gaspersic/triton
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[GH-PAGES] Updated website

Browse Source
This commit is contained in:
Philippe Tillet
2022-07-23 00:49:35 +00:00
parent 10f1d77697
commit 5f9c7bc693
165 changed files with 273 additions and 273 deletions
Download Patch File Download Diff File Expand all files Collapse all files

54
master/_sources/getting-started/tutorials/03-matrix-multiplication.rst.txt
View File

@@ -262,8 +262,8 @@ Final Result
b_ptrs += BLOCK_SIZE_K * stride_bk
# you can fuse arbitrary activation functions here
# while the accumulator is still in FP32!
if ACTIVATION:
accumulator = ACTIVATION(accumulator)
if ACTIVATION == "leaky_relu":
accumulator = leaky_relu(accumulator)
c = accumulator.to(tl.float16)
# -----------------------------------------------------------
@@ -300,7 +300,7 @@ and (1) checks any shape constraint; (2) allocates the output; (3) launches the
def matmul(a, b, activation=None):
def matmul(a, b, activation=""):
# checks constraints
assert a.shape[1] == b.shape[0], "incompatible dimensions"
assert a.is_contiguous(), "matrix A must be contiguous"
@@ -436,7 +436,7 @@ We can now compare the performance of our kernel against that of cuBLAS. Here we
)
if provider == 'triton + relu':
ms, min_ms, max_ms = triton.testing.do_bench(
lambda: matmul(a, b, activation=leaky_relu)
lambda: matmul(a, b, activation="leaky_relu")
)
perf = lambda ms: 2 * M * N * K * 1e-12 / (ms * 1e-3)
return perf(ms), perf(max_ms), perf(min_ms)
@@ -459,37 +459,37 @@ We can now compare the performance of our kernel against that of cuBLAS. Here we
matmul-performance:
M cuBLAS ... Triton Triton (+ LeakyReLU)
0 256.0 2.730667 ... 2.978909 2.978909
1 384.0 7.372800 ... 7.899428 7.899428
0 256.0 2.978909 ... 2.978909 2.978909
1 384.0 7.372800 ... 8.507077 8.507077
2 512.0 14.563555 ... 15.420235 15.420235
3 640.0 22.260869 ... 24.380953 24.380953
4 768.0 32.768000 ... 34.028308 34.028308
5 896.0 39.025776 ... 40.140799 39.025776
6 1024.0 49.932191 ... 53.773130 52.428801
7 1152.0 45.242181 ... 47.396572 47.396572
7 1152.0 45.242181 ... 48.161033 47.396572
8 1280.0 51.200001 ... 57.690139 57.690139
9 1408.0 64.138541 ... 68.147202 67.305878
10 1536.0 80.430545 ... 80.430545 79.526831
10 1536.0 80.430545 ... 81.355034 79.526831
11 1664.0 63.372618 ... 63.372618 62.492442
12 1792.0 72.983276 ... 73.460287 59.467852
13 1920.0 68.776119 ... 71.626943 71.257735
14 2048.0 73.908442 ... 78.398206 77.314362
15 2176.0 83.500614 ... 87.494120 85.998493
16 2304.0 68.446623 ... 78.064941 77.307030
17 2432.0 71.125224 ... 86.179335 85.653855
18 2560.0 77.833728 ... 82.331658 81.108913
19 2688.0 83.737433 ... 91.185232 89.888756
20 2816.0 83.233216 ... 84.441840 84.197315
21 2944.0 81.564701 ... 83.758038 82.373605
22 3072.0 82.540970 ... 89.593522 88.335577
23 3200.0 83.989503 ... 95.096582 89.012517
24 3328.0 82.464255 ... 82.939284 84.596116
25 3456.0 81.932484 ... 90.994998 91.200871
26 3584.0 87.127323 ... 99.354022 92.600816
27 3712.0 84.159518 ... 89.353616 83.247783
28 3840.0 85.136259 ... 93.484358 86.738820
29 3968.0 92.302520 ... 87.976885 90.926929
30 4096.0 91.741443 ... 90.933416 91.304576
13 1920.0 68.776119 ... 71.257735 70.892307
14 2048.0 73.584279 ... 78.033565 76.959706
15 2176.0 83.155572 ... 87.494120 85.998493
16 2304.0 68.446623 ... 78.320893 77.558029
17 2432.0 71.305746 ... 86.711310 75.320281
18 2560.0 77.833728 ... 82.747477 81.715711
19 2688.0 83.552988 ... 90.532356 89.676257
20 2816.0 83.552120 ... 84.035084 83.392363
21 2944.0 81.832567 ... 83.758038 81.967162
22 3072.0 82.540970 ... 89.877939 89.170242
23 3200.0 84.321474 ... 96.822991 95.380032
24 3328.0 83.034941 ... 85.806075 84.596116
25 3456.0 82.183044 ... 91.928814 87.632137
26 3584.0 87.381330 ... 92.696281 96.891584
27 3712.0 84.694652 ... 87.244203 88.092894
28 3840.0 85.136259 ... 88.900318 90.279183
29 3968.0 88.008611 ... 92.547541 84.268854
30 4096.0 93.368854 ... 87.781379 86.592080
[31 rows x 5 columns]
@@ -499,7 +499,7 @@ We can now compare the performance of our kernel against that of cuBLAS. Here we
.. rst-class:: sphx-glr-timing
**Total running time of the script:** ( 6 minutes 44.471 seconds)
**Total running time of the script:** ( 6 minutes 21.318 seconds)
.. _sphx_glr_download_getting-started_tutorials_03-matrix-multiplication.py: