Skip to content

[XPU] use get_max_ptr function, test=develop, test=xpu #7482

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

Merged
zhupengyang merged 1 commit into from
Oct 28, 2021
Merged

[XPU] use get_max_ptr function, test=develop, test=xpu #7482

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
9 changes: 6 additions & 3 deletions lite/kernels/xpu/__xpu__conv2d_compute.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -52,17 +52,20 @@ bool QuantFilter<int8_t>(const float* filter_on_host,

template <typename T, PrecisionType PType>
void XPUConv2dCompute<T, PType>::PrepareForRun() {
auto& ctx = this->ctx_->template As<XPUContext>();
auto& param = this->template Param<param_t>();
auto filter_ptr = param.filter->template data<float>();
auto filter_len = param.filter->numel();
// max
float max_f = paddle::lite::xpu::math::FindMaxAbs(filter_ptr, filter_len);
std::vector<float> max_f_v(4, max_f);
filter_max_guard_ = TargetWrapperXPU::MallocScratchPad(4 * sizeof(float));
int max_ptr_size = get_max_ptr_size(ctx.GetRawContext());
std::vector<float> max_f_v(max_ptr_size, max_f);
filter_max_guard_ =
TargetWrapperXPU::MallocScratchPad(max_ptr_size * sizeof(float));
filter_max_ = reinterpret_cast<float*>(filter_max_guard_->addr_);
XPU_CALL(xpu_memcpy(filter_max_,
max_f_v.data(),
4 * sizeof(float),
max_ptr_size * sizeof(float),
XPUMemcpyKind::XPU_HOST_TO_DEVICE));
// quant
quant_filter_guard_ =
Expand Down
14 changes: 8 additions & 6 deletions lite/kernels/xpu/__xpu__fc_compute.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,6 +25,7 @@ namespace kernels {
namespace xpu {

void XPUFcCompute::PrepareForRun() {
auto& ctx = this->ctx_->As<XPUContext>();
auto& param = this->Param<param_t>();
auto w_ptr = param.w->data<float>();
auto w_len = param.w->numel();
Expand All @@ -36,15 +37,16 @@ void XPUFcCompute::PrepareForRun() {
// max
if (!quant_int8) {
w_max = paddle::lite::xpu::math::FindMaxAbs(w_ptr, w_len);
std::vector<float> w_max_v(lite::XPU_QUANT_SCALE_NUM, w_max);
weight_max_guard_ = TargetWrapperXPU::MallocScratchPad(
lite::XPU_QUANT_SCALE_NUM * sizeof(float));
int max_ptr_size = get_max_ptr_size(ctx.GetRawContext());
std::vector<float> w_max_v(max_ptr_size, w_max);
weight_max_guard_ =
TargetWrapperXPU::MallocScratchPad(max_ptr_size * sizeof(float));
XPU_CALL(xpu_memcpy(reinterpret_cast<float*>(weight_max_guard_->addr_),
w_max_v.data(),
lite::XPU_QUANT_SCALE_NUM * sizeof(float),
max_ptr_size * sizeof(float),
XPUMemcpyKind::XPU_HOST_TO_DEVICE));
input_max_guard_ = TargetWrapperXPU::MallocScratchPad(
lite::XPU_QUANT_SCALE_NUM * sizeof(float));
input_max_guard_ =
TargetWrapperXPU::MallocScratchPad(max_ptr_size * sizeof(float));
}
// transpose
if (quant_int8) {
Expand Down
8 changes: 5 additions & 3 deletions lite/kernels/xpu/match_matrix_tensor_compute.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,14 +23,16 @@ namespace kernels {
namespace xpu {

void MatchMatrixTensorCompute::PrepareForRun() {
auto& ctx = this->ctx_->As<XPUContext>();
auto& param = this->Param<param_t>();
float w_max = param.__xpu__w_max;
std::vector<float> w_max_v(XPU_QUANT_SCALE_NUM, w_max);
int max_ptr_size = get_max_ptr_size(ctx.GetRawContext());
std::vector<float> w_max_v(max_ptr_size, w_max);
weight_max_xpu_guard_ =
TargetWrapperXPU::MallocScratchPad(XPU_QUANT_SCALE_NUM * sizeof(float));
TargetWrapperXPU::MallocScratchPad(max_ptr_size * sizeof(float));
XPU_CALL(xpu_memcpy(reinterpret_cast<float*>(weight_max_xpu_guard_->addr_),
w_max_v.data(),
XPU_QUANT_SCALE_NUM * sizeof(float),
max_ptr_size * sizeof(float),
XPUMemcpyKind::XPU_HOST_TO_DEVICE));

offset_l_xpu_guard_ =
Expand Down