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

Do not hardcode vectorisation width in layouts #176

Merged
merged 1 commit into from
Nov 15, 2023
Merged

Do not hardcode vectorisation width in layouts #176

merged 1 commit into from
Nov 15, 2023

Conversation

thomasfaingnaert
Copy link
Member

No description provided.

@thomasfaingnaert thomasfaingnaert marked this pull request as draft November 15, 2023 13:26
@thomasfaingnaert thomasfaingnaert marked this pull request as ready for review November 15, 2023 16:08
@maleadt
Copy link
Member

maleadt commented Nov 15, 2023

Benchmark results for commit 7d1b0d7 (comparing to 8b52f30):

test master PR Δmin
WMMA GEMM Float16*Float16+Float16=Float16 (2048×2048) · (2048×2048) (TN) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 230.0 μs ± 0.764% (224.0 … 234.0 μs) 221.0 μs ± 4.33% (206.0 … 235.0 μs) -8.0% ✅
FPU GEMM Int32*Int32=Int32 (128×128) · (128×128) (TN) OP (8, 16, 2), base shape (4, 8, 1) 7.38 μs ± 1.59% (7.15 … 7.87 μs) 8.36 μs ± 2.03% (8.11 … 8.82 μs) +13.3% ❌
FPU GEMM Float32*Float32=Float64 (128×128) · (128×128) (TN) OP (8, 16, 2), base shape (4, 8, 1) 157.0 μs ± 0.666% (154.0 … 159.0 μs) 173.0 μs ± 6.91% (163.0 … 359.0 μs) +6.4% ❌
WMMA GEMM Float16*Float16+Float16=Float16 (2048×2048) · (2048×2048) (TT) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 225.0 μs ± 0.724% (221.0 … 228.0 μs) 217.0 μs ± 3.82% (205.0 … 229.0 μs) -7.4% ✅
FPU GEMM Int16*Int16=Int16 (128×128) · (128×128) (TN) OP (8, 16, 2), base shape (4, 8, 1) 8.82 μs ± 1.98% (8.58 … 9.06 μs) 9.86 μs ± 1.59% (9.54 … 10.0 μs) +11.1% ❌
WMMA Complex GEMM Float16*Float16=Float32 (128×128) · (128×128) (TN) OP (16, 16, 16) 8.53 μs ± 1.91% (8.34 … 8.82 μs) 9.64 μs ± 1.65% (9.3 … 9.78 μs) +11.4% ❌
FPU GEMM Float32*Float32=Float64 (128×128) · (128×128) (NT) OP (8, 16, 2), base shape (4, 8, 1) 156.0 μs ± 0.496% (154.0 … 159.0 μs) 179.0 μs ± 4.86% (177.0 … 341.0 μs) +15.2% ❌
Tropical GEMM Float32*Float32=Float32 (256×256) · (256×256) (TT) OP (8, 16, 2), base shape (4, 8, 1) 120.0 μs ± 0.136% (120.0 … 121.0 μs) 122.0 μs ± 5.08% (110.0 … 127.0 μs) -8.2% ✅
