leonarski_fandClaude Opus 5.5
c644c4e38a
The speculative geometry probe (StartSpeculativeGeometryProbe) runs an indexing-only pass on a copy of the run beside the pass's scaling merge. It holds ~4.5 GB of engines (25 image-sized spot-finding buffers, the per-engine tables, FFT indexers) plus what its streams' pool retains, for a few seconds. When a merge allocation landed in that window and did not fit, RotationScaleMergeGPU::Impl::Alloc reported "needs more GPU memory than this card has ... too large for GPU scaling", although the per-observation arrays were 1.3 GB on a 16.6 GB card and the set runs fine alone. Timing-dependent: one full-battery failure, not reproduced in ~15 plain reruns; reproduced deterministically by letting the probe hold extra device memory. - GPUWorkBeside (common/CUDAWrapper): a process-wide count of GPU work running beside the main line, with a condition variable signalled when the last one ends. The speculative probe holds one for its whole pass. - Alloc: on failure, wait for that work to end (bounded, 10 min, then a "GPU busy" error), then ask for the same buffer once more. Nothing else in flight means no wait, so a genuine shortage still fails at once. The computation is the same whenever the allocation succeeds, so results do not depend on the wait. - "Too large for this card" is now said only by the early check of the per-observation arrays against total device memory; a later shortage with nothing running beside gets its own message (buffer size, free/total, and that another program or the set's size is the cause). Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01D1G8gJVAy6gp1K5Dz3NE5C
68 lines
3.3 KiB
C++
68 lines
3.3 KiB
C++
// SPDX-FileCopyrightText: 2024 Filip Leonarski, Paul Scherrer Institute <filip.leonarski@psi.ch>
|
|
// SPDX-License-Identifier: GPL-3.0-only
|
|
|
|
#pragma once
|
|
|
|
#include <chrono>
|
|
#include <cstdint>
|
|
#include <string>
|
|
#include <vector>
|
|
|
|
int32_t get_gpu_count();
|
|
|
|
// Names of the visible GPUs, in device order and one entry per device, so repeated cards repeat.
|
|
// Empty without CUDA and on a machine with no device, which is also what get_gpu_count() == 0 says.
|
|
std::vector<std::string> get_gpu_names();
|
|
|
|
// The same list collapsed for a person: "4x NVIDIA A100-SXM4-80GB", or several such groups separated
|
|
// by ", " on a mixed machine. Empty when no GPU is visible.
|
|
std::string get_gpu_description();
|
|
void set_gpu(int32_t dev_id);
|
|
|
|
// Pin the calling thread to the next GPU in round-robin order, using a process-wide counter
|
|
// (counter++ % get_gpu_count()). Call once per thread; no thread id needed. No-op when no GPU
|
|
// is visible. Honours CUDA_VISIBLE_DEVICES via get_gpu_count().
|
|
void pin_gpu();
|
|
|
|
// From here on, pin_gpu() also keeps the calling thread on the CPUs of the NUMA node its GPU is
|
|
// attached to (ThreadAffinity.h) - on a machine with one node, or without CUDA, nothing changes.
|
|
// Off unless a program asks for it.
|
|
void enable_gpu_numa_binding();
|
|
|
|
// pin_gpu() onto a given device: set_gpu(dev_id), and the NUMA binding above where it is enabled.
|
|
// For a worker that takes its card by index rather than round-robin.
|
|
void pin_gpu(int32_t dev_id);
|
|
|
|
// Have every GPU's host threads BLOCK in a synchronisation (cudaDeviceScheduleBlockingSync) instead
|
|
// of spinning on a core until the device finishes. Must be called before anything creates a CUDA
|
|
// context; a device that already has one keeps the flags it was created with. No-op without CUDA.
|
|
void set_gpu_blocking_sync();
|
|
|
|
// Drop the error CUDA has recorded for the calling thread. Call it where a CUDA failure has been
|
|
// HANDLED - a device route that fell back to the host, an indexing attempt whose failure was turned
|
|
// into a result - because the error otherwise stays as the thread's last error and the next
|
|
// cuda_err(cudaGetLastError()) after some later kernel launch reports it, over work that went fine.
|
|
// A sticky error (an illegal access, say) is not cleared by this, and nothing here pretends it is:
|
|
// the context is gone in that case and every later call fails on its own. No-op without CUDA.
|
|
void cuda_clear_error();
|
|
|
|
// Throw if the CUDA context is lost - a sticky error (an illegal access, say) that every later call
|
|
// on this device will fail with. For a device route about to fall back to the host: there is no
|
|
// fallback from that, and falling back only moves the failure somewhere less clear. A handled,
|
|
// non-sticky failure passes. No-op without CUDA.
|
|
void cuda_throw_if_context_lost();
|
|
|
|
// GPU work that runs beside the main line of a process and gives all of its device memory back when it
|
|
// ends: a probe pass run on a copy of the run while the run itself scales and merges. Hold one for as
|
|
// long as that work runs. Build-independent - it only counts.
|
|
class GPUWorkBeside {
|
|
public:
|
|
GPUWorkBeside();
|
|
~GPUWorkBeside();
|
|
GPUWorkBeside(const GPUWorkBeside &) = delete;
|
|
GPUWorkBeside &operator=(const GPUWorkBeside &) = delete;
|
|
};
|
|
|
|
// Wait until no GPUWorkBeside is held, for at most `timeout`. True once none is - at once if none was.
|
|
bool wait_for_gpu_work_beside(std::chrono::seconds timeout);
|