-
Notifications
You must be signed in to change notification settings - Fork 1
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
follow these tests: #5
- Loading branch information
Showing
1 changed file
with
220 additions
and
0 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,220 @@ | ||
| .POSIX: | ||
| .SUFFIXES: | ||
|
|
||
|
|
||
|
|
||
| # mycommand="./mean29-generic -h R4nNmL0Xjir09QswGFtqc7wwyuk10OfI23OlwznvsAw=yYJP/pjF9b8= -n 137 -t 483 -r -1" | ||
| # perf record $mycommand 1 | ||
| # sudo perf report | ||
| # sudo mpiexec -demux select -f /nfs/hosts -n 3 perf record --no-buffering --count=10 --freq=2249 --quiet --no-samples --no-buildid --no-buildid-cache -e cycles ./mean29-generic -h PfNTPRoCf+EOLNMQSEu7IP0oHZbz2MmN74kj7pMMQ5c=J/R5YAGZZzk= -n 0 -t 483 -r -1 | ||
|
|
||
| # -r --exclude-perf --dry-run | ||
| # perf record --dry-run -e can act as a BPF script compiler if llvm.dump-obj in config file is set to true. | ||
| # perf mem -e cycles ./mean29-generic -h PfNTPRoCf+EOLNMQSEu7IP0oHZbz2MmN74kj7pMMQ5c=J/R5YAGZZzk= -n 0 -t 483 -r -1 | ||
| OPT ?= -Os -fprefetch-loop-arrays -freorder-blocks-algorithm=stc -ffast-math \ | ||
| -fbranch-count-reg \ | ||
| -fif-conversion -fif-conversion2 \ | ||
| -finline-functions-called-once \ | ||
| -fmove-loop-invariants -fssa-phiopt \ | ||
| -ftree-bit-ccp -ftree-pta -ftree-sra \ | ||
| -fgcse-las -faggressive-loop-optimizations -fdeclone-ctor-dtor -fdelete-null-pointer-checks -fno-lifetime-dse \ | ||
| -flive-range-shrinkage -fira-algorithm=priority -fira-region=all -fira-hoist-pressure -fno-sched-interblock -fno-sched-spec -fsched2-use-superblocks \ | ||
| -freschedule-modulo-scheduled-loops -fsel-sched-pipelining -fno-semantic-interposition -fipa-pta -floop-nest-optimize \ | ||
| -floop-nest-optimize -floop-parallelize-all -ftree-loop-distribution -ftree-loop-im -funswitch-loops -ftree-loop-ivcanon -fivopts \ | ||
| -ftree-parallelize-loops=80 -fsimd-cost-model=cheap -fsplit-ivs-in-unroller -fvariable-expansion-in-unroller -fno-printf-return-value \ | ||
| -Wformat-overflow -Wformat-truncation -fstrict-aliasing | ||
|
|
||
| # -flto-partition=none -flto-compression-level=9 | ||
|
|
||
| #-fprofile-use -fprofile-use=perf.data | ||
|
|
||
| # -falign-functions -falign-jumps -falign-labels -falign-loops | ||
| #-fsched-pressure | ||
|
|
||
| # -fdevirtualize-speculatively | ||
|
|
||
| # -fno-ira-share-spill-slots -fno-ira-share-save-slots | ||
|
|
||
| # -fira-loop-pressure | ||
|
|
||
| # -flifetime-dse=2 | ||
|
|
||
| #-fdevirtualize-at-ltrans # slower | ||
|
|
||
| #-fstack-protector-explicit | ||
|
|
||
| GCC_ARCH_FLAGS ?= -march=native | ||
| GPP_ARCH_FLAGS ?= -march=native | ||
|
|
||
| # -Wno-deprecated-declarations shuts up Apple OSX clang | ||
| FLAGS ?= -Wall -Wno-format -Wno-deprecated-declarations -D_POSIX_C_SOURCE=200112L $(OPT) -DPREFETCH -I. $(CPPFLAGS) -pthread | ||
|
|
||
| #GPP ?= mpic++ $(GPP_ARCH_FLAGS) -std=c++11 $(FLAGS) | ||
|
|
||
| #GPP ?= gcc $(GPP_ARCH_FLAGS) $(FLAGS) | ||
|
|
||
| GPP ?= g++ $(GPP_ARCH_FLAGS) -std=c++11 $(FLAGS) | ||
|
|
||
| CFLAGS ?= -Wall -Wno-format -fomit-frame-pointer $(OPT) | ||
| GCC ?= gcc $(GCC_ARCH_FLAGS) -std=gnu11 $(CFLAGS) | ||
| BLAKE_2B_SRC ?= ../crypto/blake2b-ref.c | ||
| LIBS = -L. -lblake2b | ||
|
|
||
| # 2 to 5 Problem | ||
| # time ./cuda29_generic_64 -h 324D6F373754576B3546684B772F493865493343416B71497531492B624A4B6155567A6543424B3278414D3D -n 18335313330172576564 -r 120 -g | ||
|
|
||
| #cuda specific | ||
| #all: cuda29_generic_64 | ||
| all: cuda29_generic_64 cuda29_generic_test cuda29_generic_orig cuda29-generic cuda29_GTX_970 cuda29_GTX_980 cuda29_GTX_1030 cuda29_GTX_1050 cuda29_GTX_1060 cuda29_GTX_1070 cuda29_GTX_1080 cuda29_GTX_2080 cuda29_GTX_titan | ||
|
|
||
| # strip -s cuda29_generic_64 && strip -s cuda29_generic_test && strip -s cuda29_generic_orig && strip -s cuda29-generic && strip -s cuda29_GTX_970 && strip -s cuda29_GTX_980 && strip -s cuda29_GTX_1030 && strip -s cuda29_GTX_1050 && strip -s cuda29_GTX_1060 && strip -s cuda29_GTX_1070 && strip -s cuda29_GTX_1080 && strip -s cuda29_GTX_2080 && strip -s cuda29_GTX_titan | ||
|
|
||
| #32_bit apps only works on 32 bit systems | ||
| #all : cuda29-mean_sm70_32 cuda29-mean_sm60_32 cuda29-mean_sm50_32 cuda29-mean_sm35_32 | ||
|
|
||
| #cuda29 lcuda29 mean29-generic | ||
| #all : simpletest leantest mean29-generic mean29x1_test verify30 mean29-generic lean29-generic verify30 | ||
|
|
||
| simpletest: simple19 | ||
| ./simple19 -n 38 | ||
|
|
||
| leantest: lean19 | ||
| ./lean19 -n 38 | ||
|
|
||
| mean29x1_test: mean29x1_p1 | ||
|
|
||
| verify30: ../crypto/siphash.h cuckoo.h cuckoo.c Makefile | ||
| $(GCC) -o $@ -DEDGEBITS=29 cuckoo.c $(BLAKE_2B_SRC) | ||
|
|
||
| simple19: ../crypto/siphash.h cuckoo.h cyclebase.hpp simple.cpp Makefile | ||
| $(GPP) -o $@ -DIDXSHIFT=0 -DPROOFSIZE=42 -DEDGEBITS=19 simple.cpp $(BLAKE_2B_SRC) | ||
|
|
||
| mean29-generic: cuckoo.h ../crypto/siphash.h mean.hpp mean.cpp Makefile | ||
| $(GPP) -o $@ -mno-avx2 -DSAVEEDGES -DNSIPHASH=1 -DEDGEBITS=29 mean.cpp $(BLAKE_2B_SRC) | ||
|
|
||
| lean29-generic: cuckoo.h ../crypto/siphash.h lean.hpp lean.cpp Makefile | ||
| $(GPP) -o $@ -mno-avx2 -DATOMIC -DEDGEBITS=29 lean.cpp $(BLAKE_2B_SRC) | ||
|
|
||
| simple29: ../crypto/siphash.h cuckoo.h cyclebase.hpp simple.cpp Makefile | ||
| $(GPP) -o $@ -DIDXSHIFT=0 -DPROOFSIZE=42 -DEDGEBITS=29 simple.cpp $(BLAKE_2B_SRC) | ||
|
|
||
| lean19: ../crypto/siphash.h cuckoo.h lean.cpp Makefile | ||
| $(GPP) -o $@ -DATOMIC -DEDGEBITS=19 lean.cpp $(BLAKE_2B_SRC) | ||
|
|
||
| lean29x8: ../crypto/siphash.h cuckoo.h lean.cpp Makefile | ||
| $(GPP) -o $@ -mavx2 -DNSIPHASH=8 -DATOMIC -DEDGEBITS=29 lean.cpp $(BLAKE_2B_SRC) | ||
|
|
||
| mean19x8: cuckoo.h ../crypto/siphash.h mean.hpp mean.cpp Makefile | ||