WMMA diagonal GEMM diag(Float16)*Float16=Float32 (128×128) · (128×128) (NN) OP (16, 16, 16) 9.93 μs ± 1.34% (9.54 … 10.3 μs) 11.3 μs ± 1.08% (11.0 … 11.7 μs) +15.0% ❌
Tropical GEMM Float32*Float32=Float32 (128×256) · (256×128) (NT) OP (8, 16, 2), base shape (4, 8, 1) 127.0 μs ± 0.142% (127.0 … 128.0 μs) 126.0 μs ± 7.8% (112.0 … 299.0 μs) -11.9% ✅
WMMA GEMM Float16*Float16+Float16=Float16 (2048×2048) · (2048×2048) (NT) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 216.0 μs ± 3.78% (205.0 … 229.0 μs) 225.0 μs ± 0.921% (221.0 … 231.0 μs) +7.7% ❌
FPU GEMM Float16*Float16=Float32 (128×128) · (128×128) (TN) OP (8, 16, 2), base shape (4, 8, 1) 6.91 μs ± 0.858% (6.44 … 7.15 μs) 7.81 μs ± 2.11% (7.63 … 8.11 μs) +18.5% ❌
FPU GEMM Int16*Int16=Int16 (128×128) · (128×128) (NT) OP (8, 16, 2), base shape (4, 8, 1) 9.56 μs ± 1.7% (9.3 … 9.78 μs) 10.9 μs ± 1.36% (10.7 … 11.2 μs) +15.4% ❌
FPU GEMM Float16*Float16=Float32 (128×128) · (128×128) (NT) OP (8, 16, 2), base shape (4, 8, 1) 9.21 μs ± 1.47% (8.82 … 9.54 μs) 10.4 μs ± 1.2% (10.0 … 10.7 μs) +13.5% ❌
FPU GEMM Int32*Int32=Int32 (128×128) · (128×128) (NT) OP (8, 16, 2), base shape (4, 8, 1) 9.42 μs ± 1.57% (9.06 … 9.78 μs) 10.7 μs ± 1.29% (10.5 … 11.0 μs) +15.8% ❌
FPU GEMM Float16*Float16=Float32 (128×128) · (128×128) (TT) OP (8, 16, 2), base shape (4, 8, 1) 8.2 μs ± 1.54% (7.87 … 8.58 μs) 9.29 μs ± 1.74% (9.06 … 9.54 μs) +15.2% ❌
WMMA Complex GEMM Float16*Float16=Float32 (128×128) · (128×128) (NT) OP (16, 16, 16) 8.44 μs ± 1.76% (8.11 … 8.82 μs) 9.5 μs ± 1.44% (9.06 … 9.78 μs) +11.8% ❌
Tropical GEMM Float32*Float32=Float32 (128×128) · (128×128) (NT) OP (8, 16, 2), base shape (4, 8, 1) 56.8 μs ± 0.284% (56.5 … 57.2 μs) 65.5 μs ± 16.9% (64.4 … 243.0 μs) +13.9% ❌
Tropical GEMM Float32*Float32=Float32 (256×256) · (256×256) (NN) OP (8, 16, 2), base shape (4, 8, 1) 110.0 μs ± 0.149% (109.0 … 110.0 μs) 126.0 μs ± 7.56% (125.0 … 335.0 μs) +13.9% ❌
WMMA Complex GEMM Float16*Float16=Float32 (128×128) · (128×128) (TT) OP (16, 16, 16) 8.35 μs ± 1.99% (8.11 … 8.82 μs) 9.47 μs ± 1.7% (9.06 … 9.78 μs) +11.8% ❌
FPU GEMM Int16*Int16=Int16 (128×128) · (128×128) (TT) OP (8, 16, 2), base shape (4, 8, 1) 8.53 μs ± 1.98% (8.34 … 8.82 μs) 9.66 μs ± 1.54% (9.3 … 9.78 μs) +11.4% ❌

