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

Check number of threads before launching kernel #174

Merged
merged 1 commit into from
Nov 15, 2023
Merged

Check number of threads before launching kernel #174

merged 1 commit into from
Nov 15, 2023

Conversation

thomasfaingnaert
Copy link
Member

No description provided.

@maleadt
Copy link
Member

maleadt commented Nov 15, 2023

Benchmark results for commit 2f556ce (comparing to 912fa05):

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 ± 0.731% (225.0 … 235.0 μs) 226.0 μs ± 3.66% (207.0 … 235.0 μs) -8.1% ✅
FPU GEMM Int32*Int32=Int32 (128×128) · (128×128) (TN) OP (8, 16, 2), base shape (4, 8, 1) 7.39 μs ± 1.61% (7.15 … 7.63 μs) 8.36 μs ± 2.06% (8.11 … 8.82 μs) +13.3% ❌
WMMA GEMM Float16*Float16+Float16=Float16 (2048×2048) · (2048×2048) (NN) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 221.0 μs ± 4.43% (207.0 … 239.0 μs) 231.0 μs ± 1.06% (224.0 … 238.0 μs) +8.3% ❌
WMMA GEMM Float16*Float16+Float16=Float16 (2048×2048) · (2048×2048) (TT) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 220.0 μs ± 3.45% (207.0 … 228.0 μs) 225.0 μs ± 0.735% (221.0 … 228.0 μs) +6.9% ❌
FPU GEMM Float64*Float64=Float64 (128×128) · (128×128) (NN) OP (8, 16, 2), base shape (4, 8, 1) 144.0 μs ± 4.94% (129.0 … 149.0 μs) 148.0 μs ± 0.144% (147.0 … 149.0 μs) +13.8% ❌
FPU GEMM Int16*Int16=Int16 (128×128) · (128×128) (TN) OP (8, 16, 2), base shape (4, 8, 1) 8.73 μs ± 1.64% (8.34 … 9.06 μs) 9.87 μs ± 1.59% (9.54 … 10.0 μs) +14.3% ❌
Tropical GEMM Float32*Float32=Float32 (256×256) · (256×256) (TT) OP (8, 16, 2), base shape (4, 8, 1) 110.0 μs ± 0.144% (110.0 … 111.0 μs) 125.0 μs ± 0.119% (125.0 … 126.0 μs) +14.1% ❌
WMMA GEMM+bias Float16*Float16+Float32=Float32 (128×128) · (128×128) (TT) OP (16, 16, 16) 10.3 μs ± 1.76% (9.78 … 11.0 μs) 9.14 μs ± 2.26% (8.82 … 9.54 μs) -9.8% ✅
WMMA diagonal GEMM diag(Float16)*Float16=Float32 (128×128) · (128×128) (NN) OP (16, 16, 16) 11.2 μs ± 1.14% (11.0 … 11.7 μs) 9.94 μs ± 1.3% (9.54 … 10.0 μs) -13.0% ✅
WMMA GEMM+bias Float16*Float16+Float32=Float32 (128×128) · (128×128) (NT) OP (16, 16, 16) 9.05 μs ± 1.8% (8.82 … 9.3 μs) 10.3 μs ± 1.12% (9.78 … 10.7 μs) +10.8% ❌
FPU GEMM Float64*Float64=Float64 (128×128) · (128×128) (TN) OP (8, 16, 2), base shape (4, 8, 1) 130.0 μs ± 0.272% (129.0 … 131.0 μs) 148.0 μs ± 4.66% (147.0 … 306.0 μs) +13.5% ❌
WMMA GEMM Float16*Float16+Float16=Float16 (256×256) · (256×256) (TN) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 8.42 μs ± 1.68% (8.11 … 8.82 μs) 7.43 μs ± 1.51% (7.15 … 7.87 μs) -11.8% ✅
Tropical GEMM Float32*Float32=Float32 (128×256) · (256×128) (TT) OP (8, 16, 2), base shape (4, 8, 1) 122.0 μs ± 5.07% (109.0 … 126.0 μs) 125.0 μs ± 0.128% (125.0 … 126.0 μs) +13.9% ❌
Tropical GEMM Float32*Float32=Float32 (256×256) · (256×256) (NT) OP (8, 16, 2), base shape (4, 8, 1) 127.0 μs ± 0.137% (127.0 … 128.0 μs) 112.0 μs ± 0.148% (112.0 … 113.0 μs) -11.8% ✅
FPU GEMM Int16*Int16=Int16 (128×128) · (128×128) (NT) OP (8, 16, 2), base shape (4, 8, 1) 9.56 μs ± 1.68% (9.3 … 9.78 μs) 10.8 μs ± 1.26% (10.7 … 11.2 μs) +15.4% ❌