| $(GPP) -o $@ -mavx2 -DXBITS=2 -DNSIPHASH=8 -DEDGEBITS=19 mean.cpp $(BLAKE_2B_SRC) | ||
|
|
||
| mean29x4: cuckoo.h ../crypto/siphash.h mean.hpp mean.cpp Makefile | ||
| $(GPP) -o $@ -mno-avx2 -DNSIPHASH=4 -DEDGEBITS=29 mean.cpp $(BLAKE_2B_SRC) | ||
|
|
||
| mean29x8: cuckoo.h ../crypto/siphash.h mean.hpp mean.cpp Makefile | ||
| $(GPP) -o $@ -mavx2 -DNSIPHASH=8 -DEDGEBITS=29 mean.cpp $(BLAKE_2B_SRC) | ||
|
|
||
| mean29x1: cuckoo.h ../crypto/siphash.h mean.hpp mean.cpp Makefile | ||
| $(GPP) -o $@ -DNSIPHASH=1 -DEDGEBITS=29 mean.cpp $(BLAKE_2B_SRC) | ||
|
|
||
| mean29x1_p1: cuckoo.h ../crypto/siphash.h mean.hpp mean.cpp Makefile | ||
| $(GPP) -o $@ -DNSIPHASH=1 -DPROOFSIZE=1 -DEDGEBITS=29 mean.cpp $(BLAKE_2B_SRC) | ||
|
|
||
| lcuda29: ../crypto/siphash.cuh lean.cu Makefile | ||
| nvcc -o $@ -DEDGEBITS=29 -arch sm_35 lean.cu $(BLAKE_2B_SRC) | ||
|
|
||
| cuda29_orig: ../crypto/siphash.cuh mean.cu Makefile | ||
| nvcc -o $@ -DEDGEBITS=29 -arch sm_35 mean.cu $(BLAKE_2B_SRC) | ||
|
|
||
|
|
||
| #64_bit apps | ||
| #all : cuda29-mean_sm75 cuda29-mean_sm72 cuda29-mean_sm70 cuda29-mean_sm60 cuda29-mean_sm50 cuda29-mean_sm35 | ||
|
|
||
| # 64 Bit | ||
| # -arch sm_75 | ||
| cuda29-mean_sm75: ../crypto/siphash.cuh mean.cu Makefile | ||
| nvcc mean.cu $(BLAKE_2B_SRC) -gencode arch=compute_75,code=sm_75 --verbose -m64 -DEDGEBITS=29 -o $@ | ||
|
|
||
| # -arch sm_72 | ||
| cuda29-mean_sm72: ../crypto/siphash.cuh mean.cu Makefile | ||
| nvcc mean.cu $(BLAKE_2B_SRC) -gencode arch=compute_72,code=sm_72 --verbose -m64 -DEDGEBITS=29 -o $@ | ||
|
|
||
| # -arch sm_70 | ||
| cuda29-mean_sm70: ../crypto/siphash.cuh mean.cu Makefile | ||
| nvcc mean.cu $(BLAKE_2B_SRC) -gencode arch=compute_70,code=sm_70 --verbose -m64 -DEDGEBITS=29 -o $@ | ||
|
|
||
| # -arch sm_60 | ||
| cuda29-mean_sm60: ../crypto/siphash.cuh mean.cu Makefile | ||
| nvcc mean.cu $(BLAKE_2B_SRC) -gencode arch=compute_60,code=sm_60 --verbose -m64 -DEDGEBITS=29 -o $@ | ||
|
|
||
| # -arch sm_50 | ||
| cuda29-mean_sm50: ../crypto/siphash.cuh mean.cu Makefile | ||
| nvcc mean.cu $(BLAKE_2B_SRC) -gencode arch=compute_50,code=sm_50 --verbose -m64 -DEDGEBITS=29 -o $@ | ||
|
|
||
| # -arch sm_35 | ||
| cuda29-mean_sm35: ../crypto/siphash.cuh mean.cu Makefile | ||
| nvcc -std=c++11 --verbose -m64 --use_fast_math -arch sm_35 mean.cu $(BLAKE_2B_SRC) -DEDGEBITS=29 -o $@ | ||
|
|
||
|
|
||
| BUILD_INFO= -D__BUILD_VERSION=\"$@\" -D__BUILD_DATE=\"$$(date +'%Y%m%d-%H:%M:%S')\" -D__BUILD_BY=\"zwilla\" -D__BUILD_SHA=\"e99ea684f258c053f2bda842ddef176696ed0d87e22e9ee2e17469e76f94a231\" | ||
|
|
||
|
|
||
| #Driver specific | ||
| # -arch sm_35 | ||
| cuda29_generic: ../crypto/siphash.cuh mean.cu Makefile | ||
| nvcc mean.cu $(BLAKE_2B_SRC) -arch sm_35 --verbose -m64 -DEDGEBITS=29 -o $@ $(BUILD_INFO) | ||
|
|
||
| cuda29_generic_orig: ../crypto/siphash.cuh mean_orig.cu Makefile | ||
