AMD Developer - GPU Developer Tools

GPUPerfAPI 2.9 is now available for download

plohrmann — 2012-01-20T15:19:30 -05.00

AMD GPU Developer Tools is proud to announce the release of version 2.9 of the GPU Performance API.

The new version can be downloaded from: http://developer.amd.com/tools/GPUPerfAPI/Pages/default.aspx

Version 2.9 (1/18/12)

Add support for AMD Radeon 7000 Series hardware.
Add support for more FirePro, Mobility, and APU devices.
Improved memory consumption and performance during profiling.
Reduce memory footprint of the GPUPerfAPI DLLs.
Correct counters are now exposed on systems with dual GPUs of different hardware generations.
DirectX 10:
- Fixed shader related counters on AMD Radeon HD 2000-5000 Series hardware.
- Fixed PrimitivesIn counter on AMD Radeon HD 2000 Series hardware.

APP and FPGA DMA

legalto — 2012-01-16T19:52:23 -05.00

Is DMA possible between AMD GPUs and any FPGA boards? I've seen no mention of it being done thus far. If this capability is not available, does anyone know if AMD is looking into this?

AMD APP KernelAnalyzer crashes with float3 args to local mem assignment

anlu — 2012-01-13T13:39:56 -05.00

I came across a bug in the AMD APP Kernel Analyzer 1.11.1172 the other day:

When you have an float3 argument in your kernel parameter list,

and try to assign any component of it into a local memory variable,

the AMD APP Kernel Analyzer crashes, although the runtime compilation

with the AMD Runtime works just fine. I attached the code which causes

the crash.

I would be glad if someone look into this matter.

Regards,

No kernel debugging

npays — 2012-01-11T10:32:08 -05.00

Hi,

first off all, why was my first subject deleted ? I like to know what I do wrong when I do, so I can avoid making mistakes several times.

So, here's my problem : I can't get gDebugger to work properly. I have the latests catalyst pro on a v7900, and it keeps starting with kernel mode deactivated. Is there something I didn't install ?

Second problem : I did try some gDebugger some time ago, where I had access to a lot of gl counters like culled primitives & co. Is that unavailable now because I don't have kernel debugging ?

Regards,

kernel analyzer in virtual system

nou — 2012-01-08T17:17:52 -05.00

i had instaled kernel analyzer in virtualized windows (i am developing on linux) and checked the kernels in there. now i updated to latest 1.11 and it print out error message about missing aticalrt.dll. i also instaled AMD APP SDK but it didn't help. is there a way run kernel analyzer on system without AMD card and catalyst?

"GPA_OpenContext()" is returning "GPA_STATUS_ERROR_HARDWARE_NOT_SUPPORTED"

sandytelxsi02 — 2012-01-05T07:27:11 -05.00

Hello,

I am trying to get GPU utilization using "GPUPerf" library.

I modified the sample application "fgl_glxgears" , such that after opengl context is created I call api "GPA_OpenContext" . But this api always returns "GPA_STATUS_ERROR_HARDWARE_NOT_SUPPORTED".

GPA_Initialize is called before the opengl context is created and this returns success.

Attaching the part of sample application for reference "fgl_glxgears.c".(Plz search for GPU in the sample code)

PS:

OS : Fedora Core 15 64bit

Kernel : 2.6.38

Driver : ATI catalyst driver version 11.11

Platform : AMD A6-3410MX APU

GDebugger 6.1 - ENV Path

ahagan — 2012-01-04T13:30:30 -05.00

In VS2010 I set the environment variables under the project settings Debugging page.

When I start my opengl program with gdebugger via the gedebugger menu ("Start OpenGL Degugging"), it doesn't seem that the variables get merged into the current session.

The old version of gdebugger (5.8) had the ability to set env. variables for a debugging session, is there a solution to this for the 6.1 version of gdebugger ?

Thanks for the help.

Function selection not working in KernelAnalyzer 11.11

Marix — 2011-12-29T06:05:35 -05.00

If multiple kernels are contained within one source file loaded into the Kernel Analyzer the Compiler Statistics window always displays the data from the first kernel in the file, independent of which kernel is selected in the function drop down menu. The Object Code window seems to be updated correctly.

The problem can be reproduced using the attached code. Not that the Fetch column stays at 2 if the vecMulAdd-Function is selected via the dropdown. If the order of the kernels is changed it changes to 3, and stays at that value even if vecAdd is selected via the dropdown.

I am aware that the code needs to be recompiled after the function selection for the statistics to be updated, however it seems only to update the Object Code.

Please not that I am using the workaround described in http://forums.amd.com/devforum/messageview.cfm?catid=347&threadid=157841&enterthread=y to run the Kernel Analyzer on a system without an AMD GPU. If the issue does not exist on a system with an AMD GPU this would also be valuable information to me.

