Skip to content

Update fused quant broadcast logic (#20171) #20171

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

Open
DrJessop wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Update fused quant broadcast logic (#20171) #20171

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
14 changes: 3 additions & 11 deletions backends/cadence/fused_quant/op_add.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -43,19 +43,16 @@ Tensor& add_out(
ScalarType inp_dtype,
int64_t inp_quant_min,
int64_t inp_quant_max,
optional<int64_t> inp_axis,
const optional<Tensor>& other_scale,
const optional<Tensor>& other_zero_point,
ScalarType other_dtype,
int64_t other_quant_min,
int64_t other_quant_max,
optional<int64_t> other_axis,
const optional<Tensor>& out_scale,
const optional<Tensor>& out_zero_point,
ScalarType out_dtype,
int64_t out_quant_min,
int64_t out_quant_max,
optional<int64_t> out_axis,
double alpha,
Tensor& out) {
int64_t numel = inp.numel();
Expand All @@ -72,7 +69,7 @@ Tensor& add_out(
}
inp_buf.resize(numel);
QParams qp = extract_qparams(
inp_scale, inp_zero_point, inp_quant_min, inp_quant_max, inp_axis, inp);
inp_scale, inp_zero_point, inp_quant_min, inp_quant_max, inp);
FUSED_QUANT_DTYPE_SWITCH(
inp.scalar_type(),
scalar_t,
Expand All @@ -88,12 +85,7 @@ Tensor& add_out(
}
other_buf.resize(numel);
QParams qp = extract_qparams(
other_scale,
other_zero_point,
other_quant_min,
other_quant_max,
other_axis,
other);
other_scale, other_zero_point, other_quant_min, other_quant_max, other);
FUSED_QUANT_DTYPE_SWITCH(
other.scalar_type(),
scalar_t,
Expand All @@ -107,7 +99,7 @@ Tensor& add_out(
add_kernel(inp_float, other_float, result_float.data(), numel, alpha_f);

QParams qp = extract_qparams(
out_scale, out_zero_point, out_quant_min, out_quant_max, out_axis, out);
out_scale, out_zero_point, out_quant_min, out_quant_max, out);
FUSED_QUANT_DTYPE_SWITCH(
out.scalar_type(),
scalar_t,
Expand Down
3 changes: 0 additions & 3 deletions backends/cadence/fused_quant/op_add.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,20 +24,17 @@ executorch::aten::Tensor& add_out(
executorch::aten::ScalarType inp_dtype,
int64_t inp_quant_min,
int64_t inp_quant_max,
executorch::aten::optional<int64_t> inp_axis,
const executorch::aten::optional<executorch::aten::Tensor>& other_scale,
const executorch::aten::optional<executorch::aten::Tensor>&
other_zero_point,
executorch::aten::ScalarType other_dtype,
int64_t other_quant_min,
int64_t other_quant_max,
executorch::aten::optional<int64_t> other_axis,
const executorch::aten::optional<executorch::aten::Tensor>& out_scale,
const executorch::aten::optional<executorch::aten::Tensor>& out_zero_point,
executorch::aten::ScalarType out_dtype,
int64_t out_quant_min,
int64_t out_quant_max,
executorch::aten::optional<int64_t> out_axis,
double alpha,
executorch::aten::Tensor& out);

Expand Down
14 changes: 3 additions & 11 deletions backends/cadence/fused_quant/op_bmm.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -53,19 +53,16 @@ Tensor& bmm_out(
ScalarType inp_dtype,
int64_t inp_quant_min,
int64_t inp_quant_max,
optional<int64_t> inp_axis,
const optional<Tensor>& other_scale,
const optional<Tensor>& other_zero_point,
ScalarType other_dtype,
int64_t other_quant_min,
int64_t other_quant_max,
optional<int64_t> other_axis,
const optional<Tensor>& out_scale,
const optional<Tensor>& out_zero_point,
ScalarType out_dtype,
int64_t out_quant_min,
int64_t out_quant_max,
optional<int64_t> out_axis,
Tensor& out) {
int64_t batch = inp.size(0);
int64_t M = inp.size(1);
Expand All @@ -87,7 +84,7 @@ Tensor& bmm_out(
}
inp_buf.resize(inp_numel);
QParams qp = extract_qparams(
inp_scale, inp_zero_point, inp_quant_min, inp_quant_max, inp_axis, inp);
inp_scale, inp_zero_point, inp_quant_min, inp_quant_max, inp);
FUSED_QUANT_DTYPE_SWITCH(
inp.scalar_type(),
scalar_t,
Expand All @@ -104,12 +101,7 @@ Tensor& bmm_out(
}
other_buf.resize(other_numel);
QParams qp = extract_qparams(
other_scale,
other_zero_point,
other_quant_min,
other_quant_max,
other_axis,
other);
other_scale, other_zero_point, other_quant_min, other_quant_max, other);
FUSED_QUANT_DTYPE_SWITCH(other.scalar_type(),
scalar_t,
dequantize_buffer(
Expand All @@ -126,7 +118,7 @@ Tensor& bmm_out(
bmm_kernel(inp_float, other_float, result_float.data(), batch, M, K, N);

