard tracking and synchronization
API tries to hide the async nature of GPU
Lots of little things, memory model, binding model, etc
Analysis of features like instancing indicate that it is
unreliable and tends to speed up only the fastest
frames, correlation between batches and driver perf is
casual
Can’t RETRO fit old APIS

DIVING INTO NITROUS
Nitrous = Oxide’s custom engine
Specifically designed for high throughput
Core neutral. Main thread acts only as lightweight
sequencer
All work divided up into small jobs, which are in the
microsecond range
Can produce lots of jobs, 10,000+ range per frame

STAR SWARM
 Nitrous Engine demo
 Free to download,
experiment
 Proof of concept for
modern API design
 Represents 2 AI
opponents, thus
application CPU load
is realistic
 10,000 units possible
 100,000+ batches
possible

SECRETS BEHIND STAR SWARM
Much of what is required for high performance isn’t specific to
Mantle
Star Swarm originally not based on Mantle
If engine is structured in certain ways, Mantle support is
straight-forward and intuitive. Maybe even fun.
Work done to restructure engine will have benefits outside of
Mantle support

ADDING NITROUS TO THE ENGINE
 Rendering broken into jobs which generate autonomous command buffers
 CPU to GPU data streamlined – constants, texture updates go into GPU frame memory
 Shader bindings standardized
 Shaders, state, bundled into blocks
 Resources grouped into sets
 Graphics commands streamlined, restricted bind points
 Stateless command format
 Expensive state transitioned rarely
 Much attention paid to cache usage, lockless data structures
 All hazards detangled, all buffers considered non persistent

MULTI-CORE CPU BASICS
Be Wary, There Is A Lot Of Very Bad Advice In The Wild
Spawning threads to handle tasks
Relying OS preemptive scheduler, heavy weight OS synchronization primitives
Functional threading in general
Your Survival Guide
OK: Multi-thread read of same location
OK: Multi-thread write to different locations
OK: Multi-thread write to same location in ‘stamp’ mode
CAUTION: Atomic instructions
STOP: Multi-thread read/write to same location
STOP: Multi-thread write to same CACHE line

NITROUS AND MANTLE
Nitrous is NOT built around Mantle
Reverse is more true, Mantle adapts well to Nitrous
internal concepts
The concepts are what make engine fast
Results are astounding, driver time reduced up to
50x
Mantle is the harbinger of future API design, Not just
in Graphics

TASK BASED SYSTEM
 Idea is that work load is a constructed graph of much
smaller nuggets
 Many advantages
– Scales well, 32+ cores
– Easy to balance workload
– More power efficient – more slower cores just as good
 Already seeing CPUs dynamically slowing clock speed
– If enough similar work items queued, can execute same
code on cores
 Cache hit rate much higher
– End up generating a larger number of command buffers
to prevent thread serialization

HOW NITROUS GENERATES COMMANDS
Core 0 Core 1 Core 2
Cmd
Buffer 0
Cmd
Buffer 1
Cmd
Buffer 2
Cmd
Buffer 3
Frame Data
Cmd
Buffer 2
Cmd
Buffer 0
Cmd
Buffer 1
Cmd
Buffer 3

NITROUS COMMAND FORMATS
 In reality, diagram is over simplified
 Nitrous has it’s own internal command format
– Small, efficient commands
– Stateless, each command contains references to all needed state
– Inheritance unneeded
– Separates internal graphics system from any particular API
 Being Stateless, can be generated completely out of order
 Entire Frame is queued up in internal command format
 Frame is translated to GPU commands via Mantle
– Nitrous Command buffers are translated into Mantle Command Buffers at one section
 Get’s more optimal use out of instruction cache and data cache

BUILDING AROUND ASYNCRONISITY: HOW NITROUS THINKS OF A FRAME
 Entire app should be exposed to concept of asyncronisity
 The concept of a frame:
– A set of commands which will be executed on the GPU
– A set of data which will be read by the GPU
– This concept is fundamental in Nitrous, regardless of API
Frame
CMD CMD CMD CMD
Frame Data
Persistent
Textures Big
Transfer
Buffers
Resource
Sets

