add dynamic quantize gemm benchmark [step 1: minmax qparams compute] (p…

…ytorch#2297)

Summary:
Pull Request resolved: pytorch#2297

result in https://fb-my.sharepoint.com/:x:/g/personal/jiyuanz_meta_com/Ec94q-KgmslMtQ7nIYT4240BZUyWiK-iQvP1cBgzfgEDWg?e=G62cgs

Reviewed By: jspark1105

Differential Revision: D50590364

fbshipit-source-id: 0214a07fbb7bc1dd237f6af220eb863e71ec4472

fbgemm_gpu/src/qlinear_channelwise/qlinear_channelwise_mtia.cpp

@@ -41,6 +41,23 @@ static at::Tensor qlinear_quant(
   return x;
 }
 
+static at::Tensor qlinear_qparams(
+    at::Tensor x,
+    at::Tensor weight,
+    at::Tensor bias,
+    at::Tensor input_scale,
+    at::Tensor weight_scale,
+    at::Tensor weight_zero_point,
+    at::Tensor relu) {
+  assert(x.options().dtype() == at::kHalf);
+  assert(weight.options().dtype() == at::kQInt8);
+  assert(bias.options().dtype() == at::kFloat);
+  assert(input_scale.options().dtype() == at::kFloat);
+  assert(weight_scale.options().dtype() == at::kFloat);
+  assert(weight_zero_point.options().dtype() == at::kQUInt8);
+  return x;
+}
+
 TORCH_LIBRARY_FRAGMENT(fbgemm, m) {
   m.def(
       "qlinear_channelwise(Tensor x, Tensor weight, Tensor "
@@ -58,4 +75,13 @@ TORCH_LIBRARY_FRAGMENT(fbgemm, m) {
   m.impl(
       "qlinear_quant",
       torch::dispatch(c10::DispatchKey::CPU, TORCH_FN(qlinear_quant)));
+
+  m.def(
+      "qlinear_qparams(Tensor x, Tensor weight, Tensor "
+      "bias, Tensor input_scale, Tensor weight_scale, Tensor "
+      "weight_zero_point, Tensor relu) -> Tensor");
+
+  m.impl(
+      "qlinear_qparams",
+      torch::dispatch(c10::DispatchKey::CPU, TORCH_FN(qlinear_qparams)));
 }