m2s-4.1
DESCRIPTION
Using OpenCL in Multi2simTRANSCRIPT
Conference title 1
Rafael Ubal, David Kaeli
Simulation of OpenCL andAPUs on Multi2Sim 4.1
ISCA 2013, Tel-Aviv 2
Introduction
Simulation methodology
Outline
Part 2 – Simulation of a Southern Islands GPU
OpenCL from host to device
Disassembler
Emulation
Timing simulation
Visualization tool
Case study: ND-Range virtualization
Part 3 – Concluding Remarks
Additional Projects
The Multi2Sim Community
Part 1 – Simulation of an x86 CPU
Disassembler
Emulation
Timing simulation
Memory hierarchy
Visualization tool
ISCA 2013, Tel-Aviv 3
IntroductionGetting Started
• User accounts for demos
Follow ourdemos!
─ Machine: fusion1.ece.neu.edu
─ User: isca1, isca2, isca3, ...
─ Password: isca2013
ISCA 2013, Tel-Aviv 4
IntroductionGetting Started
$ ssh isca<N>@fusion1.ece.neu.edu -X
(Notice the X forwarding for later demos using graphics)
• Connect to our server
$ wget http://www.multi2sim.org/files/multi2sim-4.1.tar.gz$ tar -xzf multi2sim-4.1.tar.gz$ cd multi2sim-4.1$ ./configure && make
• Download and compile Multi2Sim
$ lsdemo1 demo2 demo3 demo4 demo5 demo6 demo7 README
All files needed for each demo are present in its corresponding directory.README files describe commands to run and interpretation of outputs.
• Demo descriptions
ISCA 2013, Tel-Aviv 5
IntroductionFirst Execution
#include <stdio.h>
int main(int argc, char **argv){
int i;printf("Number of arguments: %d\n", argc);for (i = 0; i < argc; i++)
printf("\targv[%d] = %s\n", i, argv[i]);return 0;
}
Demo 1
$ test-args hello there
Number of arguments: 4 arg[0] = 'test-args' arg[1] = 'hello' arg[2] = 'there'
$ m2s test-args hello there
< Simulator message in stderr >Number of arguments: 4 arg[0] = 'test-args' arg[1] = 'hello' arg[2] = 'there'< Simulator statistics >
• Source code
• Native execution • Execution on Multi2Sim
ISCA 2013, Tel-Aviv 6
; This is a comment.
[ Section 0 ]Color = RedHeight = 40
[ OtherSection ]Variable = Value
IntroductionSimulator Input/Output Files
• Example of INI file format
• Multi2Sim uses INI file for
─ Configuration files.
─ Output statistics.
─ Statistic summary in standard error output.
ISCA 2013, Tel-Aviv 7
Simulation MethodologyApplication-Only vs. Full-System
Virtual izat ion of User-space subset of ISA System call interface
Full-systemsimulator core
Application-onlysimulator core
Guestprogram 1
Guestprogram 2
Full O.S.
Guestprogram 1
Guestprogram 2
... ...
Virtualizat ion of Complete processor ISA I/O hardware
• Full-system simulation
An entire OS runs on top of the simulator. The simulator models the entire ISA, and virtualizes native hardware devices, similar to a virtual machine. Very accurate simulations, but extremely slow.
• Application-only simulation
Only an application runs on top of the simulator. The simulator implements a subset of the ISA, and needs to virtualize the system call interface (ABI). Multi2Sim falls in this category.
ISCA 2013, Tel-Aviv 8
Simulation MethodologyFour-Stage Simulation Process
• Modular implementation
─ Four clearly different software modules per architecture (x86, MIPS, ...)
─ Each module has a standard interface for stand-alone execution, or interaction with other modules.
Disassembler
Instructionbytes
Instructionfields
Emulator(or functional
simulator)
Run oneinstruction
Instructioninformation
Timingsimulator
(or detailed/architectural)
Pipelinetrace
Visualtool
ExectuableELF file
Instructionsdump
Exectuable file,program arguments
Programoutput
Executable file,program arguments,
processor configuration
Performancestatistics
Userinteraction
Cycle navigation,timing diagrams
ISCA 2013, Tel-Aviv 9
Simulation MethodologyCurrent Architecture Support
• In our latest Multi2Sim SVN repository
─ 4 GPU + 3 CPU architectures supported or in progress.
─ This tutorial will focus on x86 and AMD Southern Islands.
