Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Compare with cuBLAS during benchmarking #169

Merged
merged 1 commit into from
Nov 10, 2023
Merged

Compare with cuBLAS during benchmarking #169

merged 1 commit into from
Nov 10, 2023

Conversation

thomasfaingnaert
Copy link
Member

@thomasfaingnaert thomasfaingnaert commented Nov 6, 2023 •
edited
Loading

depends on #168

@maleadt
Copy link
Member

maleadt commented Nov 7, 2023 •
edited
Loading

Benchmark results for commit 7420ebb (comparing to 5dfb8ab):

test master PR Δmin
WMMA GEMM+bias Float16*Float16+Float32=Float32 (128×128) · (128×128) (TN) OP (16, 16, 16) 8.91 μs ± 1.92% (8.58 … 9.3 μs) 10.1 μs ± 1.38% (9.78 … 10.3 μs) +13.9% ❌
WMMA GEMM Float16*Float16=Float16 (2048×2048) · (2048×2048) (TN) OP (16, 16, 16) 263.0 μs ± 6.03% (248.0 … 286.0 μs) 283.0 μs ± 0.342% (281.0 … 287.0 μs) +13.5% ❌
WMMA GEMM Float16*Float16=Float32 (128×128) · (128×128) (NN) OP (16, 16, 16) 8.04 μs ± 2.05% (7.87 … 8.34 μs) 9.17 μs ± 1.69% (8.82 … 9.54 μs) +12.1% ❌
Tropical GEMM Float32*Float32=Float32 (128×256) · (256×128) (TT) OP (8, 16, 2), base shape (4, 8, 1) 124.0 μs ± 2.89% (109.0 … 126.0 μs) 125.0 μs ± 0.126% (125.0 … 126.0 μs) +14.2% ❌
Tropical GEMM Float32*Float32=Float32 (128×256) · (256×128) (NN) OP (8, 16, 2), base shape (4, 8, 1) 125.0 μs ± 0.12% (124.0 … 126.0 μs) 125.0 μs ± 1.06% (109.0 … 125.0 μs) -12.1% ✅
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NN) OP (4, 8, 1), base shape (4, 8, 1) 15.3 μs ± 1.1% (15.0 … 15.5 μs) 13.7 μs ± 4.97% (13.1 … 16.7 μs) -12.7% ✅
WMMA Complex GEMM Float16*Float16=Float32 (128×128) · (128×128) (NT) OP (16, 16, 16) 8.51 μs ± 2.14% (8.11 … 8.82 μs) 9.59 μs ± 1.73% (9.3 … 9.78 μs) +14.7% ❌
Tropical GEMM Float32*Float32=Float32 (128×128) · (128×128) (NT) OP (8, 16, 2), base shape (4, 8, 1) 64.8 μs ± 0.264% (64.4 … 65.1 μs) 56.8 μs ± 0.288% (56.5 … 57.2 μs) -12.2% ✅
FPU GEMM Float32*Float32=Float64 (128×128) · (128×128) (NN) OP (8, 16, 2), base shape (4, 8, 1) 180.0 μs ± 0.221% (178.0 … 182.0 μs) 172.0 μs ± 6.1% (157.0 … 181.0 μs) -11.9% ✅

Comparison with baseline

test GemmKernels Baseline %
FPU GEMM Float32*Float32=Float32 (256×256) · (256×256) (TT) OP (8, 16, 2), base shape (4, 8, 1) 15.0 μs ± 1.13% (14.8 … 15.3 μs) 7.18 μs ± 1.63% (6.91 … 7.63 μs) 46.8
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NT) OP (4, 32, 1), base shape (1, 32, 1) 16.1 μs ± 0.967% (15.7 … 16.2 μs) 4.17 μs ± 3.67% (4.05 … 4.53 μs) 25.8
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TT) OP (16, 16, 1), base shape (8, 4, 1) 14.5 μs ± 1.08% (14.3 … 14.8 μs) 4.03 μs ± 3.41% (3.81 … 4.29 μs) 26.7
FPU GEMM Float32*Float32=Float32 (2048×2048) · (2048×2048) (NN) OP (8, 16, 2), base shape (4, 8, 1) 1.86 ms ± 3.13% (1.82 … 2.02 ms) 1.0 ms ± 1.28% (0.94 … 1.03 ms) 51.7
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TT) OP (32, 4, 1), base shape (32, 1, 1) 13.1 μs ± 0.959% (12.6 … 13.4 μs) 4.05 μs ± 3.41% (3.81 … 4.29 μs) 30.2
WMMA GEMM Float16*Float16=Float32 (256×256) · (256×256) (TT) OP (16, 16, 16) 12.2 μs ± 1.32% (11.9 … 12.6 μs) 3.94 μs ± 3.94% (3.58 … 4.29 μs) 30.0
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NN) OP (32, 4, 1), base shape (16, 2, 1) 19.1 μs ± 0.885% (18.8 … 19.3 μs) 4.08 μs ± 2.67% (3.81 … 4.29 μs) 20.3
FPU GEMM Float16*Float16=Float32 (2048×2048) · (2048×2048) (TT) OP (8, 16, 2), base shape (4, 8, 1) 1.84 ms ± 2.35% (1.82 … 2.01 ms) 292.0 μs ± 0.244% (290.0 … 294.0 μs) 15.9
FPU GEMM Float64*Float64=Float64 (2048×2048) · (2048×2048) (TN) OP (8, 16, 2), base shape (4, 8, 1) 46.1 ms ± 0.868% (46.0 … 52.4 ms) 40.8 ms ± 0.00577% (40.8 … 40.8 ms) 88.7
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TN) OP (32, 4, 1), base shape (16, 2, 1) 19.1 μs ± 0.919% (18.8 … 19.3 μs) 4.29 μs ± 3.52% (4.05 … 4.53 μs) 21.5
FPU GEMM Float64*Float64=Float64 (256×256) · (256×256) (TN) OP (8, 16, 2), base shape (4, 8, 1) 277.0 μs ± 6.27% (255.0 … 294.0 μs) 109.0 μs ± 0.161% (109.0 … 109.0 μs) 42.7
WMMA GEMM Float16*Float16=Float32 (256×128) · (128×256) (TN) OP (16, 16, 16) 9.04 μs ± 1.87% (8.82 … 9.3 μs) 2.87 μs ± 4.88% (2.62 … 3.1 μs) 29.7
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NN) OP (8, 16, 2), base shape (4, 8, 1) 15.3 μs ± 1.14% (15.0 … 15.7 μs) 4.07 μs ± 2.83% (3.81 … 4.29 μs) 25.4
WMMA GEMM Float16*Float16=Float32 (128×128) · (128×128) (TN) OP (16, 16, 16) 7.91 μs ± 1.64% (7.63 … 8.34 μs) 2.35 μs ± 6.59% (2.15 … 2.86 μs) 28.1
WMMA GEMM Float16*Float16=Float16 (512×256) · (256×512) (NN) OP (16, 16, 16) 19.0 μs ± 0.905% (18.8 … 19.3 μs) 4.98 μs ± 3.51% (4.77 … 5.25 μs) 25.3
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TN) OP (4, 8, 1), base shape (4, 8, 1) 15.2 μs ± 1.12% (15.0 … 15.5 μs) 4.3 μs ± 3.39% (4.05 … 4.53 μs) 27.0
FPU GEMM Float32*Float32=Float32 (2048×2048) · (2048×2048) (TT) OP (8, 16, 2), base shape (4, 8, 1) 1.88 ms ± 2.62% (1.85 … 2.05 ms) 1.12 ms ± 1.91% (1.04 … 1.16 ms) 56.1