FPU GEMM Int32*Int32=Int32 (128×128) · (128×128) (NT) OP (8, 16, 2), base shape (4, 8, 1) 9.51 μs ± 1.49% (9.3 … 9.78 μs) 10.7 μs ± 1.32% (10.5 … 11.0 μs) +12.8% ❌
WMMA Complex GEMM Float16*Float16=Float32 (128×128) · (128×128) (NT) OP (16, 16, 16) 8.44 μs ± 1.8% (8.11 … 8.82 μs) 9.62 μs ± 1.67% (9.3 … 10.0 μs) +14.7% ❌
FPU GEMM Int16*Int16=Int16 (128×128) · (128×128) (TT) OP (8, 16, 2), base shape (4, 8, 1) 8.54 μs ± 1.96% (8.34 … 8.82 μs) 9.66 μs ± 1.55% (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 ± 1.16% (14.8 … 15.3 μs) 7.23 μs ± 1.79% (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 ± 0.977% (15.7 … 16.2 μs) 4.19 μs ± 3.74% (4.05 … 4.77 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) 226.0 μs ± 3.66% (207.0 … 235.0 μs) 190.0 μs ± 1.76% (180.0 … 194.0 μs) 87.1
WMMA GEMM Float16*Float16+Float32=Float32 (128×128) · (128×128) (TN) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 8.96 μs ± 1.57% (8.58 … 9.54 μs) 2.34 μs ± 6.06% (1.91 … 2.62 μs) 22.2
WMMA GEMM Float16*Float16+Float16=Float16 (2048×2048) · (2048×2048) (NT) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 218.0 μs ± 3.7% (203.0 … 228.0 μs) 184.0 μs ± 2.8% (172.0 … 192.0 μs) 84.8
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) 4.04 μs ± 3.36% (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) 1.85 ms ± 3.25% (1.8 … 2.23 ms) 1.01 ms ± 2.1% (0.933 … 1.22 ms) 51.8
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TT) OP (32, 4, 1), base shape (32, 1, 1) 13.1 μs ± 0.948% (12.6 … 13.4 μs) 4.04 μs ± 3.41% (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 ± 0.892% (18.8 … 19.3 μs) 4.09 μs ± 2.43% (3.81 … 4.53 μs) 20.3
FPU GEMM Float16*Float16=Float32 (2048×2048) · (2048×2048) (TT) OP (8, 16, 2), base shape (4, 8, 1) 1.86 ms ± 3.6% (1.8 … 2.22 ms) 293.0 μs ± 5.96% (290.0 … 509.0 μs) 16.1
FPU GEMM Float64*Float64=Float64 (2048×2048) · (2048×2048) (TN) OP (8, 16, 2), base shape (4, 8, 1) 46.2 ms ± 1.26% (46.0 … 55.2 ms) 41.1 ms ± 2.73% (40.8 … 56.9 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 ± 0.936% (18.8 … 19.3 μs) 4.3 μs ± 3.47% (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) 287.0 μs ± 5.48% (255.0 … 460.0 μs) 109.0 μs ± 0.158% (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) 7.45 μs ± 1.88% (7.15 … 7.87 μs) 3.8 μs ± 3.22% (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 ± 1.14% (15.0 … 15.7 μs) 4.09 μs ± 2.39% (3.81 … 4.29 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) 8.72 μs ± 1.85% (8.34 … 9.06 μs) 5.17 μs ± 3.05% (5.01 … 5.48 μs) 60.0
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TN) OP (4, 8, 1), base shape (4, 8, 1) 15.2 μs ± 1.07% (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 ± 6.1% (1.83 … 4.16 ms) 1.1 ms ± 3.32% (1.03 … 1.63 ms) 56.5
FPU GEMM Float64*Float64=Float64 (256×256) · (256×256) (NN) OP (8, 16, 2), base shape (4, 8, 1) 280.0 μs ± 5.76% (256.0 … 292.0 μs) 109.0 μs ± 0.157% (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) 4.05 μs ± 3.41% (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) 375.0 μs ± 5.06% (345.0 … 612.0 μs) 293.0 μs ± 4.52% (290.0 … 483.0 μs) 84.2