Disasm. EmulationTiming
simulationGraphicpipelines
ARM X In progress – –
MIPS X ––
x86 X X X X
AMD Evergreen X X X X
AMD Southern Islands X X X X
NVIDIA Fermi X In progress ––
NVIDIA Kepler In progress –––
X
ISCA 2013, Tel-Aviv 10
Part 1
Simulation of anx86 CPU
ISCA 2013, Tel-Aviv 11
DisassemblerEmulator
(or functionalsimulator)
Timingsimulator
(or detailed/architectural)
Visualtool
The x86 Disassembler
ISCA 2013, Tel-Aviv 12
08048900 <_start>: 8048900: 31 ed xor ebp,ebp 8048902: 5e pop esi 8048903: 89 e1 mov ecx,esp 8048905: 83 e4 f0 and esp,0xfffffff0 8048908: 50 push eax 8048909: 54 push esp 804890a: 52 push edx 804890b: 68 70 91 04 08 push 0x8049170 ...
• Verification of common output
$ objdump -S -M intel test-args $ m2s --x86-disasm test-args
• GNU x86 disassembler • Multi2Sim x86 disassembler
The x86 DisassemblerMethodology
─ Implementation of an efficient instruction decoder based on lookup tables.
─ When used as a stand-alone tool, the output is provided with exactly the same format as the GNU x86 disassembler for automatic verification.
ISCA 2013, Tel-Aviv 13
The x86 Emulator
DisassemblerEmulator
(or functionalsimulator)
Timingsimulator
(or detailed/architectural)
Visualtool
ISCA 2013, Tel-Aviv 14
The x86 EmulatorProgram Loading
1) Parse ELF executable
─ Read ELF sections and symbols.
─ Initialize code and data.
2) Initialize stack
─ Program headers.
─ Arguments.
─ Environment variables.
3) Initialize registers
─ Program entry → eip
─ Stack pointer → esp
Stack
Program args.Env. variables
mmap region
(not initialized)
Heap
Initialized data
Text
Initialized data0x08000000
0x08xxxxxx
0x40000000
0xc0000000
eax
ebx
eax
ecx
esp
eip
Initial virtual memory image
Initial values for x86 registers
Stac
k po
inte
rIn
stru
ctio
n po
inte
r
ISCA 2013, Tel-Aviv 15
The x86 EmulatorEmulation Loop
• Emulation of x86 instructions
─ Update x86 registers.
─ Update memory map if needed.
─ Example: add [bp+16], 0x5
• Emulation of Linux system calls
─ Analyze system call code and arguments.
─ Update memory map.
─ Update register eax with return value.
─ Example: read(fd, buf, count)
Demo 2
Read instr.at eip
Instr.bytes
Decodeinstruction
Instr.fields
Instr. isint 0x80
No Yes
Emulatesystem call
Emulatex86 instr.
Move eipto next instr.
ISCA 2013, Tel-Aviv 16
The x86 Timing Simulator
DisassemblerEmulator
(or functionalsimulator)
Timingsimulator
(or detailed/architectural)
Visualtool
ISCA 2013, Tel-Aviv 17
The x86 Timing SimulatorSuperscalar Processor
• Superscalar x86 pipelines
─ 6-stage pipeline with configurable latencies.
─ Supported features include speculative execution, branch prediction, micro-instruction generation, trace caches, out-of-order execution, …
─ Modeled structures include fetch queues, reorder buffer, load-store queues, register files, register mapping tables, ...
Instr.cache
Fetch
Tracecache
···
···
Fetchqueue
Tracequeue
Decode ···
Uopqueue
Dispatch
WritebackDatacache
Reg.file
Issue ALU
Commit
Instruction queue
Load/store queue
Reorder buffer··· ···
ISCA 2013, Tel-Aviv 18
The x86 Timing SimulatorMultithreaded and Multicore Processors
• Multicore
─ Fully replicated superscalar pipelines, connected through caches.
─ Running multiple programs concurrently, or one program spawning child threads (using OpenMP, pthread, etc.)
• Multithreading
─ Replicated superscalar pipelines with partially shared resources.
─ Fine-grain, coarse-grain, and simultaneous multithreading.
Demo 3
Superscalar Core
Shared resources:
Reorder bufferInstruction queueLoad/store queue
Register fileFunctional units
Private resources:
Hardware Thread
Program counterRegister aliasing tableTLB
Node 0
Node 1
Node m – 1
Nodes mto 2m - 1
Nodes(n – 1)m
to nm – 1
···
···
···
···
···
···
m threads
n cores
Mem
ory
hier
arch
y
ISCA 2013, Tel-Aviv 19
The x86 Timing SimulatorBenchmark Support
• Multithreaded applications
─ SPLASH-2 benchmark suite with pre-compiled x86 executables and data files available on the website.
─ PARSEC-2.1 with pre-compiled x86 executables and data files.
• Single-threaded applications
─ SPEC 2000 and SPEC 2006 benchmarks are fully supported. Pre-compiled x86 binaries are available on the website.
─ The Mediabench suite includes program binaries and data files, with all you need to run them.
ISCA 2013, Tel-Aviv 20
The Memory HierarchyConfiguration
• Flexible hierarchies
─ Any number of caches organized in any number of levels.
─ Cache levels connected through default cross-bar interconnects, or complex custom interconnect configurations.
─ Each architecture undergoing a timing simulation specifies its ownentry point (cache memory) in the memory hierarchy, for data or instructions.
─ Cache coherence is guaranteed with an implementation of the 5-state MOESI protocol.
ISCA 2013, Tel-Aviv 21
The Memory HierarchyConfiguration Examples
Example 1
Three CPU cores with private L1 caches, two L2 caches, and default cross-bar based interconnects. Cache L2-0 serves physical address range [0, 7ff...ff], and cache L2-1 serves [80...00, ff...ff].
Core 0 Core 1 Core 2
L1-0 L1-1 L1-2
Switch
L2-0 L2-1
Switch
Main Memory
Demo 4
ISCA 2013, Tel-Aviv 22
Example 2
Four CPU cores with private L1 data caches, L1 instruction caches and L2 caches shared every 2 cores (serving the whole address space), and four main memory modules, connected with a custom network on a ring topology.
DataL1-0
Core 0 Core 1
DataL1-1
Switch
L2-0 L2-1
sw1 sw2sw0 sw3
MM-0 MM-1 MM-2 MM-3
Inst.L1-0
Switch
DataL1-2
Core 2 Core 3
DataL1-3
Inst.L1-1
n0
n2 n3 n4 n5
n1
The Memory HierarchyConfiguration Examples
ISCA 2013, Tel-Aviv 23
s3
s0 s1
s2
n0 n1
n2n3
Example 3
Ring connection between four switches associated with end-nodes with routing tables calculated automatically based on shortest paths. The resulting routing algorithm can contain cycles, potentially leading to routing deadlocks at runtime.
The Memory HierarchyConfiguration Examples
ISCA 2013, Tel-Aviv 24
n3
s1
s2s3
s0 n1
n2
n0
Virtual Channel 0
Virtual Channel 1
Example 4
Ring connection between for switches associated with end nodes, where a routing cycle has been removed by adding an additional virtual channel.
The Memory HierarchyConfiguration Examples
ISCA 2013, Tel-Aviv 25
Pipeline Visualization Tool
DisassemblerEmulator
(or functionalsimulator)
Timingsimulator
(or detailed/architectural)
Visualtool
ISCA 2013, Tel-Aviv 26
Pipeline Visualization ToolPipeline Diagrams
─ Cycle bar on main window for navigation.
─ Panel on main window shows software contexts mapped to hardware cores.
─ Clicking on the Detail button opens a secondary window with a pipeline diagram.
ISCA 2013, Tel-Aviv 27
Pipeline Visualization ToolMemory Hierarchy
─ Panel on main window shows how memory accesses traverse the memory hierarchy.
─ Clicking on a Detail button opens a secondary window with the cache memory representation.
─ Each row is a set, each column is a way.
─ Each cell shows the tag and state (color) of a cache block.
─ Additional columns show the number of sharers and in-flight accesses.
Demo 5
ISCA 2013, Tel-Aviv 28
Part 2
Simulation of aSouthern Islands GPU
ISCA 2013, Tel-Aviv 29
OpenCL on the HostExecution Framework
─ Multi2Sim 4.1 includes a new execution framework for OpenCL, developed in collaboration with University of Toronto.
─ The new framework is a more accurate analogy to a native execution, and is fully AMD-compliant.
─ When working with x86 kernel binaries, the OpenCL runtime can perform both native and simulated execution correctly.
─ When run natively, an OpenCL call to clGetDeviceIDs returns only thex86 device.
─ When run on Multi2Sim, clGetDeviceIDs returns one device per supported architecture: x86, Evergreen, and Southern Islands devices (more to be added).
ISCA 2013, Tel-Aviv 30
OpenCL on the HostExecution Framework
─ The following slides show the modular organization of the OpenCL execution framework, based on 4 software/hardware entities.
─ In each case, we compare native execution withsimulated execution.
Userapplication
API call
Devicedriver
ABI call
Hardware
Internalinterface
Runtimelibrary
Use
r-le
vel c
ode
OS-
leve
l cod
e
ISCA 2013, Tel-Aviv 31
OpenCL on the HostThe OpenCL CPU Host Program
NativeAn x86 OpenCL host program performs an OpenCL API call.
Multi2SimExact same scenario.
ISCA 2013, Tel-Aviv 32
OpenCL on the HostThe OpenCL Runtime Library
NativeAMD's OpenCL runtime library handles the call, and communicates with the driver through system calls ioctl, read, write, etc. These are referred to as ABI calls.
Multi2SimMulti2Sim's OpenCL runtime library, running with guest code, transparently intercepts the call. It communicates with the Multi2Sim driver using system calls with codes not reserved in Linux.
ISCA 2013, Tel-Aviv 33
OpenCL on the HostThe OpenCL Device Driver
NativeThe AMD Catalyst driver (kernel module) handles the ABI call and communicates with the GPU through the PCIe bus.