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 … 15.5 μs) 7.22 μs ± 1.76% (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 ± 0.965% (15.7 … 16.2 μs) 4.19 μs ± 3.69% (4.05 … 4.53 μs) 25.8
WMMA GEMM Float16*Float16+Float16=Float16 (2048×2048) · (2048×2048) (TN) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 221.0 μs ± 4.33% (206.0 … 235.0 μs) 191.0 μs ± 1.99% (180.0 … 196.0 μs) 87.4
WMMA GEMM Float16*Float16+Float32=Float32 (128×128) · (128×128) (TN) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 7.93 μs ± 1.7% (7.63 … 8.34 μs) 2.08 μs ± 6.89% (1.91 … 2.38 μs) 25.0
WMMA GEMM Float16*Float16+Float16=Float16 (2048×2048) · (2048×2048) (NT) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 225.0 μs ± 0.921% (221.0 … 231.0 μs) 184.0 μs ± 2.81% (171.0 … 193.0 μs) 77.5
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TT) OP (16, 16, 1), base shape (8, 4, 1) 14.5 μs ± 1.06% (14.3 … 14.8 μs) 4.03 μs ± 3.27% (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 ± 3.85% (1.81 … 2.21 ms) 1.01 ms ± 2.16% (0.938 … 1.26 ms) 51.9
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TT) OP (32, 4, 1), base shape (32, 1, 1) 13.1 μs ± 0.92% (12.6 … 13.4 μs) 4.03 μs ± 3.53% (3.81 … 4.29 μs) 30.2
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NN) OP (32, 4, 1), base shape (16, 2, 1) 19.1 μs ± 0.905% (18.8 … 19.3 μs) 4.08 μs ± 2.58% (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) 1.85 ms ± 7.18% (1.78 … 4.31 ms) 291.0 μs ± 2.64% (289.0 … 462.0 μs) 16.3
FPU GEMM Float64*Float64=Float64 (2048×2048) · (2048×2048) (TN) OP (8, 16, 2), base shape (4, 8, 1) 46.1 ms ± 0.644% (46.0 … 52.4 ms) 40.8 ms ± 0.00684% (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 ± 0.896% (18.8 … 19.3 μs) 4.29 μs ± 3.57% (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) 289.0 μs ± 2.75% (255.0 … 293.0 μs) 109.0 μs ± 0.16% (109.0 … 109.0 μs) 42.7
WMMA GEMM Float16*Float16+Float16=Float16 (256×256) · (256×256) (NT) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 7.42 μs ± 1.94% (7.15 … 7.87 μs) 3.82 μs ± 3.0% (3.58 … 4.29 μs) 50.0
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NN) OP (8, 16, 2), base shape (4, 8, 1) 15.3 μs ± 1.1% (15.0 … 15.7 μs) 4.08 μs ± 2.81% (3.81 … 4.53 μs) 25.4
WMMA GEMM Float16*Float16+Float16=Float16 (512×256) · (256×512) (TT) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 8.73 μs ± 1.89% (8.34 … 9.06 μs) 5.17 μs ± 3.05% (4.77 … 5.48 μs) 57.1
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TN) OP (4, 8, 1), base shape (4, 8, 1) 15.2 μs ± 1.09% (15.0 … 15.5 μs) 4.29 μs ± 3.45% (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) 1.89 ms ± 3.52% (1.84 … 2.08 ms) 1.1 ms ± 2.98% (1.03 … 1.37 ms) 56.0
FPU GEMM Float64*Float64=Float64 (256×256) · (256×256) (NN) OP (8, 16, 2), base shape (4, 8, 1) 272.0 μs ± 6.38% (256.0 … 292.0 μs) 109.0 μs ± 0.154% (109.0 … 109.0 μs) 42.6
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TT) OP (4, 8, 1), base shape (4, 8, 1) 15.2 μs ± 1.11% (15.0 … 15.5 μs) 4.03 μs ± 3.34% (3.81 … 4.29 μs) 25.4
WMMA GEMM Float16*Float16+Float32=Float32 (2048×2048) · (2048×2048) (NN) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 360.0 μs ± 4.85% (344.0 … 390.0 μs) 292.0 μs ± 0.304% (290.0 … 294.0 μs) 84.3
