More tests for new FLOPS_AVX perf.group on Haswell authored by Thomas.Roehl's avatar Thomas.Roehl
......@@ -15,7 +15,7 @@ Not all performance groups can be tested for accuracy. We don't have a test appl
|-------|-------|
| BRANCH | load, triad, copy, stream, store |
| DATA | triad, copy, stream, store |
| FLOPS_AVX | triad_avx |
| FLOPS_AVX | ddot_avx, triad_avx, striad_avx, daxpy_avx |
| L2 | load, triad, copy, stream, store |
| L3 | load, triad, copy, stream, store |
......@@ -209,6 +209,26 @@ Each data size is tested 5 times, hence the first 5 entries on the x-axis corres
### Verification of Group FLOPS_AVX
#### Verification of Group FLOPS_AVX with Test ddot_avx
Benchmark performed with data sizes: 24kB, 128kB, 2MB, 1GB.
Each data size is tested 10 times, hence the first 10 entries on the x-axis correspond to the 10 runs for the first data size of 24kB and so on.
![haswell_FLOPS_AVX_ddot_avx](./images/accuracy/haswell/FLOPS_AVX_ddot_avx.jpg)
| Variant | BENCH (Min) | LIKWID (Min) | Error (Min) | BENCH (Max) | LIKWID (Max) | Error (Max) | BENCH (Avg) | LIKWID (Avg) | Error (Avg) |
| ------- | ------- | ------- | ------- | ------- | ------- | ------- | ------- | ------- | ------- |
| 24kB | 9688.00 | 9702.82 | 0.15% | 9718.00 | 9733.35 | 0.16% | 9714.20 | 9729.15 | 0.15% |
| 128kB | 8385.00 | 8388.47 | 0.04% | 9306.00 | 9309.96 | 9.86% | 9122.90 | 9126.60 | 0.04% |
| 2MB | 5774.00 | 5775.68 | 0.03% | 5798.00 | 5799.39 | 0.39% | 5791.30 | 5792.52 | 0.02% |
| 1GB | 2167.00 | 2168.24 | 0.06% | 2187.00 | 2187.80 | 0.86% | 2180.70 | 2181.39 | 0.03% |
#### Verification of Group FLOPS_AVX with Test triad_avx
......@@ -223,10 +243,50 @@ Each data size is tested 10 times, hence the first 10 entries on the x-axis corr
| Variant | BENCH (Min) | LIKWID (Min) | Error (Min) | BENCH (Max) | LIKWID (Max) | Error (Max) | BENCH (Avg) | LIKWID (Avg) | Error (Avg) |
| ------- | ------- | ------- | ------- | ------- | ------- | ------- | ------- | ------- | ------- |
| 24kB | 11812.00 | 11815.51 | 0.03% | 12174.00 | 12177.16 | 2.94% | 12109.70 | 12112.76 | 0.03% |
| 128kB | 4159.00 | 4159.86 | 0.02% | 4446.00 | 4446.68 | 6.44% | 4359.70 | 4360.63 | 0.02% |
| 2MB | 2497.00 | 2497.42 | 0.02% | 2543.00 | 2544.03 | 1.79% | 2536.10 | 2536.92 | 0.03% |
| 1GB | 941.00 | 941.57 | 0.06% | 949.00 | 949.87 | 0.78% | 947.40 | 948.13 | 0.08% |
| 24kB | 11813.00 | 11816.40 | 0.03% | 12174.00 | 12176.64 | 2.94% | 12126.70 | 12129.54 | 0.02% |
| 128kB | 3577.00 | 3577.93 | 0.03% | 4428.00 | 4428.74 | 19.20% | 4313.50 | 4314.32 | 0.02% |
| 2MB | 2541.00 | 2541.79 | 0.03% | 2545.00 | 2545.61 | 0.13% | 2543.50 | 2544.20 | 0.03% |
| 1GB | 946.00 | 946.29 | 0.03% | 950.00 | 950.33 | 0.39% | 948.40 | 949.13 | 0.08% |
#### Verification of Group FLOPS_AVX with Test striad_avx
Benchmark performed with data sizes: 24kB, 128kB, 2MB, 1GB.
Each data size is tested 10 times, hence the first 10 entries on the x-axis correspond to the 10 runs for the first data size of 24kB and so on.
![haswell_FLOPS_AVX_striad_avx](./images/accuracy/haswell/FLOPS_AVX_striad_avx.jpg)
| Variant | BENCH (Min) | LIKWID (Min) | Error (Min) | BENCH (Max) | LIKWID (Max) | Error (Max) | BENCH (Avg) | LIKWID (Avg) | Error (Avg) |
| ------- | ------- | ------- | ------- | ------- | ------- | ------- | ------- | ------- | ------- |
| 24kB | 16894.00 | 16898.54 | 0.03% | 17196.00 | 17199.79 | 1.73% | 17093.70 | 17097.89 | 0.02% |
| 128kB | 5416.00 | 5417.50 | 0.03% | 5504.00 | 5504.69 | 1.57% | 5468.90 | 5469.98 | 0.02% |
| 2MB | 3174.00 | 3174.55 | 0.02% | 3182.00 | 3183.29 | 0.23% | 3179.40 | 3180.24 | 0.03% |
| 1GB | 1055.00 | 1055.27 | 0.03% | 1197.00 | 1197.36 | 11.84% | 1174.50 | 1175.03 | 0.04% |
#### Verification of Group FLOPS_AVX with Test daxpy_avx
Benchmark performed with data sizes: 24kB, 128kB, 2MB, 1GB.
Each data size is tested 10 times, hence the first 10 entries on the x-axis correspond to the 10 runs for the first data size of 24kB and so on.
![haswell_FLOPS_AVX_daxpy_avx](./images/accuracy/haswell/FLOPS_AVX_daxpy_avx.jpg)
| Variant | BENCH (Min) | LIKWID (Min) | Error (Min) | BENCH (Max) | LIKWID (Max) | Error (Max) | BENCH (Avg) | LIKWID (Avg) | Error (Avg) |
| ------- | ------- | ------- | ------- | ------- | ------- | ------- | ------- | ------- | ------- |
| 24kB | 16053.00 | 16055.68 | 0.02% | 17364.00 | 17368.12 | 7.53% | 17220.40 | 17223.88 | 0.02% |
| 128kB | 6661.00 | 6662.48 | 0.02% | 6732.00 | 6732.72 | 1.03% | 6715.40 | 6716.62 | 0.02% |
| 2MB | 4234.00 | 4234.52 | 0.01% | 4247.00 | 4248.29 | 0.29% | 4237.30 | 4238.44 | 0.03% |
| 1GB | 1603.00 | 1603.22 | 0.01% | 1616.00 | 1616.61 | 0.79% | 1612.60 | 1613.23 | 0.04% |
### Verification of Group L2
......
......