Multi2SimAn OpenCL driver module (Multi2Sim code) intercepts the ABI call and communicates with the GPU emulator.
ISCA 2013, Tel-Aviv 34
OpenCL on the HostThe GPU Emulator
NativeThe command processor in the GPU handles the messages received from the driver.
Multi2SimThe GPU emulator updates its internal state based on the message received from the driver.
ISCA 2013, Tel-Aviv 35
OpenCL on the HostTransferring Control
• Beginning execution on the GPU
─ The key OpenCL call that effectively triggers GPU execution is clEnqueueNDRangeKernel.
• Order of events
─ The host program performs API call clEnqueueNDRangeKernel.
─ The runtime intercepts the call, and enqueues a new task in an OpenCL command queue object. A user-level thread associated with the command queue eventually processes the command, performing a LaunchKernel ABI call.
─ The driver intercepts the ABI call, reads ND-Range parameters, and launches the GPU emulator.
─ The GPU emulator enters a simulation loop until the ND-Range completes.
ISCA 2013, Tel-Aviv 36
OpenCL on the DeviceExecution Model
─ Work-items execute multiple instances of the same kernel code.
─ Work-groups are sets of work-items that can synchronize and communicate efficiently.
─ The ND-Range is composed by all work-groups, not communicating with each other and executing in any order.
Work-group
Work-group
···
···
Work-group
···
Global Memory
Work-group
Work-item
Work-item
···
···
Work-item
···
Local Memory
Work-item
···__kernel func(){
}
Private Memory
ND-Range Work-Group Work-Item
ISCA 2013, Tel-Aviv 37
OpenCL on the DeviceExecution Model
• Software-hardware mapping
─ When the kernel is launched by the Southern Islands driver, the OpenCLND-Range is mapped to the compute device (Fig. a).
─ The work-groups are mapped to the compute units (Fig. b).
─ The work-items are executed by the SIMD lanes (Fig. c).
─ This is a simplified view of the architecture, explored later in more detail.
SIMD 0Ultra-Threaded Dispatcher
ComputeUnit 0
ComputeUnit 1 ···
Global Memory
Lane
0
15···
SIMD 3
Lane
0
15···
···
Local Memory
Wav
efro
nt S
ched
uler
Register File
FunctionalUnits
(integer &float.)
a) Compute device. c) SIMD lane (stream core).
ComputeUnit 31
b) Compute unit.
ISCA 2013, Tel-Aviv 38
DisassemblerEmulator
(or functionalsimulator)
Timingsimulator
(or detailed/architectural)
Visualtool
The Southern Islands Disassembler
ISCA 2013, Tel-Aviv 39
The Southern Islands DisassemblerSIMD Execution
• Motivation
─ OpenCL follows a Single-Program Multiple-Data (SPMD) programming model, which maps to a Single-Instruction Multiple-Data (SIMD) execution model.
─ A wavefront is a fixed set of work-items forming the SIMD execution unit. One single instruction is fetched, multiple instances of it are executed.
Trade-off: 1 Wavefront = 64 Work-Items
• If the wavefront is large...
─ Amortize ports in the instruction cache and fetch hardware in general.
• If the wavefront is small...
─ Easier to create work-groups with a size equal to an exact multiple of the wavefront size, reducing waste in the wavefront tail..
─ Reduce the effects of thread divergence in conditional execution or loops.
ISCA 2013, Tel-Aviv 40
The Southern Islands DisassemblerScalar Instructions
• Motivation
─ Sometimes work-items within a wavefront do not only execute the same instruction, but also do it on the same data.
─ The compiler can detect some of these cases and emit scalar instructions, issued to a special execution unit of much lower cost.
• Scalar opportunities
─ Loading a base address of a buffer.
─ Loading values from constant memory.
─ Loop initialization, comparison, and increments.