| nvcc mean_orig.cu $(BLAKE_2B_SRC) -arch sm_35 --verbose -m64 -DEDGEBITS=29 -o $@ $(BUILD_INFO) | ||
|
|
||
| cuda29_generic_test: ../crypto/siphash.cuh mean.cu Makefile | ||
| nvcc mean.cu $(BLAKE_2B_SRC) -arch sm_61 --verbose -m64 -DEDGEBITS=29 -o $@ $(BUILD_INFO) | ||
|
|
||
| cuda29_generic_64: ../crypto/siphash.cuh mean64.cu Makefile | ||
| nvcc mean64.cu $(BLAKE_2B_SRC) -arch sm_61 --verbose -m64 -DEDGEBITS=29 -o $@ $(BUILD_INFO) | ||
|
|
||
| cuda29_GTX_970: ../crypto/siphash.cuh mean.cu Makefile | ||
| nvcc mean.cu $(BLAKE_2B_SRC) -arch sm_61 --verbose -m64 -DEDGEBITS=29 -o $@ $(BUILD_INFO) | ||
|
|
||
| cuda29_GTX_980: ../crypto/siphash.cuh mean.cu Makefile | ||
| nvcc mean.cu $(BLAKE_2B_SRC) -arch sm_61 --verbose -m64 -DEDGEBITS=29 -o $@ $(BUILD_INFO) | ||
|
|
||
| cuda29_GTX_1030: ../crypto/siphash.cuh mean.cu Makefile | ||
| nvcc mean.cu $(BLAKE_2B_SRC) -arch sm_61 --verbose -m64 -DEDGEBITS=29 -o $@ $(BUILD_INFO) | ||
|
|
||
| cuda29_GTX_1050: ../crypto/siphash.cuh mean.cu Makefile | ||
| nvcc mean.cu $(BLAKE_2B_SRC) -arch sm_61 --verbose -m64 -DEDGEBITS=29 -o $@ $(BUILD_INFO) | ||
|
|
||
| cuda29_GTX_1060: ../crypto/siphash.cuh mean.cu Makefile | ||
| nvcc mean.cu $(BLAKE_2B_SRC) -arch sm_61 --verbose -m64 -DEDGEBITS=29 -o $@ $(BUILD_INFO) | ||
|
|
||
| cuda29_GTX_1070: ../crypto/siphash.cuh mean.cu Makefile | ||
| nvcc mean.cu $(BLAKE_2B_SRC) -arch sm_61 --verbose -m64 -DEDGEBITS=29 -o $@ $(BUILD_INFO) | ||
|
|
||
| cuda29_GTX_1080: ../crypto/siphash.cuh mean.cu Makefile | ||
| nvcc mean.cu $(BLAKE_2B_SRC) -arch sm_61 --verbose -m64 -DEDGEBITS=29 -o $@ $(BUILD_INFO) | ||
|
|
||
| cuda29_GTX_2080: ../crypto/siphash.cuh mean.cu Makefile | ||
| nvcc mean.cu $(BLAKE_2B_SRC) -arch sm_61 --verbose -m64 -DEDGEBITS=29 -o $@ $(BUILD_INFO) | ||
|
|
||
| cuda29_GTX_titan: ../crypto/siphash.cuh mean.cu Makefile | ||
| nvcc mean.cu $(BLAKE_2B_SRC) -arch sm_61 --verbose -m64 -DEDGEBITS=29 -o $@ $(BUILD_INFO) | ||
|
|
||
|
|
||
| #32 Bit | ||
| # -arch sm_70 | ||
| cuda29-mean_sm70_32: ../crypto/siphash.cuh mean.cu Makefile | ||
| nvcc mean.cu $(BLAKE_2B_SRC) -gencode arch=compute_70,code=sm_70 --verbose -m32 -DEDGEBITS=29 -o $@ | ||
|
|
||
| # -arch sm_60 | ||
| cuda29-mean_sm60_32: ../crypto/siphash.cuh mean.cu Makefile | ||
| nvcc mean.cu $(BLAKE_2B_SRC) -gencode arch=compute_60,code=sm_60 --verbose -m32 -DEDGEBITS=29 -o $@ | ||
|
|
||
| # -arch sm_50 | ||
| cuda29-mean_sm50_32: ../crypto/siphash.cuh mean.cu Makefile | ||
| nvcc mean.cu $(BLAKE_2B_SRC) -gencode arch=compute_50,code=sm_50 --verbose -m32 -DEDGEBITS=29 -o $@ | ||
|
|
||
| # -arch sm_35 | ||
| cuda29-mean_sm35_32: ../crypto/siphash.cuh mean.cu Makefile | ||
| nvcc -std=c++11 --verbose -m32 --use_fast_math -arch sm_35 mean.cu $(BLAKE_2B_SRC) -DEDGEBITS=29 -o $@ | ||
|
|
||
| #'sm_20', 'sm_21', 'sm_30', 'sm_32', 'sm_35', 'sm_37', 'sm_50', 'sm_52', 'sm_53' 'sm_60', 'sm_61', 'sm_62', 'sm_70', 'sm_72', 'sm_75'. |