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NN) OP (16, 16, 1), base shape (4, 8, 1) 15.3 μs ± 0.893% (15.0 … 15.7 μs) 4.08 μs ± 2.65% (3.81 … 4.53 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 ± 2.83% (1.43 … 1.68 ms) 1.1 ms ± 2.87% (1.04 … 1.38 ms) 72.5
WMMA GEMM Float16*Float16+Float32=Float32 (128×128) · (128×128) (NN) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 8.13 μs ± 1.84% (7.87 … 8.34 μs) 2.22 μs ± 7.47% (1.91 … 2.62 μs) 24.2
FPU GEMM Float16*Float16=Float32 (2048×2048) · (2048×2048) (NN) OP (8, 16, 2), base shape (4, 8, 1) 1.85 ms ± 2.66% (1.81 … 2.02 ms) 292.0 μs ± 0.205% (290.0 … 294.0 μs) 16.0
FPU GEMM Float64*Float64=Float64 (128×128) · (128×128) (TT) OP (8, 16, 2), base shape (4, 8, 1) 130.0 μs ± 0.418% (128.0 … 131.0 μs) 28.7 μs ± 0.453% (28.4 … 29.1 μs) 22.2
WMMA GEMM Float16*Float16+Float32=Float32 (128×128) · (128×128) (TT) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 8.06 μs ± 2.08% (7.87 … 8.34 μs) 2.18 μs ± 7.86% (1.91 … 2.62 μs) 24.2
FPU GEMM Float64*Float64=Float64 (2048×2048) · (2048×2048) (TT) OP (8, 16, 2), base shape (4, 8, 1) 46.1 ms ± 0.976% (46.0 … 52.4 ms) 40.8 ms ± 0.0061% (40.8 … 40.8 ms) 88.7
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NT) OP (32, 4, 1), base shape (32, 1, 1) 12.6 μs ± 1.33% (12.4 … 12.9 μs) 4.19 μs ± 3.68% (4.05 … 4.53 μs) 32.7
WMMA GEMM Float16*Float16+Float32=Float32 (2048×2048) · (2048×2048) (NT) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 380.0 μs ± 4.86% (350.0 … 397.0 μs) 291.0 μs ± 0.243% (290.0 … 293.0 μs) 82.7
FPU GEMM Float16*Float16=Float32 (256×256) · (256×256) (TN) OP (8, 16, 2), base shape (4, 8, 1) 13.0 μs ± 1.03% (12.6 … 13.1 μs) 3.8 μs ± 3.21% (3.58 … 4.29 μs) 28.3
WMMA GEMM Float16*Float16+Float16=Float16 (256×256) · (256×256) (NN) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 8.43 μs ± 1.91% (8.11 … 8.82 μs) 3.69 μs ± 3.8% (3.58 … 4.05 μs) 44.1
WMMA GEMM Float16*Float16+Float32=Float32 (2048×2048) · (2048×2048) (TT) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 362.0 μs ± 4.89% (343.0 … 390.0 μs) 292.0 μs ± 0.246% (290.0 … 293.0 μs) 84.6
FPU GEMM Float16*Float16=Float32 (2048×2048) · (2048×2048) (NT) OP (8, 16, 2), base shape (4, 8, 1) 2.3 ms ± 2.97% (2.26 … 2.56 ms) 291.0 μs ± 0.277% (289.0 … 296.0 μs) 12.8
FPU GEMM Float16*Float16=Float32 (128×128) · (128×128) (TN) OP (8, 16, 2), base shape (4, 8, 1) 7.81 μs ± 2.11% (7.63 … 8.11 μs) 2.33 μs ± 5.85% (2.15 … 2.62 μs) 28.1
WMMA GEMM Float16*Float16+Float32=Float32 (256×128) · (128×256) (TN) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 9.04 μs ± 1.9% (8.82 … 9.3 μs) 2.89 μs ± 5.01% (2.62 … 3.1 μs) 29.7
WMMA GEMM Float16*Float16+Float32=Float32 (128×256) · (256×128) (TN) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 12.3 μs ± 1.16% (12.2 … 12.6 μs) 2.91 μs ± 5.19% (2.62 … 3.1 μs) 21.6
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NT) OP (4, 8, 2), base shape (4, 8, 1) 16.0 μs ± 1.06% (15.7 … 16.5 μs) 4.17 μs ± 3.67% (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 ± 0.981% (16.0 … 16.7 μs) 4.18 μs ± 3.69% (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 ± 0.466% (31.5 … 31.9 μs) 4.08 μs ± 2.88% (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 ± 0.918% (13.1 … 13.8 μs) 4.03 μs ± 3.41% (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 ± 1.05% (15.5 … 16.0 μs) 4.09 μs ± 2.74% (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 ± 0.848% (17.9 … 18.4 μs) 3.9 μs ± 3.57% (3.58 … 4.29 μs) 20.0
