pipeline-gpu-kernel
Über
Diese Claude Skill strukturiert einen gekachelten GPU-Kernel um, um Software-Pipelining (Double-Buffering) zu implementieren, wodurch sich globale Speicherzugriffe mit Tensor-Core-Berechnungen überlappen lassen. Sie führt Entwickler durch die Umstrukturierung von Prolog/Schleife/Epilog und hilft bei der Auswahl zwischen LDG-Register- oder cp.async-Implementierungen basierend auf dem Compute/Load-Verhältnis. Die Skill umfasst außerdem die Überprüfung von Shared-Memory-Budgets gegen Occupancy-Grenzen und die Bestätigung der Überlappung auf SASS-Ebene.
Schnellinstallation
Claude Code
Empfohlennpx skills add pjt222/agent-almanac -a claude-code/plugin add https://github.com/pjt222/agent-almanacgit clone https://github.com/pjt222/agent-almanac.git ~/.claude/skills/pipeline-gpu-kernelKopieren Sie diesen Befehl und fügen Sie ihn in Claude Code ein, um diese Fähigkeit zu installieren
Dokumentation
Pipeline GPU Kernel
Apply software pipelining (double-buffer) to tiled GPU kernel. Global loads for tile N+1 overlap Tensor Core compute on tile N. Convert sequential load-sync-compute-sync K-loop into prologue/loop/epilogue. Pick LDG-register or cp.async (LDGSTS) by compute/load ratio. Verify shared mem stays below occupancy cliff. Confirm overlap in SASS.
When Use
analyze-kernel-bottleneckflagged memory-bound kernel, low compute/load ratio per tile- Warp interleave alone won't hide DRAM latency (~300 cycles on GA104)
- Kernel has sequential load-sync-compute-sync K-loop, can restructure
- Skip when compute/load ratio high (>20:1) and 8+ warps active
Inputs
- Required: CUDA kernel source (
.cu) with tiled K-loop, separate load and compute phases - Required: Target GPU arch (e.g. GA104 / sm_86 — sets smem cliff and occupancy)
- Required: Current tile sizes (BM, BN, BK) and dtype (FP16, FP32, INT8)
- Optional: Compute/load ratio per tile (from
analyze-kernel-bottleneck; estimate if missing) - Optional: Benchmark baseline (non-pipelined perf at target size)
Steps
Step 1: Verify Preconditions
Confirm tiled K-loop has distinct load and compute phases split by __syncthreads(). Compute doubled smem cost. Verify under architecture cliff.
- Find K-loop in kernel. Must follow: load A and B tiles global → shared,
__syncthreads(), compute (HMMA/IMMA/FFMA) on shared tiles,__syncthreads(). - Record single-buffer smem sizes:
smem_a_size = BM * BK * sizeof(T)andsmem_b_size = BK * BN * sizeof(T). - Compute double-buffer cost:
smem_doubled = smem_a_size * 2 + smem_b_size * 2. - Compare to architecture cliff. GA104 (sm_86): 100 KB max smem/SM, cliff at 50 KB/block (above 50 KB → 1 block/SM = 4 warps, 2x occupancy collapse).
Single buffer: smem_a[BM*BK] + smem_b[BK*BN] = 2 KB + 2 KB = 4 KB
Double buffer: smem_a[2][BM*BK] + smem_b[2][BK*BN] = 4 KB + 4 KB = 8 KB
8 KB << 50 KB cliff -> 2 blocks/SM -> 8 warps
- Verify loop count:
num_tiles = K / BK. Pipelining needsnum_tiles >= 2(one prologue + one main iteration min).
Got: Smem budget table — single-buffer and double-buffer costs. Doubled allocation under architecture cliff. At least 2 blocks/SM occupancy.
If fail: Doubled smem above cliff? Shrink tile (halve BK or BM) until smem_doubled <= 50 KB for GA104. Or use register-only prefetch (LDG variant) — no double smem, stage in registers, write same single buffer after __syncthreads().
Step 2: Pick Variant
Choose LDG-register or cp.async (LDGSTS) by compute/load ratio per tile.
