From ba723e79d5c426fb4bc899b869bc0b497e72a152 Mon Sep 17 00:00:00 2001 From: Birte Kristina Friesel <birte.friesel@uos.de> Date: Fri, 8 Mar 2024 16:46:01 +0100 Subject: SEL: introduce alloc/load/free overhead parameters; vary them for fgbs24a --- SEL/Makefile | 5 +- SEL/host/app.c | 159 +++++++++++++++++++++++++++++++++++----------------- SEL/run-fgbs24a.sh | 10 +++- SEL/support/timer.h | 6 +- 4 files changed, 121 insertions(+), 59 deletions(-) (limited to 'SEL') diff --git a/SEL/Makefile b/SEL/Makefile index aa10246..c45bf01 100644 --- a/SEL/Makefile +++ b/SEL/Makefile @@ -2,13 +2,16 @@ NR_TASKLETS ?= 16 BL ?= 10 NR_DPUS ?= 1 ENERGY ?= 0 +WITH_ALLOC_OVERHEAD ?= 0 +WITH_LOAD_OVERHEAD ?= 0 +WITH_FREE_OVERHEAD ?= 0 COMMON_INCLUDES := support HOST_SOURCES := $(wildcard host/*.c) DPU_SOURCES := $(wildcard dpu/*.c) COMMON_FLAGS := -Wall -Wextra -g -I${COMMON_INCLUDES} -HOST_FLAGS := ${COMMON_FLAGS} -std=c11 -O3 `dpu-pkg-config --cflags --libs dpu` -DNR_TASKLETS=${NR_TASKLETS} -DNR_DPUS=${NR_DPUS} -DBL=${BL} -DENERGY=${ENERGY} +HOST_FLAGS := ${COMMON_FLAGS} -std=c11 -O3 `dpu-pkg-config --cflags --libs dpu` -DNR_TASKLETS=${NR_TASKLETS} -DNR_DPUS=${NR_DPUS} -DBL=${BL} -DENERGY=${ENERGY} -DWITH_ALLOC_OVERHEAD=${WITH_ALLOC_OVERHEAD} -DWITH_LOAD_OVERHEAD=${WITH_LOAD_OVERHEAD} -DWITH_FREE_OVERHEAD=${WITH_FREE_OVERHEAD} DPU_FLAGS := ${COMMON_FLAGS} -O2 -DNR_TASKLETS=${NR_TASKLETS} -DBL=${BL} QUIET = @ diff --git a/SEL/host/app.c b/SEL/host/app.c index 90f8bd4..7946979 100644 --- a/SEL/host/app.c +++ b/SEL/host/app.c @@ -66,55 +66,88 @@ int main(int argc, char **argv) { struct dpu_set_t dpu_set, dpu; uint32_t nr_of_dpus; - -#if ENERGY - struct dpu_probe_t probe; - DPU_ASSERT(dpu_probe_init("energy_probe", &probe)); -#endif + uint32_t nr_of_ranks; + + // Timer declaration + Timer timer; // Allocate DPUs and load binary +#if !WITH_ALLOC_OVERHEAD DPU_ASSERT(dpu_alloc(NR_DPUS, NULL, &dpu_set)); + timer.time[0] = 0; // alloc +#endif +#if !WITH_LOAD_OVERHEAD DPU_ASSERT(dpu_load(dpu_set, DPU_BINARY, NULL)); DPU_ASSERT(dpu_get_nr_dpus(dpu_set, &nr_of_dpus)); - printf("Allocated %d DPU(s)\n", nr_of_dpus); + DPU_ASSERT(dpu_get_nr_ranks(dpu_set, &nr_of_ranks)); + assert(nr_of_dpus == NR_DPUS); + timer.time[1] = 0; // load +#endif +#if !WITH_FREE_OVERHEAD + timer.time[6] = 0; // free +#endif + +#if ENERGY + struct dpu_probe_t probe; + DPU_ASSERT(dpu_probe_init("energy_probe", &probe)); +#endif unsigned int i = 0; uint32_t accum = 0; uint32_t total_count = 0; - const unsigned int input_size = p.exp == 0 ? p.input_size * nr_of_dpus : p.input_size; // Total input size (weak or strong scaling) - const unsigned int input_size_dpu_ = divceil(input_size, nr_of_dpus); // Input size per DPU (max.) + const unsigned int input_size = p.exp == 0 ? p.input_size * NR_DPUS : p.input_size; // Total input size (weak or strong scaling) + const unsigned int input_size_dpu_ = divceil(input_size, NR_DPUS); // Input size per DPU (max.) const unsigned int input_size_dpu_round = (input_size_dpu_ % (NR_TASKLETS * REGS) != 0) ? roundup(input_size_dpu_, (NR_TASKLETS * REGS)) : input_size_dpu_; // Input size per DPU (max.), 8-byte aligned // Input/output allocation - A = malloc(input_size_dpu_round * nr_of_dpus * sizeof(T)); - C = malloc(input_size_dpu_round * nr_of_dpus * sizeof(T)); - C2 = malloc(input_size_dpu_round * nr_of_dpus * sizeof(T)); + A = malloc(input_size_dpu_round * NR_DPUS * sizeof(T)); + C = malloc(input_size_dpu_round * NR_DPUS * sizeof(T)); + C2 = malloc(input_size_dpu_round * NR_DPUS * sizeof(T)); T *bufferA = A; T *bufferC = C2; // Create an input file with arbitrary data - read_input(A, input_size, input_size_dpu_round * nr_of_dpus); - - // Timer declaration - Timer timer; + read_input(A, input_size, input_size_dpu_round * NR_DPUS); printf("NR_TASKLETS\t%d\tBL\t%d\n", NR_TASKLETS, BL); // Loop over main kernel for(int rep = 0; rep < p.n_warmup + p.n_reps; rep++) { +#if WITH_ALLOC_OVERHEAD + if(rep >= p.n_warmup) { + start(&timer, 0, 0); + } + DPU_ASSERT(dpu_alloc(NR_DPUS, NULL, &dpu_set)); + if(rep >= p.n_warmup) { + stop(&timer, 0); + } +#endif +#if WITH_LOAD_OVERHEAD + if(rep >= p.n_warmup) { + start(&timer, 1, 0); + } + DPU_ASSERT(dpu_load(dpu_set, DPU_BINARY, NULL)); + if(rep >= p.n_warmup) { + stop(&timer, 1); + } + DPU_ASSERT(dpu_get_nr_dpus(dpu_set, &nr_of_dpus)); + DPU_ASSERT(dpu_get_nr_ranks(dpu_set, &nr_of_ranks)); + assert(nr_of_dpus == NR_DPUS); +#endif + // Compute output on CPU (performance comparison and verification purposes) if(rep >= p.n_warmup) - start(&timer, 0, 0); + start(&timer, 2, 0); total_count = select_host(C, A, input_size); if(rep >= p.n_warmup) - stop(&timer, 0); + stop(&timer, 2); printf("Load input data\n"); if(rep >= p.n_warmup) - start(&timer, 1, 0); + start(&timer, 3, 0); // Input arguments const unsigned int input_size_dpu = input_size_dpu_round; unsigned int kernel = 0; @@ -130,19 +163,19 @@ int main(int argc, char **argv) { } DPU_ASSERT(dpu_push_xfer(dpu_set, DPU_XFER_TO_DPU, DPU_MRAM_HEAP_POINTER_NAME, 0, input_size_dpu * sizeof(T), DPU_XFER_DEFAULT)); if(rep >= p.n_warmup) - stop(&timer, 1); + stop(&timer, 3); printf("Run program on DPU(s) \n"); // Run DPU kernel if(rep >= p.n_warmup) { - start(&timer, 2, 0); + start(&timer, 4, 0); #if ENERGY DPU_ASSERT(dpu_probe_start(&probe)); #endif } DPU_ASSERT(dpu_launch(dpu_set, DPU_SYNCHRONOUS)); if(rep >= p.n_warmup) { - stop(&timer, 2); + stop(&timer, 4); #if ENERGY DPU_ASSERT(dpu_probe_stop(&probe)); #endif @@ -161,15 +194,15 @@ int main(int argc, char **argv) { #endif printf("Retrieve results\n"); - dpu_results_t results[nr_of_dpus]; - uint32_t* results_scan = malloc(nr_of_dpus * sizeof(uint32_t)); + dpu_results_t results[NR_DPUS]; + uint32_t* results_scan = malloc(NR_DPUS * sizeof(uint32_t)); i = 0; accum = 0; if(rep >= p.n_warmup) - start(&timer, 3, 0); + start(&timer, 5, 0); // PARALLEL RETRIEVE TRANSFER - dpu_results_t* results_retrieve[nr_of_dpus]; + dpu_results_t* results_retrieve[NR_DPUS]; DPU_FOREACH(dpu_set, dpu, i) { results_retrieve[i] = (dpu_results_t*)malloc(NR_TASKLETS * sizeof(dpu_results_t)); @@ -195,11 +228,11 @@ int main(int argc, char **argv) { free(results_retrieve[i]); } if(rep >= p.n_warmup) - stop(&timer, 3); + stop(&timer, 5); i = 0; if(rep >= p.n_warmup) - start(&timer, 4, 0); + start(&timer, 6, 0); DPU_FOREACH (dpu_set, dpu) { // Copy output array DPU_ASSERT(dpu_copy_from(dpu, DPU_MRAM_HEAP_POINTER_NAME, input_size_dpu * sizeof(T), bufferC + results_scan[i], results[i].t_count * sizeof(T))); @@ -207,7 +240,21 @@ int main(int argc, char **argv) { i++; } if(rep >= p.n_warmup) - stop(&timer, 4); + stop(&timer, 6); + +#if WITH_ALLOC_OVERHEAD +#if WITH_FREE_OVERHEAD + if(rep >= p.n_warmup) { + start(&timer, 7, 0); + } +#endif + DPU_ASSERT(dpu_free(dpu_set)); +#if WITH_FREE_OVERHEAD + if(rep >= p.n_warmup) { + stop(&timer, 7); + } +#endif +#endif // Free memory free(results_scan); @@ -226,35 +273,40 @@ int main(int argc, char **argv) { if (status) { printf("[" ANSI_COLOR_GREEN "OK" ANSI_COLOR_RESET "] Outputs are equal\n"); if (rep >= p.n_warmup) { - printf("[::] SEL NMC | n_dpus=%d n_tasklets=%d e_type=%s block_size_B=%d n_elements=%d " - "| throughput_cpu_MBps=%f throughput_pim_MBps=%f throughput_MBps=%f", - nr_of_dpus, NR_TASKLETS, XSTR(T), BLOCK_SIZE, input_size, - input_size * sizeof(T) / timer.time[0], - input_size * sizeof(T) / (timer.time[2] + timer.time[3]), - input_size * sizeof(T) / (timer.time[1] + timer.time[2] + timer.time[3] + timer.time[4])); - printf(" throughput_cpu_MOpps=%f throughput_pim_MOpps=%f throughput_MOpps=%f", - input_size / timer.time[0], - input_size / (timer.time[2] + timer.time[3]), - input_size / (timer.time[1] + timer.time[2] + timer.time[3] + timer.time[4])); - printall(&timer, 4); + printf("[::] SEL UPMEM | n_dpus=%d n_ranks=%d n_tasklets=%d e_type=%s block_size_B=%d n_elements=%d", + NR_DPUS, nr_of_ranks, NR_TASKLETS, XSTR(T), BLOCK_SIZE, input_size); + printf(" b_with_alloc_overhead=%d b_with_load_overhead=%d b_with_free_overhead=%d ", + WITH_ALLOC_OVERHEAD, WITH_LOAD_OVERHEAD, WITH_FREE_OVERHEAD); + printf("| latency_alloc_us=%f latency_load_us=%f latency_cpu_us=%f latency_write_us=%f latency_kernel_us=%f latency_read_us=%f latency_free_us=%f", + timer.time[0], + timer.time[1], + timer.time[2], + timer.time[3], // write + timer.time[4], // kernel + timer.time[5] + timer.time[6], // read + timer.time[7]); + printf(" throughput_cpu_MBps=%f throughput_upmem_kernel_MBps=%f throughput_upmem_total_MBps=%f", + input_size * sizeof(T) / timer.time[2], + input_size * sizeof(T) / (timer.time[4]), + input_size * sizeof(T) / (timer.time[0] + timer.time[1] + timer.time[3] + timer.time[4] + timer.time[5] + timer.time[6] + timer.time[7])); + printf(" throughput_upmem_wxr_MBps=%f throughput_upmem_lwxr_MBps=%f throughput_upmem_alwxr_MBps=%f", + input_size * sizeof(T) / (timer.time[3] + timer.time[4] + timer.time[5] + timer.time[6]), + input_size * sizeof(T) / (timer.time[1] + timer.time[3] + timer.time[4] + timer.time[5] + timer.time[6]), + input_size * sizeof(T) / (timer.time[0] + timer.time[1] + timer.time[3] + timer.time[4] + timer.time[5] + timer.time[6])); + printf(" throughput_cpu_MOpps=%f throughput_upmem_kernel_MOpps=%f throughput_upmem_total_MOpps=%f", + input_size / timer.time[2], + input_size / (timer.time[4] + timer.time[5] + timer.time[6]), + input_size / (timer.time[0] + timer.time[1] + timer.time[3] + timer.time[4] + timer.time[5] + timer.time[6] + timer.time[7] + timer.time[8])); + printf(" throughput_upmem_wxr_MOpps=%f throughput_upmem_lwxr_MOpps=%f throughput_upmem_alwxr_MOpps=%f\n", + input_size / (timer.time[3] + timer.time[4] + timer.time[5] + timer.time[6]), + input_size / (timer.time[1] + timer.time[3] + timer.time[4] + timer.time[5] + timer.time[6]), + input_size / (timer.time[0] + timer.time[1] + timer.time[3] + timer.time[4] + timer.time[5] + timer.time[6])); } } else { printf("[" ANSI_COLOR_RED "ERROR" ANSI_COLOR_RESET "] Outputs differ!\n"); } } - // Print timing results - printf("CPU "); - print(&timer, 0, p.n_reps); - printf("CPU-DPU "); - print(&timer, 1, p.n_reps); - printf("DPU Kernel "); - print(&timer, 2, p.n_reps); - printf("Inter-DPU "); - print(&timer, 3, p.n_reps); - printf("DPU-CPU "); - print(&timer, 4, p.n_reps); - #if ENERGY double energy; DPU_ASSERT(dpu_probe_get(&probe, DPU_ENERGY, DPU_AVERAGE, &energy)); @@ -265,7 +317,10 @@ int main(int argc, char **argv) { free(A); free(C); free(C2); + +#if !WITH_ALLOC_OVERHEAD DPU_ASSERT(dpu_free(dpu_set)); - +#endif + return 0; } diff --git a/SEL/run-fgbs24a.sh b/SEL/run-fgbs24a.sh index 079ec3f..cb9ef43 100755 --- a/SEL/run-fgbs24a.sh +++ b/SEL/run-fgbs24a.sh @@ -4,6 +4,8 @@ set -e mkdir -p $(hostname) +ts=$(date +%Y%m%d) + # BL: use 2^(BL) B blocks for MRAM <-> WRAM transfers on PIM module # T: data type # -w: number of un-timed warmup iterations @@ -16,14 +18,16 @@ echo "prim-benchmarks SEL (dfatool fgbs24a edition)" echo "Started at $(date)" echo "Revision $(git describe --always)" -# >2048 is not in upstream -for nr_dpus in 2543 2304 2048; do +for nr_dpus in 2304 2048 2543; do for nr_tasklets in 16; do echo if make -B NR_DPUS=${nr_dpus} NR_TASKLETS=${nr_tasklets} BL=10; then timeout --foreground -k 1m 30m bin/host_code -w 0 -e 100 -i 251658240 -x 1 || true fi + if make -B NR_DPUS=${nr_dpus} NR_TASKLETS=${nr_tasklets} BL=10 WITH_ALLOC_OVERHEAD=1 WITH_LOAD_OVERHEAD=1 WITH_FREE_OVERHEAD=1; then + timeout --foreground -k 1m 30m bin/host_code -w 0 -e 100 -i 251658240 -x 1 || true + fi done done echo "Completed at $(date)" -) | tee "$(hostname)/fgbs24a.txt" +) | tee "$(hostname)/${ts}-fgbs24a.txt" diff --git a/SEL/support/timer.h b/SEL/support/timer.h index 4d597b9..3e826f0 100755 --- a/SEL/support/timer.h +++ b/SEL/support/timer.h @@ -37,9 +37,9 @@ typedef struct Timer{ - struct timeval startTime[7]; - struct timeval stopTime[7]; - double time[7]; + struct timeval startTime[8]; + struct timeval stopTime[8]; + double time[8]; }Timer; -- cgit v1.2.3