WMMA GEMM Float16*Float16+Float16=Float16 (2048×2048) · (2048×2048) (TT) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 217.0 μs ± 3.82% (205.0 … 229.0 μs) 188.0 μs ± 1.86% (178.0 … 195.0 μs) 87.0
WMMA GEMM Float16*Float16+Float16=Float16 (2048×2048) · (2048×2048) (NN) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 220.0 μs ± 4.42% (205.0 … 237.0 μs) 184.0 μs ± 2.86% (175.0 … 190.0 μs) 85.4
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TN) OP (4, 32, 1), base shape (2, 16, 1) 21.9 μs ± 0.733% (21.7 … 22.2 μs) 4.3 μs ± 3.31% (4.05 … 4.53 μs) 18.7
WMMA GEMM Float16*Float16+Float32=Float32 (128×256) · (256×128) (NN) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 12.2 μs ± 1.44% (11.9 … 12.4 μs) 3.1 μs ± 3.65% (2.86 … 3.34 μs) 24.0
FPU GEMM Float64*Float64=Float64 (128×128) · (128×128) (NT) OP (8, 16, 2), base shape (4, 8, 1) 146.0 μs ± 4.38% (129.0 … 149.0 μs) 28.6 μs ± 0.506% (28.4 … 29.1 μs) 21.9
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NN) OP (16, 16, 1), base shape (8, 4, 1) 14.5 μs ± 1.08% (14.3 … 14.8 μs) 4.08 μs ± 2.79% (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 ± 1.42% (12.2 … 12.9 μs) 4.03 μs ± 3.64% (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 ± 1.1% (15.3 … 15.7 μs) 4.03 μs ± 3.53% (3.81 … 4.29 μs) 25.0
WMMA GEMM Float16*Float16+Float16=Float16 (256×512) · (512×256) (NN) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 12.0 μs ± 1.35% (11.7 … 12.4 μs) 5.18 μs ± 3.09% (5.01 … 5.48 μs) 42.9
WMMA GEMM Float16*Float16+Float16=Float16 (512×512) · (512×512) (NT) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 12.4 μs ± 1.49% (11.9 … 12.6 μs) 7.25 μs ± 1.86% (6.91 … 7.63 μs) 58.0
WMMA GEMM Float16*Float16+Float16=Float16 (512×256) · (256×512) (NN) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 8.65 μs ± 2.11% (8.34 … 9.06 μs) 4.95 μs ± 3.48% (4.77 … 5.25 μs) 57.1
FPU GEMM Float64*Float64=Float64 (128×128) · (128×128) (NN) OP (8, 16, 2), base shape (4, 8, 1) 145.0 μs ± 2.55% (140.0 … 149.0 μs) 31.0 μs ± 0.319% (30.5 … 31.2 μs) 21.8
WMMA GEMM Float16*Float16+Float32=Float32 (256×256) · (256×256) (TN) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 12.4 μs ± 1.39% (12.2 … 12.6 μs) 3.8 μs ± 3.22% (3.58 … 4.29 μs) 29.4
WMMA GEMM Float16*Float16+Float32=Float32 (256×256) · (256×256) (TT) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 12.2 μs ± 1.26% (11.9 … 12.4 μs) 3.95 μs ± 3.98% (3.58 … 4.29 μs) 30.0
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TT) OP (4, 32, 1), base shape (2, 16, 1) 21.9 μs ± 0.772% (21.7 … 22.2 μs) 4.03 μs ± 3.38% (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 ± 0.865% (18.8 … 19.3 μs) 4.03 μs ± 3.58% (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) 11.8 μs ± 1.39% (11.7 … 12.2 μs) 4.29 μs ± 3.54% (4.05 … 4.53 μs) 34.7