CREATING A FRAME, USING FRAME DATA
 Create 2 copies of our frame data
 One will be read by GPU, while
other is being written to by the CPU
 Must use fence to make sure CPU
doesn’t get ahead
 More complex situations could be
explored
 Frame data includes
– Constant Data
– Small texture updates
Even Frame
Odd Frame
GPU
CPU

STARTING OUR FRAME
g_fTotalFrameTime = System::Time::TimeAsSeconds( System::Time::PeekTime()) - g_StartTotalFrameTime ;
g_uFrameBuffer = (g_uFrame % TransferMemory::BUFFERS);
if(g_bProtectMemory)
ProtectMemory(g_CmdTransferMemory.PerFrameMemory[g_uFrameBuffer].pData, g_CmdTransferMemory.Capacity, PO_READWRITE);
gr = grWaitForFences(g_Device, 1, &g_FrameFences[g_uFrameBuffer], true, MAX_FRAME_WAIT_TIME_SECONDS);
if(gr == GR_TIMEOUT) // Throw the frame out
return false;
g_StartTotalFrameTime = System::Time::TimeAsSeconds( System::Time::PeekTime());
//Reset our allocation and map the current GPU memory
HeapAndChunk HeapChunk = g_GPUTransferMemory.PerFrameMemory[g_uFrameBuffer].Memory;
MarkMemoryChunk(HeapChunk);
GR_GPU_MEMORY DynamicMemory = g_MemoryHeap.MemoryChunk[HeapChunk.Heap][HeapChunk.SubChunk].Memory;
gr = grMapMemory(DynamicMemory, 0, (GR_VOID**) &g_GPUTransferMemory.PerFrameMemory[g_uFrameBuffer].pData);
g_CmdTransferMemory.PerFrameMemory[g_uFrameBuffer ].Offset = 0;
g_GPUTransferMemory.PerFrameMemory[g_uFrameBuffer ].Offset = 0;
g_GPUTransferMemory.pCurrentMantleMemoryBase = g_GPUTransferMemory.PerFrameMemory[g_uFrameBuffer].pData;
g_GPUTransferMemory.pCurrentMantleMemoryEnd = g_GPUTransferMemory.PerFrameMemory[g_uFrameBuffer].pData +
g_GPUTransferMemory.Capacity;
g_GPUTransferMemory.CurrentMantleMemory = DynamicMemory;
g_GPUTransferMemory.TempHeapMemory = 0;
return true;

HINT
Use memory heap that has highest cpuWritePerfRating
In Debug, rather then copying directly to GPU memory,
allocate CPU memory
–Or use pinned Mantle memory
Then, use OS call Virtual Protect with PAGE_NOACCESS for
any data that effects the frame, while the frame is being
accessed by GPU, or could be being translated by the CPU
If any part of system inadvertently writes to the memory, will
throw exception

SOME EXTRA STUFF WE WILL NEED
Because we track hazards, we will want a few more buffers
A delete queue – objects are not deleted, but placed in the delete queue
–One queue per frame, once that frame is complete, items will be deleted
A state transition queue
–Used only when a resource is created, to transition it to the desired
initial state
An Unordered Command Queue
–Gets flushed before main frames command queue
–Useful for preparing resources for first time use (e.g. initialization)

INTERNAL COMMAND FORMAT
 Nitrous has it’s own internal command format
 Persistent state:
– Resource Sets
– Shader Blocks
– Various pipeline state
 Frame State, primary construct is a batch set
– Contains primitives, batches and shader sets
– Batches which reference
 Primitives
 Shader Sets
– Constant references are made into our frame memory
 Each one of these has a different, natural change frequency