- Compute ratio:
ratio = (2 * BM * BN * BK) / ((BM * BK + BK * BN) * sizeof(T))for GEMM-like kernels (2 FLOPs per multiply-add, bytes loaded per tile). - Apply rule:
LDG-register variant (ratio >= 5 or CUDA < 11.0):
- LDG tile N+1 to registers (non-blocking global loads).
- Compute on
buf[N % 2](overlaps outstanding LDGs). __syncthreads(), then STS registers tobuf[(N+1) % 2],__syncthreads().- Simpler. No pipeline API dependency.
- Adds register pressure: ~
(BM * BK + BK * BN) / BLOCK_SIZEregisters per thread for staging.
cp.async (LDGSTS) variant (ratio < 5, CUDA >= 11.0):
__pipeline_memcpy_asynctile N+1 directly tobuf[(N+1) % 2](async, bypass register file).__pipeline_commit()before compute.- Compute on
buf[N % 2]. __pipeline_wait_prior(0)+__syncthreads()after compute.- Better overlap, zero register pressure for prefetch. Needs
#include <cuda_pipeline.h>.
- Decision thresholds (measured GA104, IGEMM 4096x4096x4096):
- Ratio < 5:1 — pick cp.async (+35% measured on IGEMM).
- Ratio 5-20:1 — implement both, benchmark.
- Ratio > 20:1 — pipelining likely won't help (warp interleave enough).
Got: Picked variant with reason — based on compute/load ratio and target arch.
If fail: Ratio ambiguous (5-20:1)? Implement both, benchmark. cp.async safer default when CUDA version supports.
Step 3: Restructure K-Loop
Convert sequential load-sync-compute-sync loop into pipelined prologue/loop/epilogue.
-
Identify three sections: Original loop body splits into three:
- Prologue: Load tile 0 to
buf[0], sync, enter main loop. - Main loop: For tiles 1 through
num_tiles - 1, overlap loading tile N+1 with computing tile N. - Epilogue: Compute last tile (already loaded by final main iteration).
- Prologue: Load tile 0 to
-
LDG-register variant structure:
// === LDG-register variant ===
// Prologue: load tile 0 into buf[0]
cooperative_load_tile(smem_a[0], smem_b[0], global_a, global_b, /*k_offset=*/0);
__syncthreads();
for (int tile = 0; tile < num_tiles - 1; tile++) {
int cur_buf = tile & 1;
int next_buf = 1 - cur_buf;
// Phase 1: LDG next tile into registers (non-blocking)
float reg_a[ELEMS_PER_THREAD_A], reg_b[ELEMS_PER_THREAD_B];
prefetch_tile_to_registers(reg_a, reg_b, global_a, global_b,
(tile + 1) * BK);
// Phase 2: Compute on current buffer (overlaps with LDG flight)
tensor_core_mma(smem_a[cur_buf], smem_b[cur_buf], acc);
// Phase 3: Drain registers into next buffer
__syncthreads();
store_registers_to_smem(smem_a[next_buf], smem_b[next_buf],
reg_a, reg_b);
__syncthreads();
}
// Epilogue: compute last tile
tensor_core_mma(smem_a[(num_tiles - 1) & 1], smem_b[(num_tiles - 1) & 1], acc);
- cp.async variant structure:
// === cp.async variant ===
#include <cuda_pipeline.h>
// Prologue: async load tile 0 into buf[0]
cpasync_load_tile(smem_a[0], smem_b[0], global_a, global_b, /*k_offset=*/0);
__pipeline_commit();
__pipeline_wait_prior(0);
__syncthreads();
for (int tile = 0; tile < num_tiles - 1; tile++) {
int cur_buf = tile & 1;
int next_buf = 1 - cur_buf;
// Phase 1: cp.async next tile into next buffer (async, direct to smem)
cpasync_load_tile(smem_a[next_buf], smem_b[next_buf],
global_a, global_b, (tile + 1) * BK);
__pipeline_commit();
// Phase 2: Compute on current buffer (overlaps with LDGSTS in flight)
tensor_core_mma(smem_a[cur_buf], smem_b[cur_buf], acc);
// Phase 3: Wait for async copies to complete
__pipeline_wait_prior(0);
__syncthreads();
}
// Epilogue: compute last tile
tensor_core_mma(smem_a[(num_tiles - 1) & 1], smem_b[(num_tiles - 1) & 1], acc);
- Verify loop count: main loop runs
num_tiles - 1iterations (tiles 0 throughnum_tiles - 2indexing compute, loading tiles 1 throughnum_tiles - 1). Epilogue computes tile loaded last iteration.