FPU GEMM Float64*Float64=Float64 (256×256) · (256×256) (NN) OP (8, 16, 2), base shape (4, 8, 1) 290.0 μs ± 2.29% (256.0 … 292.0 μs) 109.0 μs ± 0.15% (109.0 … 109.0 μs) 42.6
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TT) OP (4, 8, 1), base shape (4, 8, 1) 15.2 μs ± 1.1% (15.0 … 15.5 μs) 4.03 μs ± 3.47% (3.81 … 4.29 μs) 25.4
WMMA GEMM Float16*Float16=Float32 (256×256) · (256×256) (NN) OP (16, 16, 16) 12.3 μs ± 1.13% (11.9 … 12.6 μs) 3.95 μs ± 3.7% (3.81 … 4.29 μs) 32.0
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NN) OP (16, 16, 1), base shape (4, 8, 1) 15.3 μs ± 0.919% (15.0 … 15.7 μs) 4.08 μs ± 2.71% (3.81 … 4.29 μs) 25.4
FPU GEMM Float32*Float32=Float32 (2048×2048) · (2048×2048) (TN) OP (8, 16, 2), base shape (4, 8, 1) 1.46 ms ± 1.77% (1.43 … 1.53 ms) 1.12 ms ± 1.54% (1.05 … 1.16 ms) 73.1
WMMA GEMM Float16*Float16=Float32 (2048×2048) · (2048×2048) (NT) OP (16, 16, 16) 375.0 μs ± 5.13% (350.0 … 397.0 μs) 291.0 μs ± 0.344% (289.0 … 293.0 μs) 82.8
FPU GEMM Float16*Float16=Float32 (2048×2048) · (2048×2048) (NN) OP (8, 16, 2), base shape (4, 8, 1) 1.86 ms ± 2.76% (1.82 … 2.02 ms) 293.0 μs ± 0.272% (290.0 … 294.0 μs) 15.9
FPU GEMM Float64*Float64=Float64 (128×128) · (128×128) (TT) OP (8, 16, 2), base shape (4, 8, 1) 130.0 μs ± 0.405% (129.0 … 131.0 μs) 28.7 μs ± 0.459% (28.4 … 29.1 μs) 22.0
WMMA GEMM Float16*Float16=Float32 (2048×2048) · (2048×2048) (TT) OP (16, 16, 16) 373.0 μs ± 4.71% (342.0 … 389.0 μs) 292.0 μs ± 0.27% (290.0 … 293.0 μs) 84.9
FPU GEMM Float64*Float64=Float64 (2048×2048) · (2048×2048) (TT) OP (8, 16, 2), base shape (4, 8, 1) 46.1 ms ± 0.871% (46.0 … 52.4 ms) 40.8 ms ± 0.00639% (40.8 … 40.8 ms) 88.7
WMMA GEMM Float16*Float16=Float16 (256×256) · (256×256) (TN) OP (16, 16, 16) 18.1 μs ± 0.844% (17.9 … 18.6 μs) 3.58 μs ± 2.92% (3.1 … 4.05 μs) 17.3
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NT) OP (32, 4, 1), base shape (32, 1, 1) 12.6 μs ± 1.3% (12.4 … 12.9 μs) 4.18 μs ± 3.63% (4.05 … 4.53 μs) 32.7
WMMA GEMM Float16*Float16=Float16 (256×256) · (256×256) (NN) OP (16, 16, 16) 18.9 μs ± 0.861% (18.6 … 19.3 μs) 3.71 μs ± 3.76% (3.58 … 4.05 μs) 19.2
FPU GEMM Float16*Float16=Float32 (256×256) · (256×256) (TN) OP (8, 16, 2), base shape (4, 8, 1) 13.0 μs ± 1.15% (12.6 … 13.1 μs) 3.79 μs ± 3.24% (3.58 … 4.05 μs) 28.3
WMMA GEMM Float16*Float16=Float32 (128×128) · (128×128) (TT) OP (16, 16, 16) 8.03 μs ± 2.05% (7.87 … 8.34 μs) 2.45 μs ± 6.69% (2.15 … 2.62 μs) 27.3
WMMA GEMM Float16*Float16=Float32 (2048×2048) · (2048×2048) (NN) OP (16, 16, 16) 371.0 μs ± 4.88% (343.0 … 389.0 μs) 292.0 μs ± 0.268% (290.0 … 294.0 μs) 84.5
FPU GEMM Float16*Float16=Float32 (2048×2048) · (2048×2048) (NT) OP (8, 16, 2), base shape (4, 8, 1) 2.3 ms ± 3.47% (2.26 … 2.57 ms) 292.0 μs ± 0.237% (290.0 … 293.0 μs) 12.8
FPU GEMM Float16*Float16=Float32 (128×128) · (128×128) (TN) OP (8, 16, 2), base shape (4, 8, 1) 6.88 μs ± 1.28% (6.44 … 7.39 μs) 2.08 μs ± 6.66% (1.91 … 2.38 μs) 29.6