NITROUS MEMORY POOLS
Resources used together, created together
Multiple resource sets are often pooled
Simplifies memory management, less then
1000 total allocations
Orange Team Unit’s Memory
FIGHTER 1 CAR. REAR
CAR. FOR CARRIER MAIN
(0) Albiedo
(1) Material Mask
(2) Ambient
Occlusion
(3) Normal Map
(4) Weathering Map
(0) Albiedo
(1) Material Mask
(2) Ambient
Occlusion
(3) Normal Map
(4) Weathering Map
(0) Albiedo
(1) Material Mask
(2) Ambient
Occlusion
(3) Normal Map
(4) Weathering Map
(0) Albiedo
(1) Material Mask
(2) Ambient
Occlusion
(3) Normal Map
(4) Weathering Map

NITROUS MEMORY POOLS
 GPU resource allocation a little tricky – we don’t
know ahead of time how big something might be
 2 step process, first calculate size of resource,
then allocate pool based on that size
 Does not map 1:1 to Mantle memory allocations
 Instead, Pool is created with default page size
 When a new resource is added, either it places
inside current allocation, or if resource is bigger
then the page size, creates a new allocation that
fits the resource
 A memory pool in Nitrous = a list of allocations in
Mantle
 If able to size ahead of time, only 1 allocation
Unit Textures
Diffuse
Specular
Mask
AO
Normal
Mantle
Alloc
Mantle
Alloc
Mantle
Alloc

CREATING A RESOURCE
//Setup our resource decriptions
ResourceDesc RedLUTDesc, BlueLUTDesc, GreenLUTDesc;
RedLUTDesc.Init(OX_FORMAT_R32_FLOAT, 1, 1, RT_TEXTURE_1D, v3ui(1024,1,1), pfRedData, 0);
BlueLUTDesc.Init(OX_FORMAT_R32_FLOAT, 1, 1, RT_TEXTURE_1D, v3ui(1024,1,1), pfBlueData, 0);
GreenLUTDesc.Init(OX_FORMAT_R32_FLOAT, 1, 1, RT_TEXTURE_1D, v3ui(1024,1,1), pfGreenData, 0);
//Calculate Size
uint64 uSize = 0;
uSize += GetResourceMemorySize(RedLUTDesc, Graphics::RF_SHADERRESOURCE, Graphics::HT_GPU);
uSize += GetResourceMemorySize(BlueLUTDesc, Graphics::RF_SHADERRESOURCE, Graphics::HT_GPU);
uSize += GetResourceMemorySize(GreenLUTDesc, Graphics::RF_SHADERRESOURCE, Graphics::HT_GPU);
//Create a GPU heap, sized to what we want
g_GPUMemory = AllocateHeap(uSize, Graphics::HT_GPU);
//Create the Resources
ColorLuts.RedLUT.Resource = CreateResource(RedLUTDesc, RF_SHADERRESOURCE, RSTATE_DEFAULT, g_GPUMemory);
ColorLuts.BlueLUT.Resource = CreateResource(BlueLUTDesc, RF_SHADERRESOURCE, RSTATE_DEFAULT, g_GPUMemory);
ColorLuts.GreenLUT.Resource = CreateResource(GreenLUTDesc, RF_SHADERRESOURCE, RSTATE_DEFAULT, g_GPUMemory);

SOME EXTRA MANAGEMENT REQUIRED
 Creating a Resource slightly more involved
 When a resource creation call occurs, check to see if we are a GPU heap
 If so, no way to directly map memory and upload resource so
– 1) Allocate, or recycle a CPU visible heap object
– 2) Create Resource and map into this heap
– 3) Create Resource on the GPU in the specified heap, (it will be uninitialized)
– 4) Issue a copy command in our Unordered Command Queue
– 5) Place temp resource in a deletion queue
 For any resource, we allow a default state to be specified
– At beginning of frame, before we execute main comands, issue any state transition queues to place
resources from default state into desired state

