update benchmark results

salykova · Nov 16, 2024 · 6fa5c50 · 6fa5c50
1 parent 4d7b9a7
commit 6fa5c50
Show file tree

Hide file tree

Showing 5 changed files with 126 additions and 120 deletions.
diff --git a/README.md b/README.md
@@ -1,15 +1,15 @@
-# Fast, Multi-threaded Matrix Multiplication in C from Scratch
+# High-Performance FP32 Matrix Multiplication on CPU
 
 > **Important note:** Please don’t expect peak performance without fine-tuning hyperparameters such as the *number of threads, kernel size and block sizes*, unless you're running it on a Ryzen 7700(X). The current implementation includes a single kernel and a parallelization strategy, both optimized for AMD Zen CPUs. For manycore processors (> 16 cores), consider utilizing nested parallelism and parallelizing 2-3 loops to increase the performance (e.g., the 5th, 3rd, and 2nd loops around the kernel). More on this in the [tutorial](https://salykova.github.io/matmul-cpu).
 
 ## Key Features
-- Step by step, beginner-friendly [tutorial](https://salykova.github.io/matmul-cpu)
+- Performance comparable to OpenBLAS and MKL
 - Simple and scalable C code
 - Supports arbitrary matrix sizes
-- Faster than OpenBLAS and MKL on Ryzen 7700
-- Efficiently parallelized with just 3 lines of OpenMP directives
-- Targets x86 processors with AVX2 and FMA3 instructions (=all modern Intel Core and AMD Ryzen CPUs)
+- Efficiently parallelized with 3 lines of OpenMP directives
+- Targets x86 processors with AVX2 and FMA3 instructions
 - Follows the [BLIS](https://github.com/flame/blis) design
+- Step by step, beginner-friendly [tutorial](https://salykova.github.io/matmul-cpu)
 
 ## Installation
 Install the following packages via `apt` if you are using a Debian-based Linux distribution
@@ -40,7 +40,7 @@ Test enviroment:
 - CPU LOCKED CLOCK SPEED: 4.5GHz
 - RAM: 32GB DDR5 6000 MHz CL36
 - OpenBLAS v.0.3.26
-- MKL v2023.1
+- MKL 2023.1
 - Compiler: GCC 11.4.0
 - OS: Ubuntu 22.04.4 LTS
 