WMMA GEMM Float16*Float16+Float16=Float16 (256×512) · (512×256) (TT) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 12.2 μs ± 1.42% (11.9 … 12.4 μs) 5.19 μs ± 3.19% (4.77 … 5.48 μs) 40.0
WMMA GEMM Float16*Float16+Float16=Float16 (512×256) · (256×512) (TN) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 8.7 μs ± 2.05% (8.34 … 9.06 μs) 5.14 μs ± 2.8% (4.77 … 5.48 μs) 57.1
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TN) OP (4, 16, 1), base shape (4, 8, 1) 15.2 μs ± 1.11% (15.0 … 15.5 μs) 4.3 μs ± 3.46% (4.05 … 4.53 μs) 27.0
WMMA GEMM Float16*Float16+Float32=Float32 (256×256) · (256×256) (NN) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 12.3 μs ± 1.21% (11.9 … 12.4 μs) 3.96 μs ± 3.97% (3.81 … 4.29 μs) 32.0
FPU GEMM Float64*Float64=Float64 (2048×2048) · (2048×2048) (NT) OP (8, 16, 2), base shape (4, 8, 1) 46.4 ms ± 0.757% (46.4 … 52.8 ms) 40.8 ms ± 0.00524% (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 ± 0.803% (15.0 … 15.7 μs) 4.09 μs ± 2.69% (3.81 … 4.29 μs) 25.4
WMMA GEMM Float16*Float16+Float16=Float16 (512×256) · (256×512) (NT) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 8.78 μs ± 2.08% (8.34 … 9.06 μs) 4.93 μs ± 3.29% (4.77 … 5.25 μs) 57.1
WMMA GEMM Float16*Float16+Float16=Float16 (512×512) · (512×512) (TN) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 12.4 μs ± 1.43% (12.2 … 12.9 μs) 7.42 μs ± 1.52% (7.15 … 7.87 μs) 58.8
WMMA GEMM Float16*Float16+Float32=Float32 (2048×2048) · (2048×2048) (TN) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 375.0 μs ± 4.96% (350.0 … 396.0 μs) 293.0 μs ± 0.235% (290.0 … 295.0 μs) 83.0
WMMA GEMM Float16*Float16+Float16=Float16 (512×512) · (512×512) (NN) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 12.3 μs ± 1.4% (11.9 … 12.6 μs) 7.28 μs ± 1.97% (7.15 … 7.63 μs) 60.0
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NN) OP (4, 8, 1), base shape (4, 8, 1) 15.2 μs ± 1.15% (15.0 … 15.7 μs) 4.08 μs ± 2.58% (3.81 … 4.53 μs) 25.4
FPU GEMM Float16*Float16=Float32 (128×128) · (128×128) (NT) OP (8, 16, 2), base shape (4, 8, 1) 10.4 μs ± 1.2% (10.0 … 10.7 μs) 2.3 μs ± 5.34% (1.91 … 2.62 μs) 19.0
WMMA GEMM Float16*Float16+Float32=Float32 (256×128) · (128×256) (NN) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 9.29 μs ± 1.71% (9.06 … 9.54 μs) 2.94 μs ± 5.26% (2.62 … 3.1 μs) 28.9
WMMA GEMM Float16*Float16+Float32=Float32 (128×128) · (128×128) (NT) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 8.03 μs ± 2.07% (7.87 … 8.34 μs) 2.31 μs ± 5.3% (1.91 … 2.62 μs) 24.2
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NT) OP (16, 16, 1), base shape (8, 4, 1) 14.6 μs ± 0.932% (14.3 … 15.0 μs) 4.19 μs ± 3.86% (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 ± 0.428% (32.7 … 33.4 μs) 4.03 μs ± 3.51% (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 ± 1.15% (15.0 … 15.5 μs) 4.03 μs ± 3.47% (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 ± 0.775% (18.8 … 19.3 μs) 4.18 μs ± 3.69% (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 ± 6.92% (2.32 … 6.08 ms) 864.0 μs ± 2.04% (819.0 … 1080.0 μs) 35.3
FPU GEMM Float64*Float64=Float64 (256×256) · (256×256) (TT) OP (8, 16, 2), base shape (4, 8, 1) 277.0 μs ± 6.38% (256.0 … 295.0 μs) 109.0 μs ± 0.148% (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 ± 1.01% (15.0 … 15.5 μs) 4.03 μs ± 3.49% (3.81 … 4.29 μs) 25.4