WMMA GEMM Float16*Float16=Float16 (512×512) · (512×512) (NN) OP (16, 16, 16) 30.7 μs ± 0.451% (30.3 … 31.0 μs) 7.32 μs ± 2.05% (7.15 … 7.63 μs) 23.6
WMMA GEMM Float16*Float16=Float16 (512×512) · (512×512) (TT) OP (16, 16, 16) 30.7 μs ± 0.484% (30.3 … 31.0 μs) 7.34 μs ± 2.0% (7.15 … 7.63 μs) 23.6
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NT) OP (4, 8, 2), base shape (4, 8, 1) 16.0 μs ± 1.03% (15.7 … 16.5 μs) 4.18 μs ± 3.73% (4.05 … 4.53 μs) 25.8
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NT) OP (8, 16, 2), base shape (4, 8, 1) 16.4 μs ± 1.0% (16.0 … 16.7 μs) 4.19 μs ± 3.71% (4.05 … 4.53 μs) 25.4
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NN) OP (32, 4, 1), base shape (32, 1, 1) 31.6 μs ± 0.461% (31.5 … 31.9 μs) 4.09 μs ± 2.72% (3.81 … 4.29 μs) 12.1
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TT) OP (8, 16, 2), base shape (4, 8, 1) 13.6 μs ± 0.994% (13.1 … 13.8 μs) 4.03 μs ± 3.54% (3.81 … 4.29 μs) 29.1
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NN) OP (4, 32, 1), base shape (1, 32, 1) 15.7 μs ± 1.06% (15.5 … 16.0 μs) 4.08 μs ± 2.59% (3.81 … 4.29 μs) 24.6
FPU GEMM Float16*Float16=Float32 (256×256) · (256×256) (NT) OP (8, 16, 2), base shape (4, 8, 1) 18.1 μs ± 0.878% (17.9 … 18.4 μs) 3.91 μs ± 3.68% (3.58 … 4.29 μs) 20.0
WMMA GEMM Float16*Float16=Float32 (128×128) · (128×128) (NT) OP (16, 16, 16) 8.04 μs ± 2.11% (7.87 … 8.34 μs) 2.31 μs ± 5.63% (1.91 … 2.62 μs) 24.2
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TN) OP (4, 32, 1), base shape (2, 16, 1) 21.9 μs ± 0.751% (21.7 … 22.2 μs) 4.3 μs ± 3.35% (4.05 … 4.53 μs) 18.7
FPU GEMM Float64*Float64=Float64 (128×128) · (128×128) (NT) OP (8, 16, 2), base shape (4, 8, 1) 143.0 μs ± 5.76% (130.0 … 149.0 μs) 28.6 μs ± 0.532% (28.4 … 28.8 μs) 21.9
WMMA GEMM Float16*Float16=Float32 (128×256) · (256×128) (NN) OP (16, 16, 16) 12.2 μs ± 1.43% (11.9 … 12.4 μs) 3.09 μs ± 3.2% (2.86 … 3.34 μs) 24.0
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NN) OP (16, 16, 1), base shape (8, 4, 1) 14.5 μs ± 1.06% (14.3 … 14.8 μs) 4.07 μs ± 2.72% (3.81 … 4.29 μs) 26.7
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TT) OP (4, 8, 2), base shape (4, 8, 1) 12.4 μs ± 1.41% (12.2 … 12.6 μs) 4.04 μs ± 3.45% (3.81 … 4.29 μs) 31.4
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TT) OP (16, 16, 1), base shape (4, 8, 1) 15.4 μs ± 1.08% (15.3 … 15.7 μs) 4.03 μs ± 3.57% (3.81 … 4.29 μs) 25.0
WMMA GEMM Float16*Float16=Float32 (128×256) · (256×128) (TN) OP (16, 16, 16) 12.3 μs ± 1.1% (12.2 … 12.6 μs) 2.9 μs ± 5.07% (2.62 … 3.1 μs) 21.6
WMMA GEMM Float16*Float16=Float16 (512×512) · (512×512) (NT) OP (16, 16, 16) 30.8 μs ± 0.495% (30.3 … 31.2 μs) 7.25 μs ± 1.95% (6.91 … 7.63 μs) 22.8
