[luci/pass] Introduce FuseMulToFullyConnectedWeightsPass #13439
Conversation
This will introduce FuseMulToFullyConnectedWeightsPass which will fuse Mul to following FullyConnected weights if possible. ONE-DCO-1.0-Signed-off-by: SaeHie Park <[email protected]>
| _mul_s->size<DT>(3); | ||
| for (uint32_t i = 0; i < 3; ++i) | ||
| { | ||
| _mul_s->at<DT>(0) = 1.0f; |
There was a problem hiding this comment.
i index is not used in for statement..!?
There was a problem hiding this comment.
Did you intend below code?
| _mul_s->at<DT>(0) = 1.0f; | |
| _mul_s->at<DT>(i) = 1.0f; |
| _fc_w->size<DT>(4 * 6); | ||
| for (uint32_t i = 0; i < 4 * 6; ++i) | ||
| { | ||
| _fc_w->at<DT>(0) = 1.0f; |
| _fc_b->size<DT>(6); | ||
| for (uint32_t i = 0; i < 6; ++i) | ||
| { | ||
| _fc_b->at<DT>(0) = 1.0f; |
| _fc_w->dim(0) = 3; | ||
| _fc_w->dim(1) = 4; | ||
| _fc_w->dtype(DT); | ||
| _fc_w->size<DT>(4 * 6); | ||
| for (uint32_t i = 0; i < 4 * 6; ++i) |
There was a problem hiding this comment.
The shape of _fc_w is <3x4>
Does this size have to be the same to the shape..?
| _fc_w->dim(0) = 3; | |
| _fc_w->dim(1) = 4; | |
| _fc_w->dtype(DT); | |
| _fc_w->size<DT>(4 * 6); | |
| for (uint32_t i = 0; i < 4 * 6; ++i) | |
| _fc_w->dim(0) = 3; | |
| _fc_w->dim(1) = 4; | |
| _fc_w->dtype(DT); | |
| _fc_w->size<DT>(3 * 4); | |
| for (uint32_t i = 0; i < 3 * 4; ++i) |
There was a problem hiding this comment.
or
| _fc_w->dim(0) = 3; | |
| _fc_w->dim(1) = 4; | |
| _fc_w->dtype(DT); | |
| _fc_w->size<DT>(4 * 6); | |
| for (uint32_t i = 0; i < 4 * 6; ++i) | |
| _fc_w->dim(0) = 4; | |
| _fc_w->dim(1) = 6; | |
| _fc_w->dtype(DT); | |
| _fc_w->size<DT>(4 * 6); | |
| for (uint32_t i = 0; i < 4 * 6; ++i) |
There was a problem hiding this comment.
please check about FullyConnected Op.
with torch,
import torch
tensor1 = torch.randn(3, 4)
FC = torch.nn.Linear(4, 6)
output = FC(tensor1)
print(output)
print(output.shape)gives something like
tensor([[-0.8929, 0.2467, -0.1407, -0.9917, -0.3787, 0.6504],
[-0.2442, -0.0493, -0.1175, -1.0433, 0.0100, -0.3347],
[ 0.7240, 0.0195, -0.3735, -1.5789, -0.0348, -0.9252]],
grad_fn=<AddmmBackward0>)
torch.Size([3, 6])
There was a problem hiding this comment.
more simple case is with vector,
tensor1 = torch.randn(4)
FC = torch.nn.Linear(4, 6)
output = FC(tensor1)
print(output)
print(output.shape)gives
tensor([ 0.7355, -0.1311, 0.4749, -0.1821, -0.1245, -1.0594],
grad_fn=<ViewBackward0>)
torch.Size([6])
There was a problem hiding this comment.
ah, should be
_fc_w->dim(0) = 6;
_fc_w->dim(1) = 4;
| _mul->x(input()); | ||
| _mul->y(_mul_s); | ||
| _fc->input(_mul); | ||
| _fc->weights(_fc_b); |
There was a problem hiding this comment.
Should it be _fc_w instead of _fc_b..?
| TEST_F(FuseMulToFullyConnectedWeightsPassS32Test, dtype_s32_NEG) | ||
| { | ||
| _graph.init(); | ||
|
|
||
| EXPECT_FALSE(_pass.run(_graph.g())); | ||
| } |
There was a problem hiding this comment.
This test seems same with the fuse_mul_to_fc_weights test, but it expects false.
How could it be..!?
There was a problem hiding this comment.
FuseMulToFullyConnectedWeightsPassS32Test is made with S32, which datatype we can't support.
There was a problem hiding this comment.
Ah, it was S32.
Thank you!
|
@seanshpark, |
This will introduce FuseMulToFullyConnectedWeightsPass which will
fuse Mul to following FullyConnected weights if possible.