WMMA GEMM Float16*Float16+Float16=Float16 (256×256) · (256×256) (TT) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 7.47 μs ± 1.78% (7.15 … 7.87 μs) 3.63 μs ± 3.08% (3.58 … 4.05 μs) 50.0
WMMA GEMM Float16*Float16+Float32=Float32 (128×256) · (256×128) (TT) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 12.2 μs ± 1.42% (11.9 … 12.4 μs) 2.99 μs ± 4.86% (2.62 … 3.1 μs) 22.0
FPU GEMM Float16*Float16=Float32 (128×128) · (128×128) (TT) OP (8, 16, 2), base shape (4, 8, 1) 9.29 μs ± 1.74% (9.06 … 9.54 μs) 2.46 μs ± 6.63% (2.15 … 2.62 μs) 23.7
WMMA GEMM Float16*Float16+Float16=Float16 (512×512) · (512×512) (TT) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 12.5 μs ± 1.37% (12.2 … 12.9 μs) 7.29 μs ± 1.98% (7.15 … 7.63 μs) 58.8
WMMA GEMM Float16*Float16+Float32=Float32 (256×256) · (256×256) (NT) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 12.4 μs ± 1.4% (12.2 … 12.6 μs) 3.88 μs ± 3.51% (3.58 … 4.29 μs) 29.4
FPU GEMM Float32*Float32=Float32 (256×256) · (256×256) (NN) OP (8, 16, 2), base shape (4, 8, 1) 15.3 μs ± 1.11% (15.0 … 15.5 μs) 6.77 μs ± 2.01% (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 ± 1.08% (15.3 … 15.7 μs) 4.18 μs ± 3.61% (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 ± 5.86% (256.0 … 294.0 μs) 109.0 μs ± 0.159% (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 ± 0.958% (15.0 … 15.7 μs) 4.19 μs ± 3.69% (4.05 … 4.53 μs) 27.0
WMMA GEMM Float16*Float16+Float32=Float32 (256×128) · (128×256) (TT) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 9.19 μs ± 1.54% (8.82 … 9.54 μs) 2.88 μs ± 4.94% (2.62 … 3.1 μs) 29.7
FPU GEMM Float16*Float16=Float32 (128×128) · (128×128) (NN) OP (8, 16, 2), base shape (4, 8, 1) 9.22 μs ± 1.35% (8.82 … 9.54 μs) 2.22 μs ± 7.35% (1.91 … 2.38 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) 4.29 μs ± 3.56% (4.05 … 4.53 μs) 27.0
FPU GEMM Float16*Float16=Float32 (256×256) · (256×256) (TT) OP (8, 16, 2), base shape (4, 8, 1) 15.8 μs ± 1.04% (15.5 … 16.0 μs) 3.97 μs ± 3.94% (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 ± 0.928% (15.0 … 15.5 μs) 4.19 μs ± 3.73% (4.05 … 4.53 μs) 27.0
WMMA GEMM Float16*Float16+Float32=Float32 (128×256) · (256×128) (NT) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 12.4 μs ± 1.36% (12.2 … 12.6 μs) 2.76 μs ± 5.13% (2.38 … 3.1 μs) 19.6
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NN) OP (4, 8, 2), base shape (4, 8, 1) 15.0 μs ± 1.09% (14.5 … 15.5 μs) 4.09 μs ± 2.77% (3.81 … 4.29 μs) 26.2
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TN) OP (32, 4, 1), base shape (32, 1, 1) 31.9 μs ± 21.7% (31.5 … 189.0 μs) 4.3 μs ± 3.4% (4.05 … 4.53 μs) 12.9
WMMA GEMM Float16*Float16+Float32=Float32 (256×128) · (128×256) (NT) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 9.17 μs ± 1.68% (8.82 … 9.3 μs) 2.9 μs ± 5.1% (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) 23.2 μs ± 175.0% (21.7 … 1310.0 μs) 4.19 μs ± 3.71% (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) 3.96 μs ± 3.86% (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) 1.55 ms ± 3.33% (1.51 … 1.87 ms) 293.0 μs ± 3.99% (290.0 … 466.0 μs) 19.3