RESOURCE SETS
 In real world, textures are grouped
 Nitrous has 5 bind points
– 2 for batch
– 2 for shader
– 1 for primitive
 VB is just a resource set
 Nitrous does not allow binding of individual
textures
 Clearly, maps 1:1 to a descriptor
Space Fighter 1
(0) Albiedo
(1) Material Mask
(2) Ambient Occlusion
(3) Normal Map
(4) Weathering Map

VERTEX BUFFERS
 Nitrous does not use Vertex Buffers
 Instead, Resource Set acts as VB, but with more programmatic control
 Vastly simplifies engine side management
– VBs can be saved as DDS files
– Do not require a huge amount of loading code for slightly different Vertex Formats
– Can fold Displacement maps and other geometry modifiers into Primitive Resource Set
 Not seen strong evidence on any hardware that this causes a performance issue

CONSTANT BUFFERS
 Nitrous does not have concept of constant buffers
 Instead, all constant data is thrown out every frame
– When we render an object, CPU will generate the constants needed for that frame
– Grab a piece of the Frame Memory and write to it
 Constant bindings are just references into our frame memory
 But… be careful! CPU is writing straight to GPU memory. Do NOT read it back!
 Evidence suggests no performance advantage of persisting constants across frames, regenerating every
frame is ample fast. 100k+ batches not a problem

A BATCH IN NITROUS CONSISTS OF 4 PARTS
Batch Set
Prim 0 Prim 1 Prim 2
Shader 0 Shader 1
Batch
0
Batch
1
Batch
2
Batch
3
Batch
4
Primitive
IB
Resources
Tri info
Shader
Resources (2)
Constants (2)
Shader Block
Batch
Primitive
Shader
Resources (2)
Constants (2)
Batch Set
Batches
Primitives
Shaders
RTs
Blend State

DESCRIPTOR TABLE LAYOUT FOR NITROUS
Descriptor 0
*Batch Resource Set 0
*Batch Resource Set 1
Batch Constants 1
Batch Constants 2
*Shader Resource Set 0
*Shader Resource Set 1
Shader Constants 0
Shader Constants 1
*UAV
*Samplers (only 1 global bank)
Descriptor 1
*Primitive VB
Dynamic Const
Batch Constants 0

DESCRIPTOR BINDING STRATEGY
 Remember: Descriptors are just structures on GPU memory, so need to double buffer as well
 Create 1 giant descriptor table, start update at beginning of frame
 Recognize that we have a resource bind vector of only 9 items
 Each bind vector can be built into a descriptor table, but don’t need unique one
 Check to see if this bind vector has been built before(During this frame), e.g. resident in a small cache, if
so, just reference it
 If not, build a new descriptor table, and place in cache
 Dynamic constants, batch constant 0, uses grCmdBindDynamicMemoryView
– Usually, this will change every call (e.g. some part of the batch is changing or else it’s the same batch)
 Using grCmdBindDynamicMemoryView, for 100k batches, about 5-10k descriptors actually need to get
built per frame

TRACKING RESOURCE USAGE
 Apps responsibility to track what resources get used
 Simple strategy: Stamp a frame number on each
memory pool anytime it is bound
 Traverse the complete resource list, anything which
matches current frame must be resident
 Quick as long as we keep # of heaps reasonable
 Important: Frame # should be padded into a cache line
to avoid serialization
Heap description Last Frame Used
UI Textures intro 2401
UI Textures in Game 17204
Orange Faction Units 17204
Purple Faction Units 17204
Weapon effects 16392
Post Process RTs 17204
Terrain Heightmap 17204

DEALING WITH STATE TRANSITIONS
 Most important, difficult part of Mantle
 Must understand anytime a resource is getting used in a different way,
 Read After Write
 Write After Write