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NN) OP (16, 16, 1), base shape (4, 8, 1) 15.3 μs ± 0.887% (15.0 … 15.7 μs) 4.09 μs ± 2.51% (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 ± 6.19% (1.43 … 3.81 ms) 1.1 ms ± 2.42% (1.04 … 1.36 ms) 72.4
WMMA GEMM Float16*Float16+Float32=Float32 (128×128) · (128×128) (NN) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 9.18 μs ± 1.65% (8.82 … 9.54 μs) 2.23 μs ± 7.33% (1.91 … 2.38 μs) 21.6
FPU GEMM Float16*Float16=Float32 (2048×2048) · (2048×2048) (NN) OP (8, 16, 2), base shape (4, 8, 1) 1.86 ms ± 3.18% (1.82 … 2.02 ms) 293.0 μs ± 0.237% (291.0 … 294.0 μs) 16.0
FPU GEMM Float64*Float64=Float64 (128×128) · (128×128) (TT) OP (8, 16, 2), base shape (4, 8, 1) 145.0 μs ± 4.79% (129.0 … 149.0 μs) 28.7 μs ± 0.448% (28.4 … 29.1 μs) 22.0
WMMA GEMM Float16*Float16+Float32=Float32 (128×128) · (128×128) (TT) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 9.07 μs ± 1.96% (8.82 … 9.54 μs) 2.46 μs ± 6.6% (2.15 … 2.62 μs) 24.3
FPU GEMM Float64*Float64=Float64 (2048×2048) · (2048×2048) (TT) OP (8, 16, 2), base shape (4, 8, 1) 46.3 ms ± 2.34% (46.0 … 61.6 ms) 40.9 ms ± 0.55% (40.8 … 44.1 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 ± 1.26% (12.4 … 12.9 μs) 4.2 μs ± 3.76% (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) 372.0 μs ± 5.96% (350.0 … 547.0 μs) 292.0 μs ± 4.39% (289.0 … 531.0 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 ± 1.17% (12.6 … 13.4 μs) 3.8 μs ± 3.17% (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) 8.44 μs ± 1.89% (8.11 … 8.82 μs) 3.72 μs ± 3.77% (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) 375.0 μs ± 4.45% (344.0 … 390.0 μs) 292.0 μs ± 0.303% (290.0 … 294.0 μs) 84.3
FPU GEMM Float16*Float16=Float32 (2048×2048) · (2048×2048) (NT) OP (8, 16, 2), base shape (4, 8, 1) 2.35 ms ± 11.3% (2.26 … 5.6 ms) 291.0 μs ± 0.337% (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 ± 2.2% (7.63 … 8.11 μs) 2.33 μs ± 5.89% (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 ± 1.85% (8.82 … 9.3 μs) 2.89 μs ± 4.9% (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 ± 1.13% (12.2 … 12.6 μs) 2.91 μs ± 5.37% (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 … 16.2 μs) 4.2 μs ± 3.87% (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 ± 0.996% (16.0 … 16.7 μs) 4.2 μs ± 3.64% (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 ± 0.472% (31.5 … 31.9 μs) 4.09 μs ± 2.63% (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 ± 0.96% (13.1 … 13.8 μs) 4.04 μs ± 3.35% (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 ± 1.07% (15.5 … 16.0 μs) 4.1 μs ± 2.62% (3.81 … 4.53 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 ± 0.84% (17.9 … 18.4 μs) 3.92 μs ± 3.79% (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) 225.0 μs ± 0.735% (221.0 … 228.0 μs) 187.0 μs ± 2.54% (178.0 … 194.0 μs) 80.7
WMMA GEMM Float16*Float16+Float16=Float16 (2048×2048) · (2048×2048) (NN) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 231.0 μs ± 1.06% (224.0 … 238.0 μs) 185.0 μs ± 2.46% (175.0 … 190.0 μs) 78.2