Got: Restructured K-loop source with clear prologue, main loop, epilogue for picked variant.
If fail: Most common bug: off-by-one in buffer indexing or forget epilogue compute pass. Verify: prologue loads buf[0], first main iteration computes buf[0] and loads buf[1], second iteration computes buf[1] and loads buf[0]. Epilogue computes buf[(num_tiles - 1) & 1].
Step 4: Implement Double-Buffer
Declare double-buffered smem. Implement load functions.
- Replace single-buffer smem decls with double-buffered arrays:
// Before (single buffer)
__shared__ half smem_a[BM * BK];
__shared__ half smem_b[BK * BN];
// After (double buffer)
__shared__ half smem_a[2][BM * BK];
__shared__ half smem_b[2][BK * BN];
- For cp.async variant, implement async load with pipeline API:
__device__ void cpasync_load_tile(half* dst_a, half* dst_b,
const half* src_a, const half* src_b,
int k_offset) {
// Each thread copies its portion (16 bytes = 8 half values per cp.async)
int tid = threadIdx.x;
int bytes_per_thread = 16; // cp.async.cg supports 4, 8, or 16 bytes
// A tile: BM * BK elements, distributed across BLOCK_SIZE threads
int elems_a = BM * BK / BLOCK_SIZE;
for (int i = 0; i < elems_a; i += 8) {
int idx = tid * elems_a + i;
__pipeline_memcpy_async(dst_a + idx,
src_a + k_offset * BM + idx,
bytes_per_thread);
}
// B tile: BK * BN elements, distributed similarly
int elems_b = BK * BN / BLOCK_SIZE;
for (int i = 0; i < elems_b; i += 8) {
int idx = tid * elems_b + i;
__pipeline_memcpy_async(dst_b + idx,
src_b + k_offset * BN + idx,
bytes_per_thread);
}
}
- For LDG variant, implement register staging arrays and store functions:
// Declare register staging (size = elements per thread)
half reg_a[BM * BK / BLOCK_SIZE];
half reg_b[BK * BN / BLOCK_SIZE];
// Prefetch: LDG from global to registers (non-blocking, issued early)
for (int i = 0; i < BM * BK / BLOCK_SIZE; i++) {
int idx = threadIdx.x * (BM * BK / BLOCK_SIZE) + i;
reg_a[i] = global_a[k_offset * BM + idx];
}
// ... similarly for reg_b
// Store: STS from registers to shared memory (after __syncthreads)
for (int i = 0; i < BM * BK / BLOCK_SIZE; i++) {
int idx = threadIdx.x * (BM * BK / BLOCK_SIZE) + i;
smem_a[next_buf][idx] = reg_a[i];
}
- Keep
__launch_bounds__(BLOCK_SIZE)on kernel — gives compiler accurate occupancy info. - Compile:
nvcc --cubin -arch=sm_86 -O2 -o kernel.sm_86.cubin kernel.cu.
Got: Compilable kernel with double-buffered smem and picked load mechanism. Cubin builds, no errors.
If fail: Compile fails on pipeline API? Verify #include <cuda_pipeline.h> present, CUDA toolkit >= 11.0. Register spills (check nvcc --resource-usage)? Shrink register staging array sizes — bigger BLOCK_SIZE or smaller BK.
Step 5: Verify Correctness
Run pipelined kernel against CPU reference. Confirm same numerical output.
- Compile bench:
nvcc -arch=sm_86 -O2 -o bench bench.cu -lcuda -I../../phase2/common. - Run small first (512x512x512). Catch indexing bugs before scaling.
- Apply tolerance for dtype:
- INT8 Tensor Core (IMMA):
abs=0.5, rel=0.1 - FP16 Tensor Core (HMMA):
abs=1e-2, rel=1e-2 - FP32 scalar (FFMA):
abs=1e-3, rel=1e-3
- INT8 Tensor Core (IMMA):
- Pipelining doesn't change arithmetic — only reorders loads. Correctness fails? Bug in buffer indexing, not compute logic.