ISCA 2013, Tel-Aviv 41
The Southern Islands DisassemblerVector Addition Kernel
s_buffer_load_dword s0, s[4:7], 0x04 s_buffer_load_dword s1, s[4:7], 0x18 s_buffer_load_dword s4, s[8:11], 0x00 s_buffer_load_dword s5, s[8:11], 0x04 s_buffer_load_dword s6, s[8:11], 0x08 s_load_dwordx4 s[8:11], s[2:3], 0x58 s_load_dwordx4 s[16:19], s[2:3], 0x60 s_load_dwordx4 s[20:23], s[2:3], 0x50 s_waitcnt lgkmcnt(0) s_min_u32 s0, s0, 0x0000ffff v_mov_b32 v1, s0 v_mul_i32_i24 v1, s12, v1 v_add_i32 v0, vcc, v0, v1 v_add_i32 v0, vcc, s1, v0 v_lshlrev_b32 v0, 2, v0 v_add_i32 v1, vcc, s4, v0 v_add_i32 v2, vcc, s5, v0 v_add_i32 v0, vcc, s6, v0 tbuffer_load_format_x v1, v1, s[8:11], 0 tbuffer_load_format_x v2, v2, s[16:19], 0 s_waitcnt vmcnt(0) v_add_i32 v1, vcc, v1, v2 tbuffer_store_format_x v1, v0, s[20:23], 0 s_endpgm
Scal
ar in
stru
ctio
ns
The loads
The additionThe store
Vect
or in
stru
ctio
ns Vector registers
Scalar registers__kernel void vector_add( __read_only __global int *src1, __read_only __global int *src2, __write_only __global int *dst){ int id = get_global_id(0); dst[id] = src1[id] + src2[id];}
• Source code • Assembly code
ISCA 2013, Tel-Aviv 42
The Southern Islands DisassemblerConditional Statements
__kernel void if_kernel(__global int *v)
{ uint id = get_global_id(0); if (id < 5) v[id] = 10;}
• Source code • Assembly code
s_buffer_load_dword s0, s[8:11], 0x04 s_buffer_load_dword s1, s[8:11], 0x18 s_waitcnt lgkmcnt(0) s_min_u32 s0, s0, 0x0000ffff v_mov_b32 v1, s0 v_mul_i32_i24 v1, s16, v1 v_add_i32 v0, vcc, v0, v1 v_add_i32 v0, vcc, s1, v0 s_buffer_load_dword s0, s[12:15], 0x00 v_cmp_lt_u32 s[2:3], v0, 5 s_and_saveexec_b64 s[2:3], s[2:3] v_lshlrev_b32 v0, 2, v0 s_waitcnt lgkmcnt(0) v_add_i32 v0, vcc, s0, v0 v_mov_b32 v1, 10 tbuffer_store_format_x v1, v0, s[4:7], 0 s_mov_b64 exec, s[2:3] s_endpgm
The comparison.Save active mask.
Store value 10.
Restore active mask.
ISCA 2013, Tel-Aviv 43
The Southern Islands Emulator
DisassemblerEmulator
(or functionalsimulator)
Timingsimulator
(or detailed/architectural)
Visualtool
ISCA 2013, Tel-Aviv 44
The Southern Islands EmulatorProgram Loading
Userapplication
API call
Devicedriver
ABI call
Hardware
Internalinterface
Runtimelibrary
• Responsible entity
─ The device driver is the responsible module for setting up an initial state for the hardware, leaving it ready to run the first ISA instruction.
─ Natively, it writes on hardware registers and global memory locations. On Multi2Sim, it calls initialization functions of the emulator.
• Setup
─ Instruction memories in compute units, each with one copy of the ISA section of the kernel binary.
─ Initial global memory image, copying global buffers from CPU to GPU memory.
─ Kernel arguments.
─ ND-Range topology, including number of dimensions and sizes.
ISCA 2013, Tel-Aviv 45
Split ND-Rangeinto work-groups
Work-grouppool
Anywork-groups
left?
Grab work-group andsplit in wavefronts
Wavefrontpool
Anywavefront
left?
For each running wavefront not stalled in a barrier:
● Read instruction @PC● Emulate (update mem. + regs.)● Advance PC
Yes
No
Yes
NoEnd
The Southern Islands EmulatorEmulation Loop
Demo 6
• Work-group execution
─ Work-groups can execute in any order. This order is irrelevant for emulation purposes.
─ The chosen policy is executing one work-group at a time, in increasing order of ID for each dimension.
• Wavefront execution
─ Wavefronts within a work-group can also execute in any order, as long as synchronizations are considered.
─ The chosen policy is executing one wavefront at a time until it hits a barrier, if any.
ISCA 2013, Tel-Aviv 46
Soutern Islands Timing Simulation
DisassemblerEmulator
(or functionalsimulator)
Timingsimulator
(or detailed/architectural)
Visualtool
ISCA 2013, Tel-Aviv 47
Southern Islands Timing SimulationThe GPU Architecture
Command Processor
Ultra-Threaded Dispatcher
ComputeUnit 0
ComputeUnit 1
ComputeUnit 31···
L1Cache
L1Cache
L1Cache···
Crossbar
Main Memory Hierarchy(L2 caches, memory controllers,
video memory)
─ A command processor receives and processes commands from the host.
─ When the ND-Range is created, an ultra-threaded dispatcher (scheduler) assignswork-groups into compute units while new available slots occur.
─ The following slides present the architecture on each of the compute units.
ISCA 2013, Tel-Aviv 48
Southern Islands Timing SimulationThe Compute Unit
─ The instruction memory of each compute unit contains the OpenCL kernel.
─ A front-end fetches instructions and sends them to the appropriate execution unit.
─ There is one scalar unit, vector-memory unit, branch unit, LDS (local data store) unit.
