update loongarch64 benchmarking result.

strongtz · May 6, 2024 · d8e6db7 · d8e6db7
1 parent d57479f
commit d8e6db7
Showing 1 changed file with 14 additions and 12 deletions.
diff --git a/benchmark_result/loongarch64.md b/benchmark_result/loongarch64.md
@@ -12,12 +12,12 @@ Number Threads: 1
 Thread Pool Binding: 0
 -----------------------------------------------------------
 | Instruction Set | Core Computation   | Peak Performance |
-| LASX            | fmadd(f32,f32,f32) | 79.89 GFLOPS     |
-| LASX            | fmadd(f64,f64,f64) | 39.95 GFLOPS     |
-| LSX             | fmadd(f32,f32,f32) | 39.952 GFLOPS    |
-| LSX             | fmadd(f64,f64,f64) | 19.964 GFLOPS    |
-| FP_SP           | fmadd(f32,f32,f32) | 9.9916 GFLOPS    |
-| FP_DP           | fmadd(f64,f64,f64) | 9.9919 GFLOPS    |
+| LASX            | fmadd(f32,f32,f32) | 119.92 GFLOPS    |
+| LASX            | fmadd(f64,f64,f64) | 59.957 GFLOPS    |
+| LSX             | fmadd(f32,f32,f32) | 59.958 GFLOPS    |
+| LSX             | fmadd(f64,f64,f64) | 29.98 GFLOPS     |
+| FP_SP           | fmadd(f32,f32,f32) | 14.99 GFLOPS     |
+| FP_DP           | fmadd(f64,f64,f64) | 14.99 GFLOPS     |
 -----------------------------------------------------------
 </pre>
 
@@ -29,12 +29,14 @@ Number Threads: 4
 Thread Pool Binding: 0 2 4 6
 -----------------------------------------------------------
 | Instruction Set | Core Computation   | Peak Performance |
-| LASX            | fmadd(f32,f32,f32) | 319.72 GFLOPS    |
-| LASX            | fmadd(f64,f64,f64) | 159.82 GFLOPS    |
-| LSX             | fmadd(f32,f32,f32) | 159.72 GFLOPS    |
-| LSX             | fmadd(f64,f64,f64) | 79.858 GFLOPS    |
-| FP_SP           | fmadd(f32,f32,f32) | 39.94 GFLOPS     |
-| FP_DP           | fmadd(f64,f64,f64) | 39.943 GFLOPS    |
+| LASX            | fmadd(f32,f32,f32) | 479.7 GFLOPS     |
+| LASX            | fmadd(f64,f64,f64) | 239.86 GFLOPS    |
+| LSX             | fmadd(f32,f32,f32) | 239.85 GFLOPS    |
+| LSX             | fmadd(f64,f64,f64) | 119.92 GFLOPS    |
+| FP_SP           | fmadd(f32,f32,f32) | 59.963 GFLOPS    |
+| FP_DP           | fmadd(f64,f64,f64) | 59.963 GFLOPS    |
 -----------------------------------------------------------
 </pre>
 
+NOTE: Loongson 3A6000 has 4 vector pipelines. Two of them are floating mul-add, while the other two are floating add. The numbers of muls and adds are not equal.
+