QParams qp = extract_qparams(
out_scale, out_zero_point, out_quant_min, out_quant_max, out_axis, out);
out_scale, out_zero_point, out_quant_min, out_quant_max, out);
FUSED_QUANT_DTYPE_SWITCH(out.scalar_type(),
scalar_t,
quantize_buffer(
Expand Down
3 changes: 0 additions & 3 deletions backends/cadence/fused_quant/op_bmm.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,20 +24,17 @@ executorch::aten::Tensor& bmm_out(
executorch::aten::ScalarType inp_dtype,
int64_t inp_quant_min,
int64_t inp_quant_max,
executorch::aten::optional<int64_t> inp_axis,
const executorch::aten::optional<executorch::aten::Tensor>& other_scale,
const executorch::aten::optional<executorch::aten::Tensor>&
other_zero_point,
executorch::aten::ScalarType other_dtype,
int64_t other_quant_min,
int64_t other_quant_max,
executorch::aten::optional<int64_t> other_axis,
const executorch::aten::optional<executorch::aten::Tensor>& out_scale,
const executorch::aten::optional<executorch::aten::Tensor>& out_zero_point,
executorch::aten::ScalarType out_dtype,
int64_t out_quant_min,
int64_t out_quant_max,
executorch::aten::optional<int64_t> out_axis,
executorch::aten::Tensor& out);

} // namespace native
Expand Down
6 changes: 2 additions & 4 deletions backends/cadence/fused_quant/op_hardswish.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,13 +40,11 @@ Tensor& hardswish_out(
ScalarType inp_dtype,
int64_t inp_quant_min,
int64_t inp_quant_max,
optional<int64_t> inp_axis,
const optional<Tensor>& out_scale,
const optional<Tensor>& out_zero_point,
ScalarType out_dtype,
int64_t out_quant_min,
int64_t out_quant_max,
optional<int64_t> out_axis,
Tensor& out) {
int64_t numel = inp.numel();

Expand All @@ -60,7 +58,7 @@ Tensor& hardswish_out(
}
inp_buf.resize(numel);
QParams qp = extract_qparams(
inp_scale, inp_zero_point, inp_quant_min, inp_quant_max, inp_axis, inp);
inp_scale, inp_zero_point, inp_quant_min, inp_quant_max, inp);
FUSED_QUANT_DTYPE_SWITCH(
inp.scalar_type(),
scalar_t,
Expand All @@ -74,7 +72,7 @@ Tensor& hardswish_out(
hardswish_kernel(inp_float, result_float.data(), numel);

QParams qp = extract_qparams(
out_scale, out_zero_point, out_quant_min, out_quant_max, out_axis, out);
out_scale, out_zero_point, out_quant_min, out_quant_max, out);
FUSED_QUANT_DTYPE_SWITCH(
out.scalar_type(),
scalar_t,
Expand Down
2 changes: 0 additions & 2 deletions backends/cadence/fused_quant/op_hardswish.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,13 +23,11 @@ executorch::aten::Tensor& hardswish_out(
executorch::aten::ScalarType inp_dtype,
int64_t inp_quant_min,
int64_t inp_quant_max,
executorch::aten::optional<int64_t> inp_axis,
const executorch::aten::optional<executorch::aten::Tensor>& out_scale,
const executorch::aten::optional<executorch::aten::Tensor>& out_zero_point,
executorch::aten::ScalarType out_dtype,
int64_t out_quant_min,
int64_t out_quant_max,
executorch::aten::optional<int64_t> out_axis,
executorch::aten::Tensor& out);