diff --git a/assets/perf.png b/assets/perf.png
diff --git a/assets/ryzen7700_perf/benchmark_matmul.txt b/assets/ryzen7700_perf/benchmark_matmul.txt
@@ -1,38 +1,40 @@
-200 4 730 493
-400 146 980 936
-600 444 1022 1001
-800 586 1062 1043
-1000 867 1068 1061
-1200 672 1077 1054
-1400 772 1069 1051
-1600 985 1077 1070
-1800 795 1059 1045
-2000 936 1065 1053
-2200 1019 1050 1045
-2400 928 1059 1045
-2600 883 1052 1034
-2800 1036 1060 1052
-3000 1035 1058 1053
-3200 1024 1062 1057
-3400 1040 1061 1057
-3600 1015 1070 1059
-3800 1018 1070 1058
-4000 1027 1077 1069
-4200 1034 1067 1062
-4400 1049 1074 1068
-4600 1023 1075 1060
-4800 1045 1077 1067
-5000 1057 1077 1073
-5200 1049 1077 1069
-5400 1059 1077 1071
-5600 1058 1079 1076
-5800 1071 1080 1077
-6000 1054 1085 1081
-6200 1069 1078 1075
-6400 1056 1084 1076
-6600 1056 1081 1076
-6800 1080 1085 1083
-7000 1070 1082 1078
-7200 1065 1085 1078
-7400 1069 1082 1079
-7600 1072 1085 1081
+200 15 734 615
+400 150 981 935
+600 337 1026 1006
+800 785 1060 1047
+1000 761 1069 1055
+1200 863 1077 1065
+1400 938 1069 1061
+1600 1012 1076 1071
+1800 941 1064 1054
+2000 968 1071 1057
+2200 969 1050 1039
+2400 967 1060 1042
+2600 1021 1054 1049
+2800 986 1066 1057
+3000 1036 1068 1063
+3200 1017 1065 1059
+3400 978 1061 1048
+3600 1054 1071 1067
+3800 1055 1070 1065
+4000 1065 1077 1073
+4200 1056 1069 1065
+4400 1068 1076 1074
+4600 1041 1072 1065
+4800 1071 1082 1078
+5000 1069 1079 1075
+5200 1056 1077 1072
+5400 1069 1075 1073
+5600 1075 1080 1078
+5800 1068 1080 1075
+6000 1073 1085 1081
+6200 1056 1076 1071
+6400 1075 1082 1080
+6600 1075 1081 1079
+6800 1074 1085 1081
+7000 1077 1081 1079
+7200 1077 1082 1080
+7400 1076 1081 1079
+7600 1071 1082 1081
+7800 1067 1076 1073
+8000 1074 1079 1077
diff --git a/assets/ryzen7700_perf/benchmark_mkl.txt b/assets/ryzen7700_perf/benchmark_mkl.txt
@@ -1,38 +1,40 @@
-200 9 591 529
-400 700 813 803
-600 784 907 884
-800 886 950 939
-1000 884 948 944
-1200 798 988 979
-1400 867 988 963
-1600 856 976 956
-1800 890 951 933
-2000 893 935 928
-2200 877 967 935
-2400 865 974 964
-2600 736 977 866
-2800 919 961 954
-3000 771 977 871
-3200 897 931 917
-3400 724 740 731
-3600 915 941 924
-3800 721 732 727
-4000 924 941 930
-4200 716 724 719
-4400 922 934 927
-4600 712 719 715
-4800 909 925 914
-5000 715 720 717
-5200 911 925 919
-5400 710 715 713
-5600 925 943 933
-5800 721 725 723
-6000 892 943 935
-6200 714 721 718
-6400 907 918 914
-6600 723 727 725
-6800 934 943 939
-7000 717 719 718
-7200 938 948 943
-7400 723 727 725
-7600 942 951 947
+200 152 600 555
+400 653 816 780
+600 757 895 871
+800 796 941 921
+1000 896 952 949
+1200 940 993 978
+1400 908 990 973
+1600 914 986 970
+1800 904 963 951
+2000 946 963 960
+2200 890 976 957
+2400 893 982 940
+2600 813 986 915
+2800 762 838 803
+3000 756 975 854
+3200 750 960 869
+3400 735 757 747
+3600 947 965 959
+3800 727 748 742
+4000 952 969 961
+4200 726 739 734
+4400 937 971 958
+4600 717 728 725
+4800 936 961 944
+5000 715 726 721
+5200 933 944 939
+5400 716 721 718
+5600 940 960 951
+5800 721 725 724
+6000 927 957 948
+6200 717 721 719
+6400 926 939 932
+6600 724 729 727
+6800 950 964 955
+7000 720 723 722
+7200 954 965 959
+7400 725 728 727
+7600 956 968 961
+7800 725 727 726
+8000 949 959 952
diff --git a/assets/ryzen7700_perf/benchmark_openblas.txt b/assets/ryzen7700_perf/benchmark_openblas.txt
@@ -1,38 +1,40 @@
-200 52 543 514
-400 81 788 718
-600 143 698 658
-800 336 897 852
-1000 496 835 824
-1200 828 931 924
-1400 916 984 979
-1600 908 997 990
-1800 991 1061 1056
-2000 942 991 984
-2200 980 1052 1047
-2400 965 996 990
-2600 862 1041 1031
-2800 931 1013 999
-3000 1005 1038 1031
-3200 1065 1078 1074
-3400 1025 1035 1033
-3600 1049 1064 1057
-3800 982 1027 1020
-4000 1024 1056 1048
-4200 968 1034 1021
-4400 1031 1044 1039
-4600 1042 1053 1049
-4800 1011 1046 1028
-5000 1067 1069 1068
-5200 1028 1051 1046
-5400 1044 1079 1073
-5600 1030 1048 1042
-5800 1041 1069 1062
-6000 1041 1059 1057
-6200 1056 1071 1068
-6400 1081 1092 1089
-6600 1058 1067 1065
-6800 1069 1085 1080
-7000 1049 1063 1060
-7200 1063 1079 1073
-7400 1046 1062 1058
-7600 1049 1073 1068
+200 40 495 464
+400 106 765 721
+600 451 854 825
+800 760 934 920
+1000 856 947 925
+1200 903 1005 1000
+1400 983 1009 998
+1600 960 1057 1054
+1800 989 1044 1036
+2000 1012 1052 1050
+2200 1018 1057 1051
+2400 1013 1045 1030
+2600 1024 1049 1044
+2800 1027 1053 1037
+3000 1048 1071 1067
+3200 1035 1065 1055
+3400 1039 1057 1051
+3600 1042 1066 1058
+3800 1043 1053 1050
+4000 1032 1065 1053
+4200 1032 1068 1058
+4400 1059 1074 1068
+4600 1041 1062 1055
+4800 1059 1069 1065
+5000 1050 1057 1053
+5200 1062 1080 1075
+5400 1066 1076 1072
+5600 1063 1074 1065
+5800 1055 1074 1068
+6000 1061 1076 1070
+6200 1041 1070 1059
+6400 1059 1071 1068
+6600 1064 1071 1067
+6800 1067 1079 1075
+7000 1060 1072 1063
+7200 1069 1071 1070
+7400 1059 1064 1063
+7600 1060 1070 1068
+7800 1060 1070 1064
+8000 1055 1074 1067