KernelAnalyzer 11.11 on System without AMD GPU

Marix — 2011-12-22T13:48:44 -05.00

I installed the AMD Kernel Analyzer 11.11 and APP SDK 2.6 on my non-AMD-GPU equipped Laptop (actually in VMWare) to analyze some register utilization issues. While previous version (e.g. Kernal Analyzer 11.7) worked fine in this setup the 11.11 version fails with the following error:

--------------------------- AMDAPPKernelAnalyzer.exe - Systemfehler --------------------------- Das Programm kann nicht gestartet werden, da aticalrt.dll auf dem Computer fehlt. Installieren Sie das Programm erneut, um das Problem zu beheben. --------------------------- OK ---------------------------

It seems to be missing aticalrt.dll, which I assume would normally be provided by the GPU driver. However I also though the Kernel Analyzer brought it's own version of that, to be able to emulate different driver version. Am I doing something wrong, or is this kind of setup no longer supported?

APP Profiler support for fusion APU

vns — 2011-12-16T05:56:47 -05.00

Hi,

Is there support for GPU performance counters on AMD Fusion APU? In particularly BeaverCreek (Llano) and Loveland (Zacate).

Insofar I'm seeing confusing results from running sprofile. Where whether counters are to be sampled, would depend on numerous factors, including system configuration, windows or linux OS, and version of the catalyst driver.

ProfilerVersion2.3.811
Catalyst 12.12

APP SDK 2.5

Thank you.

Developers are confused about OpenGL and DirectX for new game development

VJ01 — 2011-12-10T19:52:30 -05.00

I have a dozen to two dozen simple questions.

I asked an experienced graphics developer most simple question on GameDev.net and they were caught with headlights on a deer look!

I told them to solve your errors check Exception handling on OpenGL processing unit or GPU.

GPU Perfstudio 1.2 "Error: GPU PerfServer not running on Local"

Jaebaek — 2011-12-10T01:26:31 -05.00

Hi, I tried to profile DirectX 9 application using GPU Perfstudio 1.2.

The application is normal when executing itself and it is also normal when drag it onto "OpenGL App Launcher".

But when I profile it using GPU Perfstudio 1.2, GPU Prefstudio 1.2 shows me "Error: GPU PerfServer not running on Local" message.

Definitely, "OpenGL App Launcher" works fine with my application as a local server ...

any help?

GPU and Virtual Machine suspicions

VJ01 — 2011-12-05T12:20:24 -05.00

I've had frustrated views of OpenGL ARB (Extensions ????) and Direct 2d and 3d (...)

After what i expect to be newspaper readings, looks like a feverish note taking work-out.

And then I have suspicion that a GPU has a virtual machine (i presumed a Lite version.)

The famous military game is still cool today and runs in XB 360 (directx for linux). (and wikipedia says directx is for microsoft platform). (off imagination: At times, i thought someones BXB 360/720 has galactical powers like thought control.)

GPUPerfstudio Frame Profiler doesn't support 6900M

xoofx — 2011-12-01T09:20:19 -05.00

Hi,

I'm trying to profile a Direct3D11 application with GPUPerfStudio but I get an error that my hardware is not supported (Perf counters were not found).

I have a HD6900M (running Win7 from an iMac) with Catalyst 11.11 installed.

Is this hardware is going to be supported in the future?

Here is the GPUPerfServer log:

App : C:\Code\GPUPerfStudio2.7.829.0\Server-Win32\GPUPerfServer.exe
PID: 3420
Time: 2011-Dec-01 21:51:27
21:51:27.029050: Message:     3420 Server        : Using Logfile: C:/Code/GPUPerfStudio2.7.829.0/Server-Win32/pslog.txt
21:51:27.036050: Message:     3420 Server        : Succesfully enabled DX Peformance Counters in Driver
21:51:27.037051: Message:     3420 Server        : Available Wrappers:
21:51:27.038051: Message:     3420 Server        :    DX10Server.dll   V2.7.829.0   DX10
21:51:27.038051: Message:     3420 Server        :    DX11Server.dll   V2.7.829.0   DX11
21:51:27.039051: Message:     3420 Server        :    DX9Server.dll   V2.7.829.0   DX9
21:51:27.039051: Message:     3420 Server        :    DXGIServer.dll   V2.7.829.0   DXGI
21:51:27.039051: Message:     3420 Server        :    GLServer.dll   V2.7.829.0   OpenGL

21:51:27.041051: Message:     3420 Server        : Starting web server on port 8082
21:51:27.042051: Message:     3420 Server        : About to launch: C:\Code\Temp\TestDirect3D11\TestDirect3D11_32bit.exe
21:51:27.043051: Message:     3420 Server        : Params:
21:51:27.043051: Message:     3420 Server        : Working Directory: C:\Code\Temp\TestDirect3D11\

[...]

21:52:34.439906: Message:     3420 Server        : 78: 127.0.0.1 - /8748/DX11/CommandTree.xml
21:52:36.827042: Message:     3420 Server        : 79: 127.0.0.1 - /8748/DX11/PushLayer=FrameProfiler
21:52:36.845043: Message:     3420 Server        : 80: 127.0.0.1 - /8748/DX11/FP/CounterInfo.xml
21:52:36.891046: Error:       8748               : GPA: Unrecognized device ID.
21:52:36.892046: Error:       8748               : GPA: Unable to create a known device dependent counter.
21:52:36.893046: Error:       8748               : Unable to open counters: Hardware not supported.

