diff options
| -rw-r--r-- | VA/Makefile | 3 | ||||
| -rw-r--r-- | VA/host/app.c | 14 |
2 files changed, 5 insertions, 12 deletions
diff --git a/VA/Makefile b/VA/Makefile index 0994df9..040dd4a 100644 --- a/VA/Makefile +++ b/VA/Makefile @@ -7,14 +7,13 @@ WITH_ALLOC_OVERHEAD ?= 0 WITH_LOAD_OVERHEAD ?= 0 WITH_FREE_OVERHEAD ?= 0 WITH_DPUINFO ?= 0 -SDK_SINGLETHREADED ?= 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} -D${TYPE} -DENERGY=${ENERGY} -DWITH_ALLOC_OVERHEAD=${WITH_ALLOC_OVERHEAD} -DWITH_LOAD_OVERHEAD=${WITH_LOAD_OVERHEAD} -DWITH_FREE_OVERHEAD=${WITH_FREE_OVERHEAD} -DWITH_DPUINFO=${WITH_DPUINFO} -DSDK_SINGLETHREADED=${SDK_SINGLETHREADED} +HOST_FLAGS := ${COMMON_FLAGS} -std=c11 -O3 `dpu-pkg-config --cflags --libs dpu` -DNR_TASKLETS=${NR_TASKLETS} -DNR_DPUS=${NR_DPUS} -DBL=${BL} -D${TYPE} -DENERGY=${ENERGY} -DWITH_ALLOC_OVERHEAD=${WITH_ALLOC_OVERHEAD} -DWITH_LOAD_OVERHEAD=${WITH_LOAD_OVERHEAD} -DWITH_FREE_OVERHEAD=${WITH_FREE_OVERHEAD} -DWITH_DPUINFO=${WITH_DPUINFO} DPU_FLAGS := ${COMMON_FLAGS} -O2 -DNR_TASKLETS=${NR_TASKLETS} -DBL=${BL} -D${TYPE} QUIET = @ diff --git a/VA/host/app.c b/VA/host/app.c index 93d6e57..b339bf3 100644 --- a/VA/host/app.c +++ b/VA/host/app.c @@ -34,12 +34,6 @@ #include <dpu_target_macros.h> #endif -#if SDK_SINGLETHREADED -#define DPU_ALLOC_PROFILE "nrThreadsPerRank=0" -#else -#define DPU_ALLOC_PROFILE NULL -#endif - // Pointer declaration static T* A; static T* B; @@ -81,7 +75,7 @@ int main(int argc, char **argv) { // Allocate DPUs and load binary #if !WITH_ALLOC_OVERHEAD - DPU_ASSERT(dpu_alloc(NR_DPUS, DPU_ALLOC_PROFILE, &dpu_set)); + DPU_ASSERT(dpu_alloc(NR_DPUS, NULL, &dpu_set)); timer.time[0] = 0; // alloc #endif #if !WITH_LOAD_OVERHEAD @@ -122,7 +116,7 @@ int main(int argc, char **argv) { if(rep >= p.n_warmup) { start(&timer, 0, 0); } - DPU_ASSERT(dpu_alloc(NR_DPUS, DPU_ALLOC_PROFILE, &dpu_set)); + DPU_ASSERT(dpu_alloc(NR_DPUS, NULL, &dpu_set)); if(rep >= p.n_warmup) { stop(&timer, 0); } @@ -263,8 +257,8 @@ int main(int argc, char **argv) { if (rep >= p.n_warmup) { printf("[::] VA UPMEM | n_dpus=%d n_ranks=%d n_tasklets=%d e_type=%s block_size_B=%d n_elements=%d n_elements_per_dpu=%d", nr_of_dpus, nr_of_ranks, NR_TASKLETS, XSTR(T), BLOCK_SIZE, input_size, input_size / NR_DPUS); - printf(" b_sdk_singlethreaded=%d b_with_alloc_overhead=%d b_with_load_overhead=%d b_with_free_overhead=%d ", - SDK_SINGLETHREADED, WITH_ALLOC_OVERHEAD, WITH_LOAD_OVERHEAD, WITH_FREE_OVERHEAD); + 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], |