FPU GEMM Float64*Float64=Float64 (2048×2048) · (2048×2048) (NN) OP (8, 16, 2), base shape (4, 8, 1) 46.3 ms ± 0.872% (46.3 … 52.7 ms) 40.8 ms ± 0.00606% (40.8 … 40.8 ms) 88.2
WMMA GEMM Float16*Float16+Float16=Float16 (256×512) · (512×256) (TN) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 12.1 μs ± 1.37% (11.9 … 12.4 μs) 5.11 μs ± 3.44% (4.77 … 5.72 μs) 40.0
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NN) OP (4, 32, 1), base shape (2, 16, 1) 22.0 μs ± 0.628% (21.7 … 22.2 μs) 4.08 μs ± 2.74% (3.81 … 4.29 μs) 17.6
WMMA GEMM Float16*Float16+Float16=Float16 (256×512) · (512×256) (NT) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 12.0 μs ± 1.36% (11.7 … 12.4 μs) 4.49 μs ± 3.66% (4.29 … 4.77 μs) 36.7
FPU GEMM Float32*Float32=Float32 (256×256) · (256×256) (NT) OP (8, 16, 2), base shape (4, 8, 1) 18.2 μs ± 0.797% (17.9 … 18.6 μs) 6.78 μs ± 2.03% (6.2 … 6.91 μs) 34.7
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NN) OP (4, 16, 1), base shape (4, 8, 1) 15.2 μs ± 1.12% (15.0 … 15.5 μs) 4.08 μs ± 2.84% (3.81 … 4.53 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) 4.18 μs ± 3.69% (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 ± 0.456% (32.7 … 33.4 μs) 4.29 μs ± 3.55% (4.05 … 4.77 μs) 12.4
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TN) OP (8, 16, 2), base shape (4, 8, 1) 12.1 μs ± 1.14% (11.9 … 12.4 μs) 4.3 μs ± 3.39% (4.05 … 4.53 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 ± 0.278% (129.0 … 131.0 μs) 28.7 μs ± 0.462% (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 ± 1.12% (15.0 … 15.5 μs) 4.29 μs ± 3.55% (4.05 … 4.53 μs) 27.0
WMMA GEMM Float16*Float16+Float16=Float16 (256×256) · (256×256) (TN) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 7.44 μs ± 1.55% (7.15 … 7.87 μs) 3.6 μs ± 2.84% (3.1 … 4.05 μs) 43.3
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TN) OP (16, 16, 1), base shape (8, 4, 1) 14.5 μs ± 1.05% (14.3 … 14.8 μs) 4.3 μs ± 3.45% (4.05 … 4.53 μs) 28.3
FPU GEMM Float32*Float32=Float32 (256×256) · (256×256) (TN) OP (8, 16, 2), base shape (4, 8, 1) 12.6 μs ± 1.34% (12.4 … 12.9 μs) 7.17 μs ± 1.16% (6.91 … 7.63 μs) 55.8

@codecov Codecov
Copy link

codecov bot commented Nov 15, 2023

Codecov Report

Attention: 6 lines in your changes are missing coverage. Please review.

Comparison is base (8b52f30) 34.16% compared to head (7d1b0d7) 34.23%.

Files Patch % Lines
src/layout.jl 25.00% 6 Missing ⚠️
Additional details and impacted files 
@@            Coverage Diff             @@
##           master     #176      +/-   ##
==========================================
+ Coverage   34.16%   34.23%   +0.07%     
==========================================
  Files          11       11              
  Lines         925      923       -2     
==========================================
  Hits          316      316              
+ Misses        609      607       -2

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

@thomasfaingnaert thomasfaingnaert merged commit 6a8e8cb into master Nov 15, 2023
1 check passed
@thomasfaingnaert thomasfaingnaert deleted the tf/generalise-layout branch November 15, 2023 16:17
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers
No reviews
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