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TN) OP (4, 32, 1), base shape (2, 16, 1) 21.9 μs ± 0.748% (21.7 … 22.2 μs) 4.3 μs ± 3.32% (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) 12.1 μs ± 1.4% (11.9 … 12.4 μs) 3.08 μs ± 3.63% (2.62 … 3.34 μs) 22.0
FPU GEMM Float64*Float64=Float64 (128×128) · (128×128) (NT) OP (8, 16, 2), base shape (4, 8, 1) 130.0 μs ± 0.245% (129.0 … 131.0 μs) 28.6 μs ± 0.522% (28.4 … 28.8 μs) 22.0
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NN) OP (16, 16, 1), base shape (8, 4, 1) 14.5 μs ± 1.09% (14.3 … 14.8 μs) 4.09 μs ± 2.55% (3.81 … 4.53 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 ± 1.4% (12.2 … 12.6 μs) 4.03 μs ± 3.35% (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 ± 1.07% (15.3 … 15.7 μs) 4.03 μs ± 3.36% (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 ± 1.4% (11.7 … 12.4 μs) 5.24 μs ± 3.32% (5.01 … 5.72 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) 12.3 μs ± 1.48% (11.9 … 12.6 μs) 7.25 μs ± 1.9% (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) 8.71 μs ± 2.16% (8.34 … 9.06 μs) 5.0 μs ± 3.52% (4.77 … 5.48 μs) 57.1
FPU GEMM Float64*Float64=Float64 (128×128) · (128×128) (NN) OP (8, 16, 2), base shape (4, 8, 1) 148.0 μs ± 0.144% (147.0 … 149.0 μs) 32.6 μs ± 0.536% (32.4 … 32.9 μs) 22.0
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 ± 1.38% (12.2 … 12.6 μs) 3.81 μs ± 3.31% (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 ± 1.36% (11.9 … 12.6 μs) 3.98 μs ± 3.76% (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 ± 0.762% (21.7 … 22.2 μs) 4.04 μs ± 3.31% (3.81 … 4.53 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 ± 0.853% (18.8 … 19.3 μs) 4.04 μs ± 3.42% (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 ± 1.35% (11.7 … 12.2 μs) 4.29 μs ± 3.43% (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 ± 1.41% (11.9 … 12.4 μs) 5.18 μs ± 3.17% (5.01 … 5.48 μs) 42.0
WMMA GEMM Float16*Float16+Float16=Float16 (512×256) · (256×512) (TN) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 8.72 μs ± 2.01% (8.34 … 9.06 μs) 5.13 μs ± 2.86% (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 ± 1.09% (15.0 … 15.5 μs) 4.31 μs ± 3.38% (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) 12.2 μs ± 1.27% (11.9 … 12.4 μs) 3.96 μs ± 3.84% (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 ± 0.912% (46.4 … 52.8 ms) 41.1 ms ± 2.88% (40.8 … 57.6 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 ± 0.849% (15.0 … 15.7 μs) 4.1 μs ± 2.43% (3.81 … 4.53 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) 8.77 μs ± 2.1% (8.34 … 9.06 μs) 4.92 μs ± 3.26% (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 ± 1.42% (12.2 … 12.9 μs) 7.4 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) 381.0 μs ± 4.49% (350.0 … 396.0 μs) 293.0 μs ± 0.303% (291.0 … 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 ± 1.37% (11.9 … 12.6 μs) 7.33 μs ± 1.94% (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 ± 1.12% (15.0 … 15.5 μs) 4.09 μs ± 2.49% (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) 9.21 μs ± 1.49% (8.82 … 9.54 μs) 2.31 μs ± 5.06% (1.91 … 2.62 μs) 21.6