} // namespace native
Expand Down
14 changes: 3 additions & 11 deletions backends/cadence/fused_quant/op_mul.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -42,19 +42,16 @@ Tensor& mul_out(
ScalarType inp_dtype,
int64_t inp_quant_min,
int64_t inp_quant_max,
optional<int64_t> inp_axis,
const optional<Tensor>& other_scale,
const optional<Tensor>& other_zero_point,
ScalarType other_dtype,
int64_t other_quant_min,
int64_t other_quant_max,
optional<int64_t> other_axis,
const optional<Tensor>& out_scale,
const optional<Tensor>& out_zero_point,
ScalarType out_dtype,
int64_t out_quant_min,
int64_t out_quant_max,
optional<int64_t> out_axis,
Tensor& out) {
(void)ctx;
(void)inp_dtype;
Expand All @@ -74,7 +71,7 @@ Tensor& mul_out(
}
inp_buf.resize(numel);
QParams qp = extract_qparams(
inp_scale, inp_zero_point, inp_quant_min, inp_quant_max, inp_axis, inp);
inp_scale, inp_zero_point, inp_quant_min, inp_quant_max, inp);
FUSED_QUANT_DTYPE_SWITCH(
inp.scalar_type(),
scalar_t,
Expand All @@ -90,12 +87,7 @@ Tensor& mul_out(
}
other_buf.resize(numel);
QParams qp = extract_qparams(
other_scale,
other_zero_point,
other_quant_min,
other_quant_max,
other_axis,
other);
other_scale, other_zero_point, other_quant_min, other_quant_max, other);
FUSED_QUANT_DTYPE_SWITCH(
other.scalar_type(),
scalar_t,
Expand All @@ -109,7 +101,7 @@ Tensor& mul_out(
mul_kernel(inp_float, other_float, result_float.data(), numel);

QParams qp = extract_qparams(
out_scale, out_zero_point, out_quant_min, out_quant_max, out_axis, out);
out_scale, out_zero_point, out_quant_min, out_quant_max, out);
FUSED_QUANT_DTYPE_SWITCH(
out.scalar_type(),
scalar_t,
Expand Down
3 changes: 0 additions & 3 deletions backends/cadence/fused_quant/op_mul.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,20 +24,17 @@ executorch::aten::Tensor& mul_out(
executorch::aten::ScalarType inp_dtype,
int64_t inp_quant_min,
int64_t inp_quant_max,
executorch::aten::optional<int64_t> inp_axis,
const executorch::aten::optional<executorch::aten::Tensor>& other_scale,
const executorch::aten::optional<executorch::aten::Tensor>&
other_zero_point,
executorch::aten::ScalarType other_dtype,
int64_t other_quant_min,
int64_t other_quant_max,
executorch::aten::optional<int64_t> other_axis,
const executorch::aten::optional<executorch::aten::Tensor>& out_scale,
const executorch::aten::optional<executorch::aten::Tensor>& out_zero_point,
executorch::aten::ScalarType out_dtype,
int64_t out_quant_min,
int64_t out_quant_max,
executorch::aten::optional<int64_t> out_axis,
executorch::aten::Tensor& out);

} // namespace native
Expand Down
6 changes: 2 additions & 4 deletions backends/cadence/fused_quant/op_relu.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -39,13 +39,11 @@ Tensor& relu_out(
ScalarType inp_dtype,
int64_t inp_quant_min,
int64_t inp_quant_max,
optional<int64_t> inp_axis,
const optional<Tensor>& out_scale,
const optional<Tensor>& out_zero_point,
ScalarType out_dtype,
int64_t out_quant_min,
int64_t out_quant_max,
optional<int64_t> out_axis,
Tensor& out) {
int64_t numel = inp.numel();

Expand All @@ -59,7 +57,7 @@ Tensor& relu_out(
}
inp_buf.resize(numel);
QParams qp = extract_qparams(
inp_scale, inp_zero_point, inp_quant_min, inp_quant_max, inp_axis, inp);
inp_scale, inp_zero_point, inp_quant_min, inp_quant_max, inp);
FUSED_QUANT_DTYPE_SWITCH(
inp.scalar_type(),
scalar_t,
Expand All @@ -73,7 +71,7 @@ Tensor& relu_out(
relu_kernel(inp_float, result_float.data(), numel);

QParams qp = extract_qparams(
out_scale, out_zero_point, out_quant_min, out_quant_max, out_axis, out);
out_scale, out_zero_point, out_quant_min, out_quant_max, out);
FUSED_QUANT_DTYPE_SWITCH(
out.scalar_type(),
scalar_t,
Expand Down
2 changes: 0 additions & 2 deletions backends/cadence/fused_quant/op_relu.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,13 +23,11 @@ executorch::aten::Tensor& relu_out(
executorch::aten::ScalarType inp_dtype,
int64_t inp_quant_min,
int64_t inp_quant_max,
executorch::aten::optional<int64_t> inp_axis,
const executorch::aten::optional<executorch::aten::Tensor>& out_scale,
const executorch::aten::optional<executorch::aten::Tensor>& out_zero_point,
executorch::aten::ScalarType out_dtype,
int64_t out_quant_min,
int64_t out_quant_max,
executorch::aten::optional<int64_t> out_axis,
executorch::aten::Tensor& out);

} // namespace native
Expand Down
Loading
Loading