[prev in list] [next in list] [prev in thread] [next in thread] 

List:       openjdk-graal-dev
Subject:    Re: TraceGPUInteraction print format error
From:       Doug Simon <doug.simon () oracle ! com>
Date:       2014-04-22 8:35:58
Message-ID: E90E2B3D-F001-4D8A-A3E2-5C7D80D15066 () oracle ! com
[Download RAW message or body]

I’ll push this fix - thanks for catching it.

On Apr 21, 2014, at 11:14 PM, Eric Caspole <eric.caspole@amd.com> wrote:

> Hi everybody,
> There seems  to be a little mistake in gpu.cpp where the "total initialized" is \
> printing the pointer as an int and not the intended field. What's more, I think it \
> would be better to print the count after the other work so it is clear how many \
> devices are available. Thanks,
> Eric
> 
> 
> Currently:
> 
> $ nice ./mx.sh --vm server --vmbuild product unittest -server -ea -esa \
> -XX:+TraceGPUInteraction hsail.test.IntAddTest Found supported nVidia device \
> [vendor=0x10de, device=0x1189] [CUDA] Ptx::Ptx::initialize
> [CUDA] Success: library linkage
> CUDA driver initialization: Success
> [CUDA] Number of compute-capable devices found: 1
> [CUDA] Got the handle of first compute-device
> [CUDA] Device 0 supports cuda compute capability 3.0
> [CUDA] Device 0 has Unified Addressing support
> [CUDA] Using GeForce GTX 670
> [CUDA] Success: Created context for device: 0
> [GPU] registered initialization of PTX (total initialized: 249372352)
> [HSAIL] library is /tmp/okraresource.dir_449166205788333768/libokra_x86_64.so
> [GPU] registered initialization of HSAIL (total initialized: 249372352)
> [CUDA] Ptx::get_execute_kernel_from_vm_address
> JUnit version 4.11
> .[HSAIL] heap=0x00007f2e08019f80
> [HSAIL] base=0x05a00000, capacity=210763776
> External method:com.oracle.graal.compiler.hsail.test.IntAddTest.run([I[I[II)V
> installCode0: ExternalCompilationResult
> 
> 
> With my fix:
> ]$ ./mx.sh --vm server --vmbuild debug unittest -server -ea -esa \
> -XX:+TraceGPUInteraction  -Dcom.amd.sumatra.offload.immediate=true \
> hsail.test.IntAddTest Found supported nVidia device [vendor=0x10de, device=0x1189]
> [CUDA] Ptx::Ptx::initialize
> [CUDA] Success: library linkage
> CUDA driver initialization: Success
> [CUDA] Number of compute-capable devices found: 1
> [CUDA] Got the handle of first compute-device
> [CUDA] Device 0 supports cuda compute capability 3.0
> [CUDA] Device 0 has Unified Addressing support
> [CUDA] Using GeForce GTX 670
> [CUDA] Success: Created context for device: 0
> [GPU] registered initialization of PTX (total initialized: 1)
> [HSAIL] library is /tmp/okraresource.dir_5849361204469827640/libokra_x86_64.so
> [GPU] registered initialization of HSAIL (total initialized: 2)
> [CUDA] Ptx::get_execute_kernel_from_vm_address
> JUnit version 4.11
> .[HSAIL] heap=0x00007fc61c01f978
> [HSAIL] base=0x05a00000, capacity=216530944
> External method:com.oracle.graal.compiler.hsail.test.IntAddTest.run([I[I[II)V
> 
> 
> Here is my suggested fix:
> 
> $ hg diff src/share/vm/runtime/gpu.cpp
> diff -r 9ff9f6643986 src/share/vm/runtime/gpu.cpp
> --- a/src/share/vm/runtime/gpu.cpp	Sat Apr 19 12:48:19 2014 +0200
> +++ b/src/share/vm/runtime/gpu.cpp	Mon Apr 21 17:07:53 2014 -0400
> @@ -32,10 +32,10 @@
> void Gpu::initialized_gpu(Gpu* gpu) {
> // GPUs are always initialized on the same thread so no need for locking
> guarantee(_initialized_gpus_count < MAX_GPUS, "oob");
> +  _initialized_gpus[_initialized_gpus_count++] = gpu;
> if (TraceGPUInteraction) {
> -    tty->print_cr("[GPU] registered initialization of %s (total initialized: %d)", \
> gpu->name(), _initialized_gpus); +    tty->print_cr("[GPU] registered \
> initialization of %s (total initialized: %d)", gpu->name(), \
> _initialized_gpus_count); }
> -  _initialized_gpus[_initialized_gpus_count++] = gpu;
> }
> 
> void Gpu::safepoint_event(SafepointEvent event) {


[prev in list] [next in list] [prev in thread] [next in thread]