On large BAR systems, for small-sized code-objects, we get performance
using direct memcpy due to latencies when doing the blit-copy.
[ROCm/ROCR-Runtime commit: da2607024b]
Invalidate only the address range that covers the newly copied
code-object. This avoids invalidating I$ for old code objects and thus
might increase I$ hit rate.
Invalidate only the address range that covers the newly copied
code-object. This avoids invalidating I$ for old code objects and thus
might increase I$ hit rate.
[ROCm/ROCR-Runtime commit: e969e01f54]
* Unit test fix- calculate aggregate sum based on instance size
* Update GRBM instance size
* aggregate sum
---------
Co-authored-by: Sushma Vaddireddy <svaddire@amd.com>
[ROCm/rocprofiler-sdk commit: 7045a4c689]
* Unit test fix- calculate aggregate sum based on instance size
* Update GRBM instance size
* aggregate sum
---------
Co-authored-by: Sushma Vaddireddy <svaddire@amd.com>
- Updating counters collected and tested for on Navi-based machines
- Update add CMake function to query GPU architectures
- Update decode tests to use new functions
[ROCm/rocprofiler-systems commit: 4c24975626]
- Updating counters collected and tested for on Navi-based machines
- Update add CMake function to query GPU architectures
- Update decode tests to use new functions
Create CP queue and SDMA queue should fail with invalid queue ring
buffer or ring buffer size.
Test unmap or free queue buffers should fail before queue is destroyed.
Use child process to test unmap CWSR buffer will evict queue.
Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Change-Id: I5dcd51d6b43445d19a986f8b0b82063e20348a5f
Create CP queue and SDMA queue should fail with invalid queue ring
buffer or ring buffer size.
Test unmap or free queue buffers should fail before queue is destroyed.
Use child process to test unmap CWSR buffer will evict queue.
Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Change-Id: I5dcd51d6b43445d19a986f8b0b82063e20348a5f
[ROCm/ROCR-Runtime commit: bd86fb1e63]
If unmap from GPU return failed, for example, unmap user queue buffer
while queue is active, we should not free obj->mapped_node_id_array,
otherwise, the following unmap user queue buffer after queue is
destroyed still return failed.
Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Change-Id: I32aeb18871c2e971d01900d92916c54680f5c9fa
If unmap from GPU return failed, for example, unmap user queue buffer
while queue is active, we should not free obj->mapped_node_id_array,
otherwise, the following unmap user queue buffer after queue is
destroyed still return failed.
Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Change-Id: I32aeb18871c2e971d01900d92916c54680f5c9fa
[ROCm/ROCR-Runtime commit: 3e6f51b715]
* increase hip api table version for HIP_MEMSET_NODE_PARAMS removed.
* update version in static_assert.
---------
Co-authored-by: Venkateshwar Reddy Kandula <vkandula@amd.com>
[ROCm/clr commit: ad44dd527f]
* increase hip api table version for HIP_MEMSET_NODE_PARAMS removed.
* update version in static_assert.
---------
Co-authored-by: Venkateshwar Reddy Kandula <vkandula@amd.com>
Defining int64_t, uint64_t, int32_t, uint32_t in HIPRTC
seem to result in conflicts with some apps as they use
their own definitions for these types. NVRTC also doesn't
define these. Hence remove them and replace their usage
with hip internal datatypes such as __hip_int64_t
Change-Id: Id314730bbd61c05f126f915c546052f956550a3f
[ROCm/clr commit: 946ea72c94]
Bhardwaj, Gopesh
David Yat Sin
Vaddireddy, Sushma
Baraldi, Giovanni
anujshuk-amd
Belton-Schure, Aidan
Ramakrishnan, Ranjith
Roy, Rakesh
Stojiljkovic, Vladana
Arandjelovic, Marko
Philip Yang
Kandula, Venkateshwar reddy
Gupta, Maneesh
Dittakavi, Satyanvesh