GPU shader analyzer question: how cycle count is got if there's data dependent loop

terri_yu — 2011-11-02T05:15:00 -05.00

Hi expert,

I'm using AMD shader analyzer, and it's very helpful. But I don't understand when my shader source contains data dependent loop, how the tool calculates the cycle count (min/max/average).

I don't know how the data is got, therefore I'm not sure how accurate the estimation is.

Could you explain the algorithm to me?

And for more accurate estimation, is it possible to allow us to input the data to decide the loop number for example?

Thanks,

Terri

Kernel Analyzer 1.9 bug

Meteorhead — 2011-11-01T09:37:28 -05.00

Hi!

There is a bug with Kernel Analyzer 1.9 which causes permanent lock-up of the tool:

I have tried to compile a kernel with the following two lines attached as code.

Since then, it fails to compile any kernel loaded from any place that has a for cycle, even after software restart. I have created a define that is usually a kernel parameter, but I wanted to investigate GPR usage and expected throughput of kernels and stuff, so I defined this variable so compiler has a chance of loop unroll.

However, it seems to have regarded the "define for offline" in a very bad manner, even though it was preceded by "//". Now it remembers this misbehavior and throws an error on every for cycle (even ones I load from other applications) saying:

Line 57: error: function
"offline" declared implicitly
for (uint i = *call; i < *call + 4u ; ++i)

However, the word offline isn't even present inside the kernel! So clearly it remembers something from earlier. As I have stated, this problem persists through application restart also. This is clearly a bug, so please do try to correct it.

Edit: What's more worse, is that it persists through reinstall of the application. Could you give me some hint how to make it go away?

Can only see half the device memory as global memory using OpenCL

troels_blum — 2011-10-28T07:50:30 -05.00

Hi,

System: AMD Phenom(tm) II X6, 2 x AMD Radeon HD 6990, Ubuntu 11.10, Catalyst 11.9, AMD-APP-SDK-v2.5

The two GPU devices show up as four OpenCL devices, which is OK, but they only have 1GB of global memory each. Each device has 2x2GB. How do I get access to it all?

BR Troels

Can only see GPU devices after loggin in through X (Ubuntu)

troels_blum — 2011-10-28T07:46:51 -05.00

Hi,

System: AMD Phenom(tm) II X6, 2 x AMD Radeon HD 6990, Ubuntu 11.10, Catalyst 11.9, AMD-APP-SDK-v2.5

My problem is that the GPU's are only accessible after I log in through gdm on the machine. After setting it up I (and others) will only have access to the machine through ssh. How do I get around this?

The Machine will only be used for OpenCL programming, and running these programs.

PerfStudio 2.7 shader editor error

MadrMan — 2011-10-26T04:50:46 -05.00

Hello,

I can't seem to get the shader editor to work for me, every time i load up my program and try to use it i get: "Error: D3DX11CompileFromFile failed." with no additional info.

I've tried to reproduce it on dx sdk samples but it seemed to work correctly on all the sdk samples that use fx and the "Enable shader editing" button seemed to be grayed out on the samples that didn't use fx.

Another minor problem is that perfstudio doesn't seem to like the line-ends in the sdk shaders, the whole shader appears on a single line.

I've attached the code to one of the shaders that doesn't work for me.

I'm using vista x64 sp2 with a 4850HD and the latest catalyst drivers and latest dx sdk.

Is there any way I could get additional info on what the error is?

KernelAnalyzer won't compile CPU code

aporter — 2011-10-25T12:07:52 -05.00

The following kernel won't compile on the KernelAnalyzer when x86 Assembly is chosen.

OpenCL Compile Error: clBuildProgram failed (CL_BUILD_PROGRAM_FAILURE).

Internal Error: as failed

It compiles fine if Radeon HD 5870 (Cypress) Assembly is selected.

I've seen references to having the wrong runtime installed (driver vs developer) but am unclear if it I indeed have the correct one installed. Note also that the same problem occurs if I compile a kernel in a openCL program. The GPU targeted compiles complete ok but the CPU ones return with internal error.

Thanks in advance.

See simple kernel and clinfo trace below:

__kernel void XXXX(
__global float3 *output,
const __global uchar3 *input
) {
uint w = get_global_size(0);
uint x = get_global_id(0);
uint y = get_global_id(1);
uint i = y*w+x;
output.x = (float)input.x;
output.y = (float)input.y;
output.z = (float)input.z;
}

A trace from clinfo is below:

Number of platforms:     1
Platform Profile:     FULL_PROFILE
Platform Version:     OpenCL 1.1 AMD-APP-SDK-v2.5 (732.1)
Platform Name:     AMD Accelerated Parallel Processing
Platform Vendor:     Advanced Micro Devices, Inc.
Platform Extensions:     cl_khr_icd cl_amd_event_callback cl_amd_offline_devices cl_khr_d3d10_sharing

