Archives Discussions

[oscarbarenys1]

only for GPUs5xxx tested

Well I have been holding this trick on my head for over a month now..

More info on my blog coming soon: oscarbg.blogspot.com

Really you can enable image support set:

set GPU_IMAGES_SUPPORT=1

or export

tested on 5870 and amd stream 2.0 and hotfix 9.12 only works for

2d images.. 

Similarly you can enable byte_addresable_support (but seems is not using RAW UAVs) and some Nvidia samples work (histogram64) with GPU_BYTE_ADDRESSABLE_STORE

Also doules extension reporting

GPU_DOUBLE_PRECISION

and gl

CL_KHR_GL_SHARIN

[MicahVillmow]

Just a warning to those who feel adventurous. These 'features' are not supported, are not fully tested or implemented, and using them can result in undefined behavior. I would strongly recommend waiting until our releases that have these features enabled by default before using them. Use them at your own risk.

[nou]

i look into libatiocl.so and it is look like there is plenty debug/developer options.

LOG_LEVEL
BREAK_ON_LOG_WARNING
BREAK_ON_LOG_ERROR
DEBUG_GPU_FLAGS
GPU_COMPILER_OPTS is this similiar to CPU_COMPILER_OPTIONS?
GPU_COMPILER_BACKEND_OPTIONS
GPU_MEMORY_COHERENCY
GPU_INTEROP_EMULATION
FORCE_CAL_TARGET
ATISTREAMSDKROOT
LLVM_LINK_LIBS
LLVM_UNROLL_THRESHOLD
CQ_THREAD_STACK_SIZE
CPU_ENABLE_ALL this should enable CPU without SSE3 as OpenCL device
CPU_WORKER_THREAD_STACK_SIZE
CPU_COMPILER_OPTIONS you can specify additional options to compiler for example "-g"
CPU_MAX_COMPUTE_UNITS you can set that OpenCL will use only 1,2,3 core(s) of CPU
GPU_MAX_WORKGROUP_SIZE
CPU_MAX_WORKGROUP_SIZE
CPU_STACK_ALIGNMENT
CPU_ALIGN_STRUCTS
CPU_MAX_ALLOC_SIZE
PARAMETERS_MIN_ALIGNMENT
MEMOBJ_BASE_ADDR_ALIGN
APPLE_USE_THREAD_POLICY
GPU_DOUBLE_PRECISION set to 1 and you get cl_khr_fp64 in extension list
GPU_IMAGES_SUPPORT
GPU_BYTE_ADDRESSABLE_STORE
GPU_ATOMICS set to 0 and atomics disappear from list
GPU_DEVICE_ORDINAL
REMOTE_ALLOC
GPU_REPORT_EXTENSIONS setting to 0 turn off extensions, or at least it do not report it.
CL_CAL_VERSION_MAJOR setting this to 2 lead to CAL version mismatch error
CL_CAL_VERSION_MINOR
CL_CAL_VERSION_IMPLEMENTATION
GPU_INITIAL_HEAP_SIZE
GPU_MAX_HEAP_SIZE set to 512 and OpenCL report 512MB of global memory. dunno if it will realy work.
GPU_MAX_ALLOC_SIZE
GPU_HEAP_GROWTH_INCREMENT
GPU_STAGING_BUFFER_SIZE
GPU_DUMP_DEVICE_KERNEL
GPU_BINARY_DUMP_FLA
CL_KHR_GL_SHARING CLInfo report cl_khr_gl_sharing but only for CPU device
GPU_DEBUG_OUTPUT
ENABLE_CAL_SHUTDOWN

and also i find this extensions listed. cl_khr_icd
cl_khr_fp64 cl_khr_3d_image_writes  cl_khr_fp16 cl_khr_gl_sharing cl_amd_d3d9_interop cl_amd_d3d10_interop cl_khr_select_fprounding_mode

this is from SDK 2.01

[moozoo]

>CPU_ENABLE_ALL this should enable CPU without SSE3 as OpenCL device

I'd really like to official support for at least for SSE2 Processors.

I tried CPU_ENABLE_ALL on a Northwood Pentium 4 and the SDK samples I tried all seem to work.

According to the Jan 2010 Steam hardware survey SSE2 is available on 98% of Steam players machines (~25 million)

[apollo_maverick]

GREAT! after setting the switch, with Stream SDK 2.01 and Catalyst 10.2 driver, from CLInfo.exe, it seems that Cypress chip supports 3D images.