─ There are multiple instances of SIMD units.
ISCA 2013, Tel-Aviv 49
Southern Islands Timing SimulationThe Front-End
─ Work-groups are split into wavefronts and allocated to wavefront pools.
─ Fetch and issue stages operate in a round-robin fashion.
─ There is one SIMD unit associated to each wavefront pool.
Wavefront Pool ···
···
···
···
Wavefront Pool
Wavefront Pool
Wavefront Pool
···
···
···
···
···
Fetch buffers, oneper wavefront pool
SIMD issue buffer,matchingwavefront pool
Scalar unitissue buffer
Branch unitissue buffer
Vector memoryunit issue buffer
LDS unitissue buffer
Fetch
Issue
ISCA 2013, Tel-Aviv 50
─ Runs arithmetic-logic vector instructions.
─ There are 4 SIMD units, each one associated with one of the 4 wavefront pools.
─ The SIMD unit pipeline is modeled with 5 stages: decode, read, execute, write, and complete.
─ In the execute stage, a wavefront (64 work-items max.) is split into 4 subwavefronts (16 work-items each). Subwavefronts are pipelined over the 16 stream cores in 4 consecutive cycles.
─ The vector register file is accessed in the read and write stages to consume input and produce output operands, respectively.
Southern Islands Timing SimulationThe SIMD Unit
ISCA 2013, Tel-Aviv 51
Southern Islands Timing SimulationThe SIMD Unit
Execute
Work-item 0Work-item 16Work-item 32Work-item 48
PipelinedFunctionalunits
SIMD Lane 0
SIMD Lane 1Work-items 1, 17, 33, 49
...
SIMD Lane 15Work-items 15, 31, 47, 63
···Read
···
Issuebuffer
Readbuffer
Write···
Executebuffer
···
Complete
Write
buffer
···
Decode
buffer
Decode
From
com
pute
unit
fro
nt-e
nd
Vector/scalarregister file
Vectorregister file
ISCA 2013, Tel-Aviv 52
Southern Islands Timing SimulationThe Scalar Unit
─ Runs both arithmetic-logic and memory scalar instructions.─ Modeled with 5 stages – decode, read, execute/memory, write, complete.
···Read
···
Issuebuffer
Readbuffer
···
Decode
buffer
Decode
From
com
pute
unit
fro
nt-e
nd
Scalarregister file
Execute
Memory
···
Executebuffer
Write
Complete
Write
buffer
Vectorregister file
···
ISCA 2013, Tel-Aviv 53
Southern Islands Timing SimulationThe Vector-Memory Unit
─ Runs vector memory instructions.─ Modeled with 5 stages – decode, read, memory, write, complete.─ Accesses to the global memory hierarchy happen mainly in this unit.
Read
···
Issuebuffer
Decode
buffer
Decode
From
com
pute
unit
fro
nt-e
nd
Vectorregister file
···
··· Memory
Readbuffer
···
Memorybuffer
Write
Complete
Write
buffer···
Vectorregister file
Globalmemory
ISCA 2013, Tel-Aviv 54
Southern Islands Timing SimulationThe Branch Unit
─ Runs branch instructions. These instructions decide whether to make an entire wavefront jump to a target address depending on the scalar condition code.
─ Modeled with 5 stages – decode, read, execute/memory, write, complete.
Read
···
Issuebuffer
Decode
buffer
Decode
From
com
pute
unit
fro
nt-e
nd
Scalarregister file
(condition codes)
···
··· Execute
Readbuffer
···
Executebuffer
Write
Complete
Write
buffer···
Scalar reg. file(programcounter)
ISCA 2013, Tel-Aviv 55
Southern Islands Timing SimulationThe LDS (Local Data Store) Unit
─ Runs local memory accesses instructions.─ Modeled with 5 stages – decode, read, execute/memory, write, complete.─ The memory stage accesses the compute unit local memory for read/write.
Read
···
Issuebuffer
Decode
buffer
Decode
From
com
pute
unit
fro
nt-e
nd
Vectorregister file
···
··· Memory
Readbuffer
···
Memorybuffer
Write
Complete
Write
buffer···
Vectorregister file
Localmemory
ISCA 2013, Tel-Aviv 56
Southern Islands Timing SimulationBenchmark Support
• Other applications
─ Other applications from the Rodinia and Parboil benchmark suite have been added support for by other users. Packages are under development.
• AMD OpenCL SDK
─ Host programs are statically linked x86 binaries.
─ The SDK is available for device kernels using three different architectures: Evergreen, Southern Islands, and x86.
─ Packages include data files and execution commands.
ISCA 2013, Tel-Aviv 57
Global Memory HierarchyOrganization
─ Fully configurable memory hierarchy, with default values based on theAMD Radeon HD 7970 Southern Islands GPU.
─ One 16KB data L1 per compute unit.