Platform Name:     AMD Accelerated Parallel Processing
Number of devices:     3
Device Type:      CL_DEVICE_TYPE_GPU
Device ID:      4098
Max compute units:     20
Max work items dimensions:    3
    Max work items[0]:     256
    Max work items[1]:     256
    Max work items[2]:     256
Max work group size:     256
Preferred vector width char:    16
Preferred vector width short:    8
Preferred vector width int:    4
Preferred vector width long:    2
Preferred vector width float:    4
Preferred vector width double:   2
Native vector width char:    16
Native vector width short:    8
Native vector width int:    4
Native vector width long:    2
Native vector width float:    4
Native vector width double:    2
Max clock frequency:     850Mhz
Address bits:      32
Max memory allocation:    536870912
Image support:     Yes
Max number of images read arguments:   128
Max number of images write arguments:   8
Max image 2D width:     8192
Max image 2D height:     8192
Max image 3D width:     2048
Max image 3D height:     2048
Max image 3D depth:     2048
Max samplers within kernel:    16
Max size of kernel argument:    1024
Alignment (bits) of base address:   32768
Minimum alignment (bytes) for any datatype: 128
Single precision floating point capability
    Denorms:      No
    Quiet NaNs:      Yes
    Round to nearest even:    Yes
    Round to zero:     Yes
    Round to +ve and infinity:    Yes
    IEEE754-2008 fused multiply-add:   Yes
Cache type:      None
Cache line size:     0
Cache size:      0
Global memory size:     1073741824
Constant buffer size:     65536
Max number of constant args:    8
Local memory type:     Scratchpad
Local memory size:     32768
Kernel Preferred work group size multiple: 64
Error correction support:    0
Unified memory for Host and Device:   0
Profiling timer resolution:    1
Device endianess:     Little
Available:      Yes
Compiler available:     Yes
Execution capabilities:
    Execute OpenCL kernels:    Yes
    Execute native function:    No
Queue properties:
    Out-of-Order:     No
    Profiling :      Yes
Platform ID:      000007FEE62F4348
Name:       Cypress
Vendor:      Advanced Micro Devices, Inc.
Device OpenCL C version:    OpenCL C 1.1
Driver version:     CAL 1.4.1523 (VM)
Profile:      FULL_PROFILE
Version:      OpenCL 1.1 AMD-APP-SDK-v2.5 (732.1)
Extensions:      cl_khr_fp64 cl_amd_fp64 cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_3d_image_writes cl_khr_byte_addressable_store cl_khr_gl_sharing cl_ext_atomic_counters_32 cl_amd_device_attribute_query cl_amd_vec3 cl_amd_printf cl_amd_media_ops cl_amd_popcnt cl_khr_d3d10_sharing

Device Type:      CL_DEVICE_TYPE_GPU
Device ID:      4098
Max compute units:     20
Max work items dimensions:    3
    Max work items[0]:     256
    Max work items[1]:     256
    Max work items[2]:     256
Max work group size:     256
Preferred vector width char:    16
Preferred vector width short:    8
Preferred vector width int:    4
Preferred vector width long:    2
Preferred vector width float:    4
Preferred vector width double:   2
Native vector width char:    16
Native vector width short:    8
Native vector width int:    4
Native vector width long:    2
Native vector width float:    4
Native vector width double:    2
Max clock frequency:     850Mhz
Address bits:      32
Max memory allocation:    536870912
Image support:     Yes
Max number of images read arguments:   128
Max number of images write arguments:   8
Max image 2D width:     8192
Max image 2D height:     8192
Max image 3D width:     2048
Max image 3D height:     2048
Max image 3D depth:     2048
Max samplers within kernel:    16
Max size of kernel argument:    1024
Alignment (bits) of base address:   32768
Minimum alignment (bytes) for any datatype: 128
Single precision floating point capability
    Denorms:      No
    Quiet NaNs:      Yes
    Round to nearest even:    Yes
    Round to zero:     Yes
    Round to +ve and infinity:    Yes
    IEEE754-2008 fused multiply-add:   Yes
Cache type:      None
Cache line size:     0
Cache size:      0
Global memory size:     1073741824
Constant buffer size:     65536
Max number of constant args:    8
Local memory type:     Scratchpad
Local memory size:     32768
Kernel Preferred work group size multiple: 64
Error correction support:    0
Unified memory for Host and Device:   0
Profiling timer resolution:    1
Device endianess:     Little
Available:      Yes
Compiler available:     Yes
Execution capabilities:
    Execute OpenCL kernels:    Yes
    Execute native function:    No
Queue properties:
    Out-of-Order:     No
    Profiling :      Yes
Platform ID:      000007FEE62F4348
Name:       Cypress
Vendor:      Advanced Micro Devices, Inc.
Device OpenCL C version:    OpenCL C 1.1
Driver version:     CAL 1.4.1523 (VM)
Profile:      FULL_PROFILE
Version:      OpenCL 1.1 AMD-APP-SDK-v2.5 (732.1)
Extensions:      cl_khr_fp64 cl_amd_fp64 cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_3d_image_writes cl_khr_byte_addressable_store cl_khr_gl_sharing cl_ext_atomic_counters_32 cl_amd_device_attribute_query cl_amd_vec3 cl_amd_printf cl_amd_media_ops cl_amd_popcnt cl_khr_d3d10_sharing