SHADER BLOCKS
 Shader Blocks
– Group of shaders with identical resources
– Key point : all shader stages grouped together
– All resources are bound to all stages
– For mantle, need add some extra data
 Can we blend?
 What back buffer formats might be used?
 What z buffer formats might be used?
– Create a matrix of pipeline objects based on specified
modes
 The right pipeline objet is selected based on current RT state
 RTs and blendstate already chunked, no extra state changes
introduced
ShaderGroup SimpleShader
{
ResourceSetPrimitive = VertexData;
ConstantSetDynamic[0] = DynamicData;
ResourceSetBatch[1] = UserTS;
ConstantSetShader[0] = Globals;
RenderTargetFormats = R16G16B16A16_FLOAT,
R11G11B10_FLOAT;
BlendStates = BlendOff;
DepthTargetFormats = D32_FLOAT;
Methods
{
main:
CodeBlocks = SimpleShaders;
VertexShader = SimpleVSShader;
PixelShader = SimplePSShader;
zprime:
CodeBlocks = SimpleShaders;
VertexShader = SimpleVSShader;
PixelShader = BlankSimplePSShader;
}
}

CREATING SHADER BLOCKS IN MANTLE
 Translate HLSL Byte code to Mantle IC
– All done at compile time, have a Mantle speific executable
 Creating a Mapping Table
– Batch has 5 bind points
– Shader has 4 bind points
– Batch Set has 1 bind point
– Primitive has 1 bind point
– Global Samplers have 1 bind point
 Set up our IC so all pipeline objects use exactly the same top level desciptor

WHAT ABOUT THAT PRESENT?
 Unlike other APIS, we do not need, or should, block on the present on the main thread
 Instead we spawn a job, which we block against on the next present
Void PresentJob()
{
…
result = grQueueSubmit(g_UniversalQueue, g_cCommandBuffers, g_CommandBuffers,
cMemRefs, MemRef, g_FrameFences[g_uSubmittingFrameBuffer]);
uint32 PresentFlags = 0;
if(g_bVSync)
PresentFlags = GR_WSI_WIN_PRESENT_FLIP_DONOTWAIT;
// instruct the GPU to present the backbuffer in the applications window
GR_WSI_WIN_PRESENT_INFO presentInfo =
{
g_hWnd, g_MasterResourceList.Images[DR_BACKBUFFER],
GR_WSI_WIN_PRESENT_MODE_BLT, 0, PresentFlags
};
result = grWsiWinQueuePresent(g_UniversalQueue, &presentInfo);
SignalProcessAndPresentDone(pInfo);
}
}

WHAT OUR FRAME SUBMISSION LOOKS LIKE
1) Block on last frames present’s job (e.g. NOT the fence, the actual job we spawned)
2) Process and pending resource transitions from newly created resources
3) Generate all pending unordered commands, by generating into 1 or more cmd buffers
4) Send signals to the issuers of unordered commands, to notify them the commands are submiitted
5) Begin translation of Nitrous cmds into Mantle cmds – usually 100-500 jobs across all cores
6) Flush the deletion queues for this frame (likely a few frames old at this point)
7) Any item in our master deletion queue, add to the now empty deletion queue for this frame
8) Handle memory readbacks
9) Spawn Present job

FUTURE WORK
 Now have explicit control over Multi GPU
 Can write better MGPU solutions, like split screen which will not increase latency
– We just got rid of a bunch of latency, don’t want to add it back!
 Asymetric GPU use situations are doable – e.g. using integrated graphics in tandem with Discrete GPU

RESULTS
 Star Swarm surprised both Oxide and AMD
– We were not expecting to see cases where application was 300-400% faster, still room for
optimizations
– Right now, we are clearly GPU bound, will release an update soon that increases CPU utilization a little
bit to optimize GPU, expecting 10-20% more performance out of Mantle on high end GPUs
 Driver overhead very consistent, well correlated to number of calls made
 About 2 man months of work
– For an Alpha API, likely 1 month if final version
 Especially telling on slower CPUs, surprising number of cases with high end GPUS with old CPUs
 Try for yourself: Star Swarm is free to download on Steam!

Mantle and Nitrous - Combining Efficient Engine Design with a modern API - AMD at GDC14

More Related Content

Mantle and Nitrous - Combining Efficient Engine Design with a modern API - AMD at GDC14