- Test at target size (e.g. 4096x4096x4096) — verify boundary handling.
Got: PASS at small and target sizes with same error bounds as non-pipelined baseline.
If fail: Buffer indexing bug most likely. Verify: compute reads buf[tile & 1], loads write buf[1 - (tile & 1)]. Epilogue uses (num_tiles - 1) & 1, not num_tiles & 1. For cp.async: __pipeline_wait_prior(0) must finish before __syncthreads() — else compute reads partial data.
Step 6: Benchmark and Compare
Measure pipelined vs non-pipelined baseline at target size.
- Run non-pipelined baseline. Record GFLOPS or bandwidth (depends on kernel type).
- Run each pipelined variant. Record same metric.
- Compute speedup:
speedup = pipelined_metric / baseline_metric. - Expected gains by compute/load ratio (measured on GA104):
- Low ratio (<5:1): +15-35% from cp.async (IGEMM measured: LDG +18%, cp.async +35% at 4096x4096x4096).
- Medium ratio (5-20:1): +5-15%.
- High ratio (>20:1): 0-5% or regression.
- Implemented both? Pick faster for production.
| Variant | GFLOPS | Speedup vs Baseline |
|------------------|--------|---------------------|
| Baseline | XXX | 1.00x |
| LDG-register | XXX | X.XXx |
| cp.async (LDGSTS)| XXX | X.XXx |
Got: Perf comparison table showing improvement. Picked variant shows measurable speedup matching compute/load ratio prediction.
If fail: Perf regresses? Check three things: (1) SASS for unexpected instruction overhead (extra BAR.SYNC, register spills). (2) Smem stayed below occupancy cliff — verify with nvcc --resource-usage or cuobjdump -res-usage. (3) Problem size produces enough tiles (K / BK >= 4) to amortize prologue/epilogue overhead.
Step 7: Verify SASS Overlap
Inspect compiled SASS. Confirm global loads and Tensor Core instructions overlap in main loop body.
- Disassemble:
cuobjdump -sass kernel.sm_86.cubin | grep -E 'IMMA|HMMA|LDGSTS|LDG|BAR'. - In main loop body, verify ordering:
LDGSTSorLDGinstructions appear beforeHMMAorIMMA.- No
BAR.SYNCbetween load and compute (must overlap freely in warp scheduler). BAR.SYNCappears after compute block — gates next iteration's use of loaded data.
- Check stall codes on HMMA/IMMA — S08 for HMMA pipeline delay expected, unavoidable. S01-S04 for IMMA normal. Stalls on LDG/LDGSTS should be low (S01) — warp scheduler can switch to compute while loads in flight.
- Count total HMMA/IMMA per loop iteration — must match non-pipelined version (pipelining doesn't change compute volume).
# Full SASS pipeline verification
cuobjdump -sass kernel.sm_86.cubin | grep -E 'IMMA|HMMA|LDGSTS|LDG|BAR'
# Count compute instructions per loop
cuobjdump -sass kernel.sm_86.cubin | grep -c 'HMMA\|IMMA'
# Check for register spills
nvcc --resource-usage --cubin -arch=sm_86 -O2 kernel.cu 2>&1 | grep -i spill
Got: Annotated SASS showing load-before-compute pattern. No intervening barriers. Zero register spills.
If fail: Compiler reordered loads after compute (defeats overlap)? Try: (1) #pragma unroll 1 on main loop — prevents over-aggressive unroll. (2) Split load and compute into distinct inline functions — sequencing hint. (3) Use asm volatile("" ::: "memory") as compiler fence between load and compute (last resort — may inhibit other opts).