Device Type:      CL_DEVICE_TYPE_CPU
Device ID:      4098
Max compute units:     8
Max work items dimensions:    3
    Max work items[0]:     1024
    Max work items[1]:     1024
    Max work items[2]:     1024
Max work group size:     1024
Preferred vector width char:    16
Preferred vector width short:    8
Preferred vector width int:    4
Preferred vector width long:    2
Preferred vector width float:    4
Preferred vector width double:   0
Native vector width char:    16
Native vector width short:    8
Native vector width int:    4
Native vector width long:    2
Native vector width float:    4
Native vector width double:    0
Max clock frequency:     2672Mhz
Address bits:      64
Max memory allocation:    3218887680
Image support:     Yes
Max number of images read arguments:   128
Max number of images write arguments:   8
Max image 2D width:     8192
Max image 2D height:     8192
Max image 3D width:     2048
Max image 3D height:     2048
Max image 3D depth:     2048
Max samplers within kernel:    16
Max size of kernel argument:    4096
Alignment (bits) of base address:   1024
Minimum alignment (bytes) for any datatype: 128
Single precision floating point capability
    Denorms:      Yes
    Quiet NaNs:      Yes
    Round to nearest even:    Yes
    Round to zero:     Yes
    Round to +ve and infinity:    Yes
    IEEE754-2008 fused multiply-add:   No
Cache type:      Read/Write
Cache line size:     64
Cache size:      32768
Global memory size:     12875550720
Constant buffer size:     65536
Max number of constant args:    8
Local memory type:     Global
Local memory size:     32768
Kernel Preferred work group size multiple: 1
Error correction support:    0
Unified memory for Host and Device:   1
Profiling timer resolution:    383
Device endianess:     Little
Available:      Yes
Compiler available:     Yes
Execution capabilities:
    Execute OpenCL kernels:    Yes
    Execute native function:    Yes
Queue properties:
    Out-of-Order:     No
    Profiling :      Yes
Platform ID:      000007FEE62F4348
Name:       Intel(R) Core(TM) i7 CPU         920 @ 2.67GHz
Vendor:      GenuineIntel
Device OpenCL C version:    OpenCL C 1.1
Driver version:     2.0
Profile:      FULL_PROFILE
Version:      OpenCL 1.1 AMD-APP-SDK-v2.5 (732.1)
Extensions:      cl_khr_fp64 cl_amd_fp64 cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_int64_base_atomics cl_khr_int64_extended_atomics cl_khr_byte_addressable_store cl_khr_gl_sharing cl_ext_device_fission cl_amd_device_attribute_query cl_amd_vec3 cl_amd_media_ops cl_amd_popcnt cl_amd_printf cl_khr_d3d10_sharing

Same values of WriteInsts, ALUinsts, Fetchinsts for GPUPerfAPI-2.7

ukidaveyash — 2011-10-24T22:43:13 -05.00

I have been using GPUPerfAPI-2.7 lately and have profiled my application on Radeon 5870. The numbers for the WriteInsts , FetchInsts, and ALUinsts are the same.

To verify the doubt I have tried it accross the samples of the AMD SDK. They values reported there are the same too.

Is something more to be done than what is in the User guide.

IDE asm and 'C'

bebiot — 2011-10-22T11:19:43 -05.00

Hello,
what software to write code in assembler (ATI), I would like to interface with the 'C' in my code.
I'm pretty lost.

GPU PerfStudio minimum driver version requirement

TiagoVCosta — 2011-10-13T06:38:33 -05.00

Which GPU driver version do I have to use in order to GPU PerfStudio 2.7 work? Because my notebook manufacturer doesnt update drivers very often...

--Tiago Costa

ATI Catalyst driver since 11.7 version can't work with hybrid GPU?

DimaNovikov — 2011-10-11T14:26:30 -05.00


I've got Radeon 5770 and 6570 in my desktop and ati-stream-sdk-v2.3-lnx32 on my OpenSUSE 11.3.

When I install ATI Catalyst driver with a version 11.7 or higher, OpenCL doesn't see my Radeon 5770 at all. I mean CLinfo sample tells me that there is only 2 ATI Stream devices in my comp (including AMD CPU).

Most interesting that driver 11.6 can see with all the 3 devices (2 GPU + 1 CPU) and works fine with all this hardware.

What happened to drivers since 11.7 version? Why hybrid GPU are not supported any more?

Problem using BOINC Manager and 11.8 or 11.9 Catalyst

fpabernard — 2011-10-11T07:58:58 -05.00

Hello,

I noticed that BOINC Manager (for contributing to World Community Grid) does not work any more since I installed 11.8 Catalyst Driver. I used previously 11.6 and many older versions, and it worked well.

Same problem with last 11.9.

I uninstalled 11.9 and went back to 11.6 : fine !

Same problem with either HD4650 or HD6850.