─ One scalar L1 cache shared by every 4 compute units.
─ Six L2 banks with a total size of 128KB, each connected to a DRAM module.
CU 0
L1
CU 1
L1
CU 2
L1
CU 3
L1Scalarcache
CU 28
L1
CU 29
L1
CU 30
L1
CU 31
L1Scalarcache. . .
. . .L2Bank 0
L2Bank 1
L2Bank 1
...Interconnect
ISCA 2013, Tel-Aviv 58
Global Memory HierarchyPolicies
─ Inclusive, write-back, non-coherent caches.
─ Non-coherence is implemented as a 3-state “coherence” protocol: NSI
─ Blocks in N state are merged on write-back using write bit masks.
Non-coherent – Block can be accessed for read/write access, while shared with other caches.
Shared – Block is accessible for read access, possible shared with other caches.
Invalid – Block does not contain valid data.
1111 0000 0000 1111
L2 cache
Writeback
Writeback
Cache blockmerged onwriteback
L1 cache L1 cache
ISCA 2013, Tel-Aviv 59
Southern Islands Pipeline Visualization
DisassemblerEmulator
(or functionalsimulator)
Timingsimulator
(or detailed/architectural)
Visualtool
ISCA 2013, Tel-Aviv 60
Demo 7
Southern Islands Pipeline VisualizationMain Window and Timing Diagram
─ Main window provides cycle-by-cycle navigation throughout simulation.
─ A dedicated Southern Islands panel contains one widget per compute unit, showing allocated work-groups.
─ The memory hierarchy panel shows caches connected to Southern Islands compute units, and special-purpose scalar caches.
ISCA 2013, Tel-Aviv 61
Case Study: ND-Range VirtualizationMotivation
• Could be also called...
─ Cooperative work-group execution.
─ Next-level system heterogeneity.
• The idea
─ The program writes one single OpenCL kernel.
─ The OpenCL runtime creates one kernel binary targeting each architecture present in the system.
─ The ND-Range is dynamically and automatically partitioned, and work-groups are spread across CPU and GPU cores.
• The result
─ Data-parallel programming model complexity resides in the design of the OpenCL kernel, but not in cross-device scheduling.
─ System resources (cores) better utilized and transparently managed.
ISCA 2013, Tel-Aviv 62
Case Study: ND-Range VirtualizationCurrent Heterogeneous Execution
Hardware
ND-Range
WG-0 WG-1 WG-2 WG-3 WG-N
S.I. S.I. S.I. S.I.x86x86
. . .
Hardware
ND-Range
WG-0 WG-1 WG-2 WG-3 WG-N
S.I. S.I. S.I. S.I.x86x86
. . .
S.I. S.I. S.I. S.I.x86x86
Tim
e
Hostprogram
Idle Device kernel
• Work-group mapping
─ Complete ND-Range sent to one device, selected by the user with a call to clGetDeviceIDs.
• Attained concurrency
─ Host program runs on one x86 core, device kernel runs on Southern Islands compute units.
─ Idle execution regions on x86 cores while Southern Islands compute units run the ND-Range.
ISCA 2013, Tel-Aviv 63
Case Study: ND-Range VirtualizationHeterogeneous Execution with ND-Range Virtualization
S.I. S.I. S.I. S.I.x86x86
Tim
e
Hostprogram+ kernel
KernelKernel
Hardware
ND-Range
S.I. S.I. S.I. S.I.x86x86
WG-0 WG-1 WG-2 WG-3 WG-N. . .
• Work-group mapping
─ Portions of ND-Range executed by CPU/GPU cores with different ISAs.
─ Work-groups mapped to CPU cores or GPU compute units as they become available.
• Attained concurrency
─ x86 cores run both the host program and a portion of the ND-Range.
─ Idle regions are removed during the execution of the ND-Range..
ISCA 2013, Tel-Aviv 64
Case Study: ND-Range VirtualizationThe Hardware
─ Protocols MOESI and NSI merged into a single 6-state protocol: NMOESI.
─ CPU and GPU cores with any ISA can be connected to different entry points of the memory hierarchy.
─ Processing nodes interact with the memory hierarchy with three types of accesses: load, store, and n-store.
─ GPUs and CPUs running OpenCL kernels issue n-store write accesses. Rest issue regular store accesses.
. . .
L2Bank 0
...
ARM x86 Evg. S.I. . . .Fermi
L1 L1 L1 L1
Interconnect
. . .L2Bank 1
NMOESI Interface
ISCA 2013, Tel-Aviv 65
Case Study: ND-Range VirtualizationThe Software
The OpenCL host program is simplified. Only one command queue needed to exploit system heterogeneity.
Userapplication
API call
Devicedriver
ABI call
Hardware
Internalinterface
Runtimelibrary
The runtime provides an additional virtual fused device, returned as a result of a call to clGetDeviceIDs.