WMMA GEMM Float16*Float16+Float32=Float32 (256×128) · (128×256) (NN) Block (128, 128, 64) Warps (4, 2) OP (16, 16, 16) 9.28 μs ± 1.73% (9.06 … 9.54 μs) 2.94 μs ± 5.11% (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) 9.11 μs ± 1.91% (8.82 … 9.3 μs) 2.29 μs ± 5.18% (1.91 … 2.62 μs) 21.6
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NT) OP (16, 16, 1), base shape (8, 4, 1) 14.6 μs ± 0.877% (14.3 … 14.8 μs) 4.19 μs ± 3.64% (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 ± 0.419% (32.7 … 33.4 μs) 4.03 μs ± 3.37% (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) 4.05 μs ± 3.28% (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 ± 0.785% (18.8 … 19.3 μs) 4.19 μs ± 3.72% (4.05 … 4.77 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 ± 3.87% (2.32 … 2.66 ms) 865.0 μs ± 1.22% (817.0 … 879.0 μs) 35.2
FPU GEMM Float64*Float64=Float64 (256×256) · (256×256) (TT) OP (8, 16, 2), base shape (4, 8, 1) 283.0 μs ± 5.62% (256.0 … 294.0 μs) 109.0 μs ± 0.147% (109.0 … 109.0 μs) 42.5
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (TT) OP (8, 8, 1), base shape (4, 8, 1) 15.3 μs ± 0.998% (15.0 … 15.5 μs) 4.04 μs ± 3.33% (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) 8.47 μs ± 1.84% (8.11 … 8.82 μs) 3.62 μs ± 2.8% (3.58 … 4.05 μs) 44.1
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) 3.0 μs ± 4.75% (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 ± 1.76% (9.06 … 9.54 μs) 2.45 μs ± 6.78% (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 ± 1.44% (12.2 … 12.9 μs) 7.29 μs ± 2.03% (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 ± 1.41% (12.2 … 12.6 μs) 3.91 μs ± 3.63% (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 ± 1.1% (15.0 … 15.7 μs) 6.81 μs ± 2.04% (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) 4.19 μs ± 3.59% (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) 284.0 μs ± 5.09% (256.0 … 293.0 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 ± 0.919% (15.0 … 15.7 μs) 4.21 μs ± 3.76% (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 ± 1.5% (8.82 … 9.54 μs) 2.89 μs ± 5.15% (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) 9.21 μs ± 1.34% (8.82 … 9.54 μs) 2.23 μs ± 7.4% (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 ± 1.11% (15.0 … 15.5 μs) 4.29 μs ± 3.45% (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 ± 1.06% (15.5 … 16.0 μs) 3.96 μs ± 3.81% (3.58 … 4.53 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 ± 0.969% (15.0 … 15.7 μs) 4.19 μs ± 3.72% (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) 2.75 μs ± 5.36% (2.38 … 2.86 μs) 19.6
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NN) OP (4, 8, 2), base shape (4, 8, 1) 14.9 μs ± 1.09% (14.8 … 15.5 μs) 4.09 μs ± 2.43% (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 ± 0.46% (31.5 … 31.9 μs) 4.31 μs ± 3.28% (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) 9.16 μs ± 1.75% (8.82 … 9.3 μs) 2.91 μs ± 5.33% (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 ± 0.772% (21.7 … 22.2 μs) 4.2 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 ± 1.11% (15.5 … 16.0 μs) 3.96 μs ± 3.81% (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 ± 3.47% (1.52 … 1.92 ms) 296.0 μs ± 25.1% (290.0 … 2590.0 μs) 19.1
FPU GEMM Float64*Float64=Float64 (2048×2048) · (2048×2048) (NN) OP (8, 16, 2), base shape (4, 8, 1) 46.3 ms ± 0.637% (46.3 … 52.7 ms) 40.8 ms ± 0.0322% (40.8 … 41.0 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 ± 1.35% (11.9 … 12.4 μs) 5.11 μs ± 3.4% (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 ± 0.625% (21.7 … 22.2 μs) 4.09 μs ± 2.48% (3.81 … 4.53 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 ± 1.4% (11.7 … 12.4 μs) 4.5 μs ± 3.84% (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 ± 0.801% (18.1 … 18.6 μs) 6.77 μs ± 2.02% (6.44 … 6.91 μs) 35.5
FPU GEMM Float32*Float32=Float32 (128×128) · (128×128) (NN) OP (4, 16, 1), base shape (4, 8, 1) 15.2 μs ± 1.11% (15.0 … 15.5 μs) 4.09 μs ± 2.57% (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 ± 1.09% (15.0 … 15.5 μs) 4.2 μs ± 3.82% (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 ± 0.466% (32.7 … 33.4 μs) 4.3 μs ± 3.41% (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 ± 1.22% (11.7 … 12.4 μs) 4.31 μs ± 3.27% (4.05 … 4.53 μs) 34.7
FPU GEMM Float64*Float64=Float64 (128×128) · (128×128) (TN) OP (8, 16, 2), base shape (4, 8, 1) 148.0 μs ± 4.66% (147.0 … 306.0 μs) 28.7 μs ± 0.463% (28.6 … 29.1 μs) 19.5
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) 4.3 μs ± 3.35% (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) 7.43 μs ± 1.51% (7.15 … 7.87 μs) 3.59 μs ± 2.87% (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 ± 1.04% (14.3 … 14.8 μs) 4.3 μs ± 3.28% (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) 12.7 μs ± 1.34% (12.4 … 12.9 μs) 7.2 μs ± 1.44% (6.91 … 7.63 μs) 55.8

@codecov Codecov
Copy link

codecov bot commented Nov 15, 2023

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (912fa05) 34.16% compared to head (2f556ce) 34.23%.

Additional details and impacted files 
@@            Coverage Diff             @@
##           master     #174      +/-   ##
==========================================
+ Coverage   34.16%   34.23%   +0.07%     
==========================================
  Files          11       11              
  Lines         922      923       +1     
==========================================
+ Hits          315      316       +1     
  Misses        607      607

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

maleadt
maleadt approved these changes Nov 15, 2023
View reviewed changes
Copy link
Member

@maleadt maleadt left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM.

Noisy benchmarks though...

@thomasfaingnaert
Copy link
Member Author

Noisy benchmarks though...

Yeah, I think at this point ripper was not idle, but even when it is I've seen quite high sigmas, unfortunately.
Wondering if it's because of the move to the new benchmark script, or to a new machine...
In any case, we really should fix it at some point.

@thomasfaingnaert thomasfaingnaert merged commit a7a802c into master Nov 15, 2023
1 check passed
@thomasfaingnaert thomasfaingnaert deleted the tf/check-num-threads branch November 15, 2023 10:06
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@maleadt maleadt maleadt approved these changes

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