Best regards

F.Bernard

GPU PerfStudio 2.7 is now publicly available

Lonesled — 2011-10-06T13:50:02 -05.00

GPU Developer Tools is pleased to announce that GPU PerfStudio 2.7 is publicly available from AMD Developer Central at:

http://developer.amd.com/gpu/perfstudio/Pages/default.aspx

GPU PerfStudio 2.7 Improvements:

·         OpenGL 4.0 is now supported

·         Stage Layouts

o    Pipeline Stage view now shows which stages are active at a draw call

o    New Image Resource Thumbnail viewer - allows the user to view image resources from different stages of the pipeline in one view

o    DX11/DX10 the stencil buffer has been removed from the OM layout. The Depth and Stencil images are both in the depth image in different channels

·         DX11/DX10

o    Frame Capture now supports the creation of vertex buffers inside a frame

o    DX11 Buffer Visualization support for shader inputs

o    DX11 CS UAV Visualization support for shader inputs

o    DX11 Constant buffers now display non-float data correctly

·         Profiler

o    New Profile delta feature allows the user to see the differences between two profiles

·         Shader Debugger

o    Vertex Shader debugging is now supported

Catalyst 11.8 and 11.9 for SKA

corry — 2011-10-05T15:01:15 -05.00

When can we expect this? SDK 2.6 sounds unreasonable since when looking in C:\program files (x86)\Common\AMD\GPU ShaderAnalyzer I see dlls named GPUShaderAnalyzer_CAL_11_7.dll and GPUShaderAnalyzer_CAL_11_6.dll...and so on. Seems you designed it to take plugins of sorts...so why can you not release a GPUShaderAnalyzer_CAL_11_8.dll and GPUShaderAnalyzer_CAL_11_9.dll?

On a related but seperate note, I cannot seem to get the build in cal compiler to use the bfi instruction. See posings in thread http://forums.amd.com/devforum/messageview.cfm?catid=390&threadid=143488&enterthread=y for what I have tried...This makes no sense if Micah is correct...

Minor bug

corry — 2011-09-30T09:09:14 -05.00

The CAL compiler will crash if in a hurry, you change something incorrectly, and attempt to move something into a literal.

Like if you had mov r0, l13

and in a rush to change l13 to say, l14, you "fix" it with the following

mov l14, l13

The CAL compiler will crash. Just an FYI...as I already know the deprecation status of CAL. Just figured as I find bugs in whatever tool, I'd post it...Fortunately for me, that was the only change I had made to the code from compiling and working fine to crashing....made the error obvious. Had there been many many changes, I could see this as being prety frustrating!

Is there a library/api which can be used to access Catalyst Control Center?

cryis — 2011-09-23T02:53:58 -05.00

Is there a library/api which can be used to access Catalyst Control Center Profiles(Anti-Aliasing, Mipmap, etc...)?

NVidia has a proper library which can be used to access such Nvidia Control Panel Profiles(Anti-Aliasing, Mipmap, etc...), so we can manage/create/modify the profiles when

game is launched.

However, I checked ADL and AGS, but seems like there is no way to manage/modify Catalyst Control Center Profiles(Anti-Aliasing, Mipmap, etc...).

Would appreciate any help. Thank you.

amd gDebugger free for commercial use?

krrishnarraj — 2011-09-21T02:44:51 -05.00

Wanted to know if AMD gDebugger is free for commercial use. There is no mention about the licence in the gDebugger home page or anywhere else.

Thanks

How to get the complimentary license without internet connection

weihong — 2011-09-20T20:18:40 -05.00

I am using an HTTP proxy for internet. I already set the right proxy server in proxy settings. But it still doesn't work. I was wondering if there is another way to get this complimentary license.

Thanks,

-Wei

Accuracy of estimated Throughput in KernelAnalyzer

corry — 2011-09-20T19:00:08 -05.00

I suppose I'll likely find out soon, but was just curious to get some initial idea, how accurate is that number. If it says I will get 1B threads/sec, can I expect that number? 1/2? 1/4? I don't trust the number on some architectures since on the older ones, it says I'd use 2 GP registers, and see 3-4x performance vs a cayman. I kind-of find that hard to believe. I am targetting more modern GPU's, and perhaps using instructions not present on them, do they just get "optmized out" on the older architectures showing me huge performace numbers?

Please add support for GLSL tesselation shaders in GPU ShaderAnalyzer..

oscarbarenys1 — 2011-09-17T07:00:12 -05.00

After all it's more than one year avaiable on drivers

gDEbugger crashing

Apostate — 2011-09-16T20:47:48 -05.00

I was using gDEbugger 5.8 with Catalyst 11.5 and found it incredibly useful. I discovered there was a newer version (6.0) with integration in visual studio and figured I should try it to see if there are any new features I can use. I only use it for debugging OpenGL.

I installed the new gDEBugger, but it seemed to refuse to recognize my ATI Mobility Radeon HD 4650. I don't remember the exact message it was giving me, but it was pretty similar to this one from another post "The version of Catalyst you have installed does not contain support for this version of gDEBugger. Kernel debugging will be disabled. It is recommended to install the latest versions of both Catalyst and gDEBugger". It would also give me no debug information for OpenGL.

I updated my drivers to Catalyst 11.8 and now both gDEbugger 5.8 and 6.0 started to crash. 5.8 would let me hit the pause button or a breakpoint, but would crash if I tried to open any of the viewer windows. 6.0 crashes as soon as I hit the pause button or a breakpoint.

I noticed another post saying there might be something needed in 11.7, so I uninstalled 11.8, ran driver sweeper to clean up any remaining files and installed 11.7, but I have the same exact problem as with 11.8. Both gDEbuggers still crash.

Number of platforms:     1
Platform Profile:     FULL_PROFILE
Platform Version:     OpenCL 1.1 AMD-APP-SDK-v2.5 (709.2)
Platform Name:     AMD Accelerated Parallel Processing
Platform Vendor:     Advanced Micro Devices, Inc.
Platform Extensions:     cl_khr_icd cl_amd_event_callback cl_amd_offline_devices cl_khr_d3d10_sharing

Platform Name:     AMD Accelerated Parallel Processing
Number of devices:     2
Device Type:      CL_DEVICE_TYPE_GPU
Device ID:      4098
Max compute units:     8
Max work items dimensions:    3
    Max work items[0]:     128
    Max work items[1]:     128
    Max work items[2]:     128
Max work group size:     128
Preferred vector width char:    16
Preferred vector width short:    8
Preferred vector width int:    4
Preferred vector width long:    2
Preferred vector width float:    4
Preferred vector width double:   0
Native vector width char:    16
Native vector width short:    8
Native vector width int:    4
Native vector width long:    2
Native vector width float:    4
Native vector width double:    0
Max clock frequency:     450Mhz
Address bits:      32
Max memory allocation:    134217728
Image support:     No
Max size of kernel argument:    1024
Alignment (bits) of base address:   32768
Minimum alignment (bytes) for any datatype: 128
Single precision floating point capability
    Denorms:      No
    Quiet NaNs:      Yes
    Round to nearest even:    Yes
    Round to zero:     Yes
    Round to +ve and infinity:    Yes
    IEEE754-2008 fused multiply-add:   Yes
Cache type:      None
Cache line size:     0
Cache size:      0
Global memory size:     268435456
Constant buffer size:     65536
Max number of constant args:    8
Local memory type:     Global
Local memory size:     16384
Kernel Preferred work group size multiple: 32
Error correction support:    0
Unified memory for Host and Device:   0
Profiling timer resolution:    1
Device endianess:     Little
Available:      Yes
Compiler available:     Yes
Execution capabilities:
    Execute OpenCL kernels:    Yes
    Execute native function:    No
Queue properties:
    Out-of-Order:     No
    Profiling :      Yes
Platform ID:      000007FEE79FD348
Name:       ATI RV730
Vendor:      Advanced Micro Devices, Inc.
Device OpenCL C version:    OpenCL C 1.0
Driver version:     CAL 1.4.1457
Profile:      FULL_PROFILE
Version:      OpenCL 1.0 AMD-APP-SDK-v2.5 (709.2)
Extensions:      cl_khr_gl_sharing cl_amd_device_attribute_query cl_khr_d3d10_sharing

Device Type:      CL_DEVICE_TYPE_CPU
Device ID:      4098
Max compute units:     2
Max work items dimensions:    3
    Max work items[0]:     1024
    Max work items[1]:     1024
    Max work items[2]:     1024
Max work group size:     1024
Preferred vector width char:    16
Preferred vector width short:    8
Preferred vector width int:    4
Preferred vector width long:    2
Preferred vector width float:    4
Preferred vector width double:   0
Native vector width char:    16
Native vector width short:    8
Native vector width int:    4
Native vector width long:    2
Native vector width float:    4
Native vector width double:    0
Max clock frequency:     2194Mhz
Address bits:      64
Max memory allocation:    2147483648
Image support:     Yes
Max number of images read arguments:   128
Max number of images write arguments:   8
Max image 2D width:     8192
Max image 2D height:     8192
Max image 3D width:     2048
Max image 3D height:     2048
Max image 3D depth:     2048
Max samplers within kernel:    16
Max size of kernel argument:    4096
Alignment (bits) of base address:   1024
Minimum alignment (bytes) for any datatype: 128
Single precision floating point capability
    Denorms:      Yes
    Quiet NaNs:      Yes
    Round to nearest even:    Yes
    Round to zero:     Yes
    Round to +ve and infinity:    Yes
    IEEE754-2008 fused multiply-add:   No
Cache type:      Read/Write
Cache line size:     64
Cache size:      32768
Global memory size:     4260388864
Constant buffer size:     65536
Max number of constant args:    8
Local memory type:     Global
Local memory size:     32768
Kernel Preferred work group size multiple: 1
Error correction support:    0
Unified memory for Host and Device:   1
Profiling timer resolution:    466
Device endianess:     Little
Available:      Yes
Compiler available:     Yes
Execution capabilities:
    Execute OpenCL kernels:    Yes
    Execute native function:    Yes
Queue properties:
    Out-of-Order:     No
    Profiling :      Yes
Platform ID:      000007FEE79FD348
Name:       Intel(R) Core(TM)2 Duo CPU     T6600 @ 2.20GHz
Vendor:      GenuineIntel
Device OpenCL C version:    OpenCL C 1.1
Driver version:     2.0
Profile:      FULL_PROFILE
Version:      OpenCL 1.1 AMD-APP-SDK-v2.5 (709.2)
Extensions:      cl_khr_fp64 cl_amd_fp64 cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_int64_base_atomics cl_khr_int64_extended_atomics cl_khr_byte_addressable_store cl_khr_gl_sharing cl_ext_device_fission cl_amd_device_attribute_query cl_amd_vec3 cl_amd_media_ops cl_amd_popcnt cl_amd_printf cl_khr_d3d10_sharing