The driver extends its interface to allow each physical device to run a portion of the ND-Range, at the work-group granularity.
ISCA 2013, Tel-Aviv 66
ConcludingRemarks
ISCA 2013, Tel-Aviv 67
The Multi2Sim CommunityAdditional Material
• The Multi2Sim Guide─ Complete documentation of the simulator's user interface, simulation
models, and additional tools.
• Multi2Sim forums and mailing list─ New version releases and other important information is posted on the
Multi2Sim mailing list (no spam, 1 email per month).─ Users share question and knowledge on the website forum.
• M2S-Cluster─ Automatic verification framework for Multi2Sim.─ Based on a cluster of computers running condor.
ISCA 2013, Tel-Aviv 68
The Multi2Sim CommunityMulti2C – The Multi2Sim Compiler
─ LLVM-based compiler for OpenCL and CUDA kernels.
─ Future release Multi2Sim 4.2 will include a working version.
─ Diagrams show progress as per SVN Rev. 1838
• Orange = under development• Yellow = arriving soon• Green = supported
vec-add.clOpenCL Cto LLVM
front-end
CUDAto LLVM
front-endvec-add.cu
LLVM toSouthernIslands
back-end
vec-add.llvm
vec-add.s
LLVM toFermi
back-end
LLVM toKepler
back-end
vec-add.s
vec-add.s
SouthernIslands
assembler
Fermiassembler
Keplerassembler
vec-add.bin
vec-add.cubin
vec-add.cubin
ISCA 2013, Tel-Aviv 69
The Multi2Sim CommunityAcademic Efforts at Northeastern
• The “GPU Programming and Architecture” course─ We started an unofficial seminar that students can voluntarily attend. The
syllabus covers OpenCL programming, GPU architecture, and state-of-the-art research topics on GPUs.
─ Average attendance of ~25 students per semester.
• Undergraduate directed studies─ Official alternative equivalent to a 4-credit course that an undergraduate
student can optionally enroll in, collaborating in Multi2Sim development.
• Graduate-level development─ Lots of research projects at the graduate level depend are based on
Multi2Sim, and selectively included in the development trunk for public access.
─ Simulation of OpenGL pipelines, support for new CPU/GPU architectures, among others.
ISCA 2013, Tel-Aviv 70
The Multi2Sim CommunityCollaborating Research Groups
• Universidad Politécnica de Valencia─ Pedro López, Salvador Petit, Julio Sahuquillo, José Duato.
• Northeastern University─ Chris Barton, Shu Chen, Zhongliang Chen, Tahir Diop, Xiang Gong, David
Kaeli, Nicholas Materise, Perhaad Mistry, Dana Schaa, Rafael Ubal, Mark Wilkening, Ang Shen, Tushar Swamy, Amir Ziabari.
• University of Mississippi─ Byunghyun Jang
• NVIDIA─ Norm Rubin
• University of Toronto─ Jason Anderson, Natalie Enright, Steven Gurfinkel, Tahir Diop.
• University of Texas─ Rustam Miftakhutdinov
ISCA 2013, Tel-Aviv 71
The Multi2Sim CommunityMulti2Sim Academic Publications
• Conference papers─ Multi2Sim: A Simulation Framework to Evaluate Multicore-Multithreaded
Processors, SBAC-PAD, 2007.─ The Multi2Sim Simulation Framework: A CPU-GPU Model for Heterogeneous
Computing, PACT, 2012.
• Tutorials─ The Multi2Sim Simulation Framework: A CPU-GPU Model for Heterogeneous
Computing, PACT, 2011.─ Programming and Simulating Fused Devices — OpenCL and Multi2Sim, ICPE,
2012.─ Multi-Architecture ISA-Level Simulation of OpenCL, IWOCL, 2013.─ Simulation of OpenCL and APUs on Multi2Sim, ISCA, 2013.
ISCA 2013, Tel-Aviv 72
The Multi2Sim CommunityPublished Academic Works Using Multi2Sim
• Recent─ R. Miftakhutdinov, E. Ebrahimi, Y. Patt, Predicting Performance Impact of
DVFS for Realistic Memory Systems, MICRO, 2012.─ D. Lustig, M. Martonosi, Reducing GPU Offload Latency via Fine-Grained
CPU-GPU Synchronization, HPCA, 2013.
• Other─ H. Calborean, R. Jahr, T. Ungerer, L. Vintan, A Comparison of Multi-objective
Algorithms for the Automatic Design Space Exploration of a Superscalar System, Advances in Intelligent Systems and Computing, vol. 187.
─ X. Li, C. Wang, X. Zhou, Z. Zhu, Cache Promotion Policy Using Re-reference Interval Prediction, CLUSTER, 2012.
… and 62 more citations, as per Google Scholar.
ISCA 2013, Tel-Aviv 73
The Multi2Sim CommunitySponsors