WMMA GEMM Float16*Float16=Float16 (256×512) · (512×256) (TN) OP (16, 16, 16) 28.6 μs ± 0.576% (28.4 … 29.1 μs) 5.12 μs ± 3.32% (4.77 … 5.72 μs) 16.8
FPU GEMM Float64*Float64=Float64 (128×128) · (128×128) (NN) OP (8, 16, 2), base shape (4, 8, 1) 148.0 μs ± 0.16% (147.0 … 149.0 μs) 28.7 μs ± 0.461% (28.6 … 29.1 μs) 19.4
WMMA GEMM Float16*Float16=Float32 (128×256) · (256×128) (NT) OP (16, 16, 16) 12.4 μs ± 1.35% (12.2 … 12.6 μs) 2.76 μs ± 5.24% (2.38 … 3.1 μs) 19.6
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TT) OP (4, 32, 1), base shape (2, 16, 1) 21.9 μs ± 0.752% (21.7 … 22.2 μs) 4.03 μs ± 3.46% (3.81 … 4.29 μs) 17.6
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TT) OP (32, 4, 1), base shape (16, 2, 1) 19.1 μs ± 0.876% (18.8 … 19.3 μs) 4.03 μs ± 3.46% (3.81 … 4.29 μs) 20.3
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TN) OP (4, 8, 2), base shape (4, 8, 1) 11.9 μs ± 1.4% (11.7 … 12.2 μs) 4.28 μs ± 3.57% (4.05 … 4.77 μs) 34.7
WMMA GEMM Float16*Float16=Float16 (2048×2048) · (2048×2048) (TT) OP (16, 16, 16) 308.0 μs ± 5.28% (277.0 … 324.0 μs) 189.0 μs ± 2.12% (179.0 … 196.0 μs) 64.7
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TN) OP (4, 16, 1), base shape (4, 8, 1) 15.2 μs ± 1.12% (15.0 … 15.5 μs) 4.3 μs ± 3.41% (4.05 … 4.53 μs) 27.0
WMMA GEMM Float16*Float16=Float16 (256×256) · (256×256) (TT) OP (16, 16, 16) 18.9 μs ± 0.859% (18.6 … 19.3 μs) 3.65 μs ± 3.39% (3.58 … 4.05 μs) 19.2
WMMA GEMM Float16*Float16=Float16 (512×256) · (256×512) (TN) OP (16, 16, 16) 18.3 μs ± 0.91% (18.1 … 18.6 μs) 5.14 μs ± 2.95% (4.77 … 5.48 μs) 26.3
FPU GEMM Float64*Float64=Float64 (2048×2048) · (2048×2048) (NT) OP (8, 16, 2), base shape (4, 8, 1) 46.4 ms ± 0.777% (46.4 … 52.8 ms) 40.8 ms ± 0.00547% (40.8 … 40.8 ms) 88.0
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NN) OP (8, 8, 1), base shape (4, 8, 1) 15.4 μs ± 0.85% (15.0 … 15.7 μs) 4.09 μs ± 2.66% (3.81 … 4.29 μs) 25.4
WMMA GEMM Float16*Float16=Float16 (256×256) · (256×256) (NT) OP (16, 16, 16) 18.9 μs ± 0.91% (18.6 … 19.3 μs) 3.82 μs ± 2.86% (3.58 … 4.29 μs) 19.2
WMMA GEMM Float16*Float16=Float16 (512×512) · (512×512) (TN) OP (16, 16, 16) 29.0 μs ± 0.555% (28.6 … 29.3 μs) 7.42 μs ± 1.56% (7.15 … 7.87 μs) 25.0
WMMA GEMM Float16*Float16=Float16 (256×512) · (512×256) (NN) OP (16, 16, 16) 30.4 μs ± 0.535% (30.0 … 30.8 μs) 5.21 μs ± 3.32% (5.01 … 5.72 μs) 16.7
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NN) OP (4, 8, 1), base shape (4, 8, 1) 13.7 μs ± 4.97% (13.1 … 16.7 μs) 4.08 μs ± 2.54% (3.81 … 4.53 μs) 29.1
FPU GEMM Float16*Float16=Float32 (128×128) · (128×128) (NT) OP (8, 16, 2), base shape (4, 8, 1) 9.18 μs ± 1.56% (8.82 … 9.54 μs) 2.04 μs ± 6.18% (1.67 … 2.38 μs) 18.9
WMMA GEMM Float16*Float16=Float32 (256×256) · (256×256) (TN) OP (16, 16, 16) 12.4 μs ± 1.35% (12.2 … 12.6 μs) 3.81 μs ± 3.14% (3.58 … 4.29 μs) 29.4
WMMA GEMM Float16*Float16=Float32 (256×256) · (256×256) (NT) OP (16, 16, 16) 12.4 μs ± 1.4% (12.2 … 12.6 μs) 3.9 μs ± 3.79% (3.58 … 4.29 μs) 29.4
