Files
triton/python/examples/dot.py
2019-09-03 20:44:27 -04:00

139 lines
3.8 KiB
Python

import tensorflow as tf
import triton
import numpy as np
src = """
// Templates for accessing A
#if AT == 1
#define USE_A ^a
#define STRIDE_AK lda
#define STRIDE_AM 1
#define BROADCAST_AK :, newaxis
#define BROADCAST_AM newaxis, :
#define SHAPE_A TK, TM
#else
#define USE_A a
#define STRIDE_AK 1
#define STRIDE_AM lda
#define BROADCAST_AK newaxis, :
#define BROADCAST_AM :, newaxis
#define SHAPE_A TM, TK
#endif
// Templates for accessing B
#if BT == 1
#define USE_B ^b
#define STRIDE_BK 1
#define STRIDE_BN ldb
#define BROADCAST_BK newaxis, :
#define BROADCAST_BN :, newaxis
#define SHAPE_B TN, TK
#else
#define USE_B b
#define STRIDE_BK ldb
#define STRIDE_BN 1
#define BROADCAST_BK :, newaxis
#define BROADCAST_BN newaxis, :
#define SHAPE_B TK, TN
#endif
void dot(TYPE * A, TYPE * B, TYPE * C,
int M, int N, int K,
int lda __multipleof(8),
int ldb __multipleof(8),
int ldc) {
// prologue
int ridx = get_program_id(0);
int ridy = get_program_id(1);
int rxa[TM] = ridx * TM + 0 ... TM;
int ryb[TN] = ridy * TN + 0 ... TN;
int rka[TK] = 0 ... TK;
int rkb[TK] = 0 ... TK;
float c[TM, TN] = 0;
// pointers to operands
TYPE* pa[SHAPE_A] = A + rka[BROADCAST_AK] * STRIDE_AK + rxa[BROADCAST_AM] * STRIDE_AM;
TYPE* pb[SHAPE_B] = B + rkb[BROADCAST_BK] * STRIDE_BK + ryb[BROADCAST_BN] * STRIDE_BN;
// prefetches operands
TYPE a[SHAPE_A] = *pa;
TYPE b[SHAPE_B] = *pb;
// reduction loop
for(int k = K; k > 0; k-= TK){
c += USE_A @ USE_B;
pa = pa + TK * STRIDE_AK;
pb = pb + TK * STRIDE_BK;
a = *pa;
b = *pb;
}
// epilogue
int rxc[TM] = ridx * TM + 0 ... TM;
int ryc[TN] = ridy * TN + 0 ... TN;
TYPE* pc[TM, TN] = C + ryc[newaxis, :] + rxc[:, newaxis] * ldc;
bool checkc[TM, TN] = (rxc < M)[:, newaxis] && (ryc < N)[newaxis, :];
*?(checkc) pc = c;
}
"""
class dot_op:
def __init__(self, transpose_a = False, transpose_b = False):
self.dot = triton.op(src, ['C'])
self.transpose_a = transpose_a
self.transpose_b = transpose_b
def __call__(self, a, b):
# extract shapes
shape_a = triton.shape(a)
shape_b = triton.shape(b)
M, Ka = shape_a[0], shape_a[1]
Kb, N = shape_b[0], shape_b[1]
# transpose shapes
if self.transpose_a:
M, Ka = Ka, M
if self.transpose_b:
Kb, N = N, Kb
# contiguous dimensions
lda = M if self.transpose_a else Ka
ldb = Kb if self.transpose_b else N
ldc = N
# allocate output
c = triton.empty([M, N])
# compute
return self.dot(a, b, c, M, N, Ka, lda, ldb, ldc,
lambda opt: [triton.cdiv(M, opt.d('TM')), triton.cdiv(N, opt.d('TN'))],
AT = self.transpose_a, BT = self.transpose_b, TYPE = tf.float16,
TM = [128], TN = [128], TK = [32])
def dot(a, b, transpose_a = False, transpose_b = False):
if (transpose_a, transpose_b) not in dot.ops:
dot.ops[transpose_a, transpose_b] = dot_op(transpose_a, transpose_b)
return dot.ops[transpose_a, transpose_b](a, b)
dot.ops = dict()
@tf.RegisterGradient("Dot")
def _dot_grad(op, dy):
a = op.inputs[0]
b = op.inputs[1]
return [dot_tn(dy, b), dot_nt(a, dy), None, None, None, None, None, None, None]
def run_dot():
M, N, K = 128, 128, 128
a = tf.placeholder(tf.float16, shape=[M, K])
b = tf.placeholder(tf.float16, shape=[N, K])
c = dot(a, b, transpose_a = False, transpose_b = False)
# Reference
ha = np.random.rand(M, K).astype(np.float16)
hb = np.random.rand(K, N).astype(np.float16)
# Run
sess = tf.InteractiveSession()
sess.run(tf.global_variables_initializer())
result = sess.run([c], feed_dict = {a: ha,
b: hb})[0]
# Test
print(result)
hresult = np.dot(ha, hb)
dif = np.abs(result - hresult)
np.savetxt('dif.dat', dif, '%2.4f')
print("dif: %f" % np.max(dif))
run_dot()