add int quantization for