Matthew Poremba has uploaded this change for review. (
https://gem5-review.googlesource.com/c/public/gem5/+/65251?usp=email )
Change subject: gpu-compute: Chunkify AMDKernelCode read from device
......................................................................
gpu-compute: Chunkify AMDKernelCode read from device
The AMDKernelCode object can span potentially span two pages. Currently
the copy loop from device memory only translates once at the base
address.
This changeset translates one cache line at a time before copying and
has the ancillary benefit for cleaning up this code a bit.
Change-Id: I602bc12d8f8c5d3a3e57ab3f42f7dd3df58dc144
---
M src/gpu-compute/gpu_command_processor.cc
1 file changed, 35 insertions(+), 9 deletions(-)
diff --git a/src/gpu-compute/gpu_command_processor.cc
b/src/gpu-compute/gpu_command_processor.cc
index d46ace6..02c75def 100644
--- a/src/gpu-compute/gpu_command_processor.cc
+++ b/src/gpu-compute/gpu_command_processor.cc
@@ -149,6 +149,9 @@
phys_addr, tmp_bytes, BaseMMU::Mode::Read,
is_system_page);
}
+ DPRINTF(GPUCommandProc, "kernobj vaddr %#lx paddr %#lx size %d s:%d\n",
+ disp_pkt->kernel_object, phys_addr, sizeof(AMDKernelCode),
+ is_system_page);
/**
* The kernel_object is a pointer to the machine code, whose entry
@@ -167,20 +170,27 @@
} else {
assert(FullSystem);
DPRINTF(GPUCommandProc, "kernel_object in device, using device
mem\n");
- // Read from GPU memory manager
- uint8_t raw_akc[sizeof(AMDKernelCode)];
- for (int i = 0; i < sizeof(AMDKernelCode) / sizeof(uint8_t); ++i) {
- Addr mmhubAddr = phys_addr + i*sizeof(uint8_t);
+
+ // Read from GPU memory manager one cache line at a time to prevent
+ // rare cases where the AKC spans two memory pages.
+ ChunkGenerator gen(disp_pkt->kernel_object, sizeof(AMDKernelCode),
+ system()->cacheLineSize());
+ for (; !gen.done(); gen.next()) {
+ Addr chunk_addr = gen.addr();
+ int vmid = 1;
+ unsigned dummy;
+
walker->startFunctional(gpuDevice->getVM().getPageTableBase(vmid),
+ chunk_addr, dummy, BaseMMU::Mode::Read,
+ is_system_page);
+
Request::Flags flags = Request::PHYSICAL;
- RequestPtr request = std::make_shared<Request>(
- mmhubAddr, sizeof(uint8_t), flags,
walker->getDevRequestor());
+ RequestPtr request = std::make_shared<Request>(chunk_addr,
+ system()->cacheLineSize(), flags,
walker->getDevRequestor());
Packet *readPkt = new Packet(request, MemCmd::ReadReq);
- readPkt->allocate();
+ readPkt->dataStatic((uint8_t *)&akc + gen.complete());
system()->getDeviceMemory(readPkt)->access(readPkt);
- raw_akc[i] = readPkt->getLE<uint8_t>();
delete readPkt;
}
- memcpy(&akc, &raw_akc, sizeof(AMDKernelCode));
}
DPRINTF(GPUCommandProc, "GPU machine code is %lli bytes from start of
the "
--
To view, visit
https://gem5-review.googlesource.com/c/public/gem5/+/65251?usp=email
To unsubscribe, or for help writing mail filters, visit
https://gem5-review.googlesource.com/settings
Gerrit-Project: public/gem5
Gerrit-Branch: develop
Gerrit-Change-Id: I602bc12d8f8c5d3a3e57ab3f42f7dd3df58dc144
Gerrit-Change-Number: 65251
Gerrit-PatchSet: 1
Gerrit-Owner: Matthew Poremba <matthew.pore...@amd.com>
Gerrit-MessageType: newchange
_______________________________________________
gem5-dev mailing list -- gem5-dev@gem5.org
To unsubscribe send an email to gem5-dev-le...@gem5.org