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NT) OP (16, 16, 1), base shape (8, 4, 1) 14.6 μs ± 0.91% (14.3 … 14.8 μs) 4.19 μs ± 3.72% (4.05 … 4.53 μs) 28.3
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TT) OP (4, 32, 1), base shape (1, 32, 1) 33.0 μs ± 0.421% (32.7 … 33.4 μs) 4.03 μs ± 3.5% (3.81 … 4.29 μs) 11.7
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TT) OP (4, 16, 1), base shape (4, 8, 1) 15.2 μs ± 1.09% (15.0 … 15.5 μs) 4.03 μs ± 3.49% (3.81 … 4.29 μs) 25.4
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NT) OP (32, 4, 1), base shape (16, 2, 1) 19.2 μs ± 0.801% (18.8 … 19.3 μs) 4.18 μs ± 3.71% (4.05 … 4.53 μs) 21.5
FPU GEMM Float32*Float32=Float32 (2048×2048) · (2048×2048) (NT) OP (8, 16, 2), base shape (4, 8, 1) 2.37 ms ± 3.72% (2.32 … 2.63 ms) 873.0 μs ± 1.26% (822.0 … 887.0 μs) 35.4
FPU GEMM Float64*Float64=Float64 (256×256) · (256×256) (TT) OP (8, 16, 2), base shape (4, 8, 1) 282.0 μs ± 5.87% (256.0 … 295.0 μs) 109.0 μs ± 0.154% (109.0 … 109.0 μs) 42.6
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TT) OP (8, 8, 1), base shape (4, 8, 1) 15.3 μs ± 1.03% (15.0 … 15.5 μs) 4.04 μs ± 3.59% (3.81 … 4.53 μs) 25.4
WMMA GEMM Float16*Float16=Float16 (256×512) · (512×256) (TT) OP (16, 16, 16) 30.5 μs ± 0.544% (30.0 … 30.8 μs) 5.21 μs ± 3.23% (5.01 … 5.48 μs) 16.7
FPU GEMM Float16*Float16=Float32 (128×128) · (128×128) (TT) OP (8, 16, 2), base shape (4, 8, 1) 8.17 μs ± 1.63% (7.87 … 8.34 μs) 2.17 μs ± 7.73% (1.91 … 2.38 μs) 24.2
FPU GEMM Float32*Float32=Float32 (256×256) · (256×256) (NN) OP (8, 16, 2), base shape (4, 8, 1) 15.3 μs ± 1.13% (15.0 … 15.5 μs) 6.78 μs ± 2.03% (6.44 … 6.91 μs) 42.9
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NT) OP (8, 8, 1), base shape (4, 8, 1) 15.4 μs ± 1.04% (15.3 … 15.7 μs) 4.18 μs ± 3.69% (4.05 … 4.53 μs) 26.6
FPU GEMM Float64*Float64=Float64 (256×256) · (256×256) (NT) OP (8, 16, 2), base shape (4, 8, 1) 281.0 μs ± 5.85% (256.0 … 294.0 μs) 109.0 μs ± 0.156% (109.0 … 109.0 μs) 42.6
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NT) OP (4, 16, 1), base shape (4, 8, 1) 15.3 μs ± 0.947% (15.0 … 15.7 μs) 4.19 μs ± 3.81% (4.05 … 4.53 μs) 27.0
FPU GEMM Float16*Float16=Float32 (128×128) · (128×128) (NN) OP (8, 16, 2), base shape (4, 8, 1) 9.22 μs ± 1.34% (8.82 … 9.54 μs) 2.21 μs ± 7.66% (1.91 … 2.62 μs) 21.6
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TN) OP (8, 8, 1), base shape (4, 8, 1) 15.2 μs ± 1.13% (15.0 … 15.5 μs) 4.28 μs ± 3.41% (4.05 … 4.53 μs) 27.0
WMMA GEMM Float16*Float16=Float16 (2048×2048) · (2048×2048) (TN) OP (16, 16, 16) 283.0 μs ± 0.342% (281.0 … 287.0 μs) 189.0 μs ± 3.03% (180.0 … 196.0 μs) 64.2
WMMA GEMM Float16*Float16=Float32 (256×128) · (128×256) (TT) OP (16, 16, 16) 9.23 μs ± 1.38% (8.82 … 9.54 μs) 2.9 μs ± 5.06% (2.62 … 3.1 μs) 29.7
