Optimize Outer LoopRetime to Eliminate Broadcast

for i=1 to n in place

for j=1 to k in parallel

yi += wj * xi+j-1

W1

. . . x2 x1

. . . y2 y1

W2

Wk

Similar to W2 From [Kung82]

Previous slide

Next slide

Back to first slide

View graphic version