gDEBugger problems - stepping through kernels

Aiah — 2011-09-13T03:22:35 -05.00

I am having trouble using gDEBugger with vs2010.

I set a break point at a specific call to clEnqueueNDRangeKernel (using the gDEBugger API functions breakpoints). Trying to step into this function, in order to step through the kernel, causes an error to pop up : "Could not debug kernel. This kernel uses unsupported features. Please check the release notes for more details".

I found nothing relevant in the release notes. I commented out my kernel until left with the most basic operations and the error still occurs. This error also appears with the teapot example.

When I add a breakpoint to a kernel, the breakpoint is ignored because it "could not be bound".

My setup:

AMD Radeon HD 6900

Vendor ID:    OpenCL 1.1 AMD-APP-SDK-v2.5 (684.213)

Thanks!

gDEBugger 6.0 incompatible with Catalyst 11.8?

secretlair — 2011-09-09T16:44:08 -05.00

Error Message from gDEBugger:

The version of Catalyst you have installed does not contain support for this version of gDEBugger. Kernel debugging will be disabled. It is recommended to install the latest versions of both Catalyst and gDEBugger.

gDEBugger 6.0
Catalyst 11.8
ATI Radeon HD 5700 Series

OS Name    Microsoft Windows 7 Ultimate
Version    6.1.7600 Build 7600
System Type    x64-based PC
Processor    Intel(R) Core(TM) i7 CPU         920 @ 2.67GHz, 2668 Mhz, 4 Core(s), 8 Logical Processor(s)
BIOS Version/Date    American Megatrends Inc. 0801, 9/2/2009
SMBIOS Version    2.5
Locale    United States
Hardware Abstraction Layer    Version = "6.1.7600.16385"
Installed Physical Memory (RAM)    6.00 GB
Total Physical Memory    5.99 GB
Available Physical Memory    2.93 GB
Total Virtual Memory    12.0 GB
Available Virtual Memory    7.88 GB
Page File Space    5.99 GB
Page File    C:\pagefile.sys

AMD OpenCL Kernel Analyser : CRASH !

viewon01 — 2011-09-08T07:51:42 -05.00

Hi,

I have download the last version of this tool, but when I put my kernel code and request compilation, it crash.

If you want to I can send you my kernel (By example Micah Villmow has my email).

Krys

Advanced OpenCL debugging with AMD gDEBugger webinar - 9/8, 11am central time

stroia — 2011-09-06T12:09:32 -05.00

Developing robust parallel computing applications is difficult. In this talk Yaki will give the audience a first-hand tour of the gDEBugger, an OpenCL kernel source code debugger, integrated into Visual Studio. He will display advanced debugging techniques that help locate hard-to-find OpenCL related bugs and give you an introduction to other tools on developer.amd.com that he finds useful for OpenCL development.

Don't miss the chance to ask Yaki questions during this live webinar.

Time: Thursday 9/8/11, 11:00am US Central Standard Time

Register today!

Using gDEBugger 6.0 only for OpenCL CPU Kernel debugging?

OCLUser — 2011-09-05T06:31:47 -05.00

Hi all,

is there a way to make gDEBugger 6 debug at least OpenCL CPU Kernels even though a dedicated NVIDIA GPU is installed?

When I start the gDEBugger, I get a warning: You do not have an AMD GPU installed. Kernel debugging will be disabled.

Well, I just want to debug CPU kernels...

My development environment:

NVIDIA GPU Computing SDK 4 --> for GPU kernel creation

AMD APP SDK --> for CPU kernel creation

OS: Windows 7, Visual Studio 2010 Ultimate

My OpenCL program itself runs fine, both platforms are found without any changes in my VS project properties and the kernel can be executed on my CPU as well as on my GPU.

gDEBugger license generation failed

aoooooooooon — 2011-09-04T07:24:38 -05.00

I uninstalled gDEBugger 5.8, installed 6.0 and got license request dialog.
But I get an error dialog that states following:

Error - no license for gDEBugger
Failed to generate complimentary license. Please look for support in http://developer.amd.com/gDEBugger

After pressing OK, following URL is opended in browser
http://ausdevtoolsprd.amd.com/expired.php?productName=gDEBugger&freeVersionExpired=6.0.0.0

I think old gDEBugger free license cause some problems.
How can I get new license for gDEBugger 6.0?