WMMA GEMM Float16*Float16=Float32 (128×256) · (256×128) (TT) OP (16, 16, 16) 12.2 μs ± 1.36% (11.9 … 12.4 μs) 2.99 μs ± 4.75% (2.62 … 3.1 μs) 22.0
FPU GEMM Float16*Float16=Float32 (256×256) · (256×256) (TT) OP (8, 16, 2), base shape (4, 8, 1) 15.8 μs ± 1.07% (15.5 … 16.0 μs) 3.96 μs ± 3.99% (3.58 … 4.29 μs) 23.1
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NT) OP (16, 16, 1), base shape (4, 8, 1) 15.3 μs ± 0.919% (15.0 … 15.7 μs) 4.18 μs ± 3.61% (4.05 … 4.53 μs) 27.0
WMMA GEMM Float16*Float16=Float32 (2048×2048) · (2048×2048) (TN) OP (16, 16, 16) 368.0 μs ± 4.89% (350.0 … 396.0 μs) 294.0 μs ± 0.395% (290.0 … 296.0 μs) 83.0
WMMA GEMM Float16*Float16=Float32 (128×128) · (128×128) (NN) OP (16, 16, 16) 9.17 μs ± 1.69% (8.82 … 9.54 μs) 2.21 μs ± 7.56% (1.91 … 2.38 μs) 21.6
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NN) OP (4, 8, 2), base shape (4, 8, 1) 14.9 μs ± 1.08% (14.8 … 15.3 μs) 4.09 μs ± 2.58% (3.81 … 4.53 μs) 25.8
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TN) OP (32, 4, 1), base shape (32, 1, 1) 31.6 μs ± 0.459% (31.5 … 31.9 μs) 4.3 μs ± 3.28% (4.05 … 4.53 μs) 12.9
WMMA GEMM Float16*Float16=Float32 (256×128) · (128×256) (NT) OP (16, 16, 16) 9.15 μs ± 1.69% (8.82 … 9.3 μs) 2.91 μs ± 5.15% (2.62 … 3.1 μs) 29.7
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NT) OP (4, 32, 1), base shape (2, 16, 1) 21.9 μs ± 0.761% (21.7 … 22.2 μs) 4.18 μs ± 3.69% (4.05 … 4.53 μs) 18.7
FPU GEMM Float16*Float16=Float32 (256×256) · (256×256) (NN) OP (8, 16, 2), base shape (4, 8, 1) 15.7 μs ± 1.12% (15.5 … 16.0 μs) 3.94 μs ± 3.83% (3.81 … 4.29 μs) 24.6
FPU GEMM Float16*Float16=Float32 (2048×2048) · (2048×2048) (TN) OP (8, 16, 2), base shape (4, 8, 1) 1.56 ms ± 3.07% (1.52 … 1.67 ms) 293.0 μs ± 0.387% (290.0 … 295.0 μs) 19.0
FPU GEMM Float64*Float64=Float64 (2048×2048) · (2048×2048) (NN) OP (8, 16, 2), base shape (4, 8, 1) 46.3 ms ± 0.794% (46.3 … 52.7 ms) 40.8 ms ± 0.006% (40.8 … 40.8 ms) 88.2
WMMA GEMM Float16*Float16=Float32 (256×128) · (128×256) (NN) OP (16, 16, 16) 9.31 μs ± 1.78% (9.06 … 9.54 μs) 2.92 μs ± 5.1% (2.62 … 3.1 μs) 28.9
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NN) OP (4, 32, 1), base shape (2, 16, 1) 22.0 μs ± 0.606% (21.7 … 22.4 μs) 4.08 μs ± 2.64% (3.81 … 4.29 μs) 17.6
WMMA GEMM Float16*Float16=Float16 (256×512) · (512×256) (NT) OP (16, 16, 16) 30.6 μs ± 0.563% (30.0 … 31.0 μs) 4.51 μs ± 3.82% (4.29 … 5.25 μs) 14.3
FPU GEMM Float32*Float32=Float32 (256×256) · (256×256) (NT) OP (8, 16, 2), base shape (4, 8, 1) 18.2 μs ± 0.787% (17.9 … 18.6 μs) 6.77 μs ± 2.05% (6.44 … 6.91 μs) 36.0
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NN) OP (4, 16, 1), base shape (4, 8, 1) 15.2 μs ± 1.14% (15.0 … 15.5 μs) 4.08 μs ± 2.74% (3.81 … 4.29 μs) 25.4
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NT) OP (4, 8, 1), base shape (4, 8, 1) 15.3 μs ± 1.11% (15.0 … 15.5 μs) 4.19 μs ± 3.71% (4.05 … 4.53 μs) 27.0