Checks
- Double-buffer smem under architecture cliff (GA104: 50 KB/block)
- Both buffers used alternately (
buf[tile & 1]pattern) - Prologue loads tile 0 into
buf[0] - Epilogue computes last tile from
buf[(num_tiles - 1) & 1] - Correctness PASS vs CPU reference at small and target sizes
- SASS confirms load/compute overlap (no
BAR.SYNCbetween LDGSTS/LDG and IMMA/HMMA) - Perf improved over non-pipelined baseline
- No register spill from LDG variant (check
nvcc --resource-usage)
Pitfalls
- Cross smem cliff by doubling buffers — GA104 cliff is 50 KB/block, not 64 KB. Always compute
smem_doubledbefore implementing. Kernel using 28 KB single-buffered jumps to 56 KB doubled, crosses cliff, halves occupancy. Turns +20% pipelining gain into -50% occupancy regression. - Forget epilogue compute pass — Last tile loaded in final main iteration needs own compute phase outside loop. Without it, last BK columns of K dimension silently dropped — wrong results may look like small numerical errors, not obvious failures.
- Buffer indexing off-by-one — Use
buf[tile & 1]for current compute buffer,buf[1 - (tile & 1)]for next load buffer. Common mistake:buf[(tile + 1) & 1]for next buffer — equivalent tobuf[1 - (tile & 1)]only when buffer count is 2 — but reads wrong if accidentally applied to compute index. - cp.async commit/wait order —
__pipeline_commit()BEFORE compute (seals batch of async copies).__pipeline_wait_prior(0)AFTER compute (blocks until committed copies complete). Swap them → async copies become synchronous → all overlap benefit gone. - Missing __syncthreads — In LDG variant, need
__syncthreads()between compute and STS drain (compute finishes reading current buffer before overwrite). Another__syncthreads()after STS drain (all threads finish writing before next iteration reads). In cp.async variant,__syncthreads()after__pipeline_wait_prior(0)ensures all threads see completed async copies. - Boundary handling in cp.async —
__pipeline_memcpy_asyncneeds source address valid and aligned. At matrix edges whereKnot multiple ofBK, last tile may read out of bounds. Fall back to scalar loads with bounds checking for final tile, or pad input matrices to multiple of BK.
See Also
analyze-kernel-bottleneck— identify if kernel is memory-bound. Compute the compute/load ratio that drives variant selection.
GitHub Repository
Verwandte Skills
llamaguard
AndereLlamaGuard ist Metas 7-8B-Parameter-Modell zur Moderation von LLM-Eingaben und -Ausgaben in sechs Sicherheitskategorien wie Gewalt und Hassrede. Es bietet eine Genauigkeit von 94-95 % und kann mit vLLM, Hugging Face oder Amazon SageMaker eingesetzt werden. Nutzen Sie diese Skill, um Inhaltsfilterung und Sicherheitsguardrails einfach in Ihre KI-Anwendungen zu integrieren.
cost-optimization
AndereDiese Claude Skill unterstützt Entwickler bei der Optimierung von Cloud-Kosten durch Ressourcen-Dimensionierung, Tagging-Strategien und Ausgabenanalysen. Sie bietet einen Rahmen zur Senkung von Cloud-Ausgaben und zur Implementierung von Kosten-Governance für AWS, Azure und GCP. Nutzen Sie sie, wenn Sie Infrastrukturkosten analysieren, Ressourcen richtig dimensionieren oder Budgetvorgaben einhalten müssen.
quantizing-models-bitsandbytes
AndereDiese Fähigkeit quantisiert LLMs auf 8-Bit- oder 4-Bit-Präzision mittels bitsandbytes und erreicht dabei eine Speicherreduzierung von 50–75 % bei minimalem Genauigkeitsverlust. Sie ist ideal für den Betrieb größerer Modelle mit begrenztem GPU-Speicher oder zur Beschleunigung von Inferenzvorgängen und unterstützt Formate wie INT8, NF4 und FP4. Die Fähigkeit integriert sich in HuggingFace Transformers und ermöglicht QLoRA-Training sowie 8-Bit-Optimierer.
dispatching-parallel-agents
AndereDiese Claude-Fähigkeit verteilt mehrere Agenten, um drei oder mehr unabhängige Probleme gleichzeitig zu untersuchen und zu beheben. Sie ist für Szenarien konzipiert, die unabhängige Fehler umfassen, die ohne gemeinsamen Zustand oder Abhängigkeiten gelöst werden können. Die Kernfähigkeit ist die parallele Problemlösung, bei der pro unabhängigem Problembereich ein Agent zugewiesen wird, um die Effizienz zu maximieren.