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TN) OP (4, 32, 1), base shape (1, 32, 1) 33.0 μs ± 0.445% (32.7 … 33.1 μs) 4.28 μs ± 3.56% (4.05 … 4.53 μs) 12.4
WMMA GEMM Float16*Float16=Float16 (512×256) · (256×512) (NT) OP (16, 16, 16) 19.1 μs ± 0.88% (18.8 … 19.6 μs) 4.93 μs ± 3.37% (4.77 … 5.25 μs) 25.3
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TN) OP (8, 16, 2), base shape (4, 8, 1) 12.1 μs ± 1.22% (11.9 … 12.4 μs) 4.3 μs ± 3.33% (4.05 … 4.77 μs) 34.0
FPU GEMM Float64*Float64=Float64 (128×128) · (128×128) (TN) OP (8, 16, 2), base shape (4, 8, 1) 130.0 μs ± 0.273% (129.0 … 131.0 μs) 28.7 μs ± 0.476% (28.4 … 29.1 μs) 22.0
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TN) OP (16, 16, 1), base shape (4, 8, 1) 15.2 μs ± 1.15% (15.0 … 15.5 μs) 4.29 μs ± 3.46% (4.05 … 4.53 μs) 27.0
WMMA GEMM Float16*Float16=Float16 (512×256) · (256×512) (TT) OP (16, 16, 16) 19.2 μs ± 0.868% (18.8 … 19.6 μs) 5.2 μs ± 3.22% (5.01 … 5.48 μs) 26.6
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TN) OP (16, 16, 1), base shape (8, 4, 1) 14.5 μs ± 1.06% (14.3 … 14.8 μs) 4.3 μs ± 3.36% (4.05 … 4.53 μs) 28.3
WMMA GEMM Float16*Float16=Float16 (2048×2048) · (2048×2048) (NT) OP (16, 16, 16) 295.0 μs ± 6.14% (278.0 … 324.0 μs) 186.0 μs ± 2.36% (172.0 … 193.0 μs) 61.9
WMMA GEMM Float16*Float16=Float16 (2048×2048) · (2048×2048) (NN) OP (16, 16, 16) 319.0 μs ± 3.32% (288.0 … 335.0 μs) 186.0 μs ± 2.49% (175.0 … 191.0 μs) 61.0
FPU GEMM Float32*Float32=Float32 (256×256) · (256×256) (TN) OP (8, 16, 2), base shape (4, 8, 1) 12.6 μs ± 1.29% (12.4 … 12.9 μs) 7.15 μs ± 1.2% (6.68 … 7.63 μs) 53.8

@codecov Codecov
Copy link

codecov bot commented Nov 7, 2023 •
edited
Loading

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (5dfb8ab) 33.58% compared to head (7420ebb) 33.58%.

Additional details and impacted files 
@@           Coverage Diff           @@
##           master     #169   +/-   ##
=======================================
  Coverage   33.58%   33.58%           
=======================================
  Files          11       11           
  Lines         914      914           
=======================================
  Hits          307      307           
  Misses        607      607

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

@thomasfaingnaert thomasfaingnaert marked this pull request as ready for review November 7, 2023 19:56
@thomasfaingnaert thomasfaingnaert merged commit d08656f into master Nov 10, 2023
1 check passed
@thomasfaingnaert thomasfaingnaert deleted the tf/benchmarks-baseline branch November 10, 2023 16:01
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@maleadt maleadt Awaiting requested review from maleadt

Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
2 participants
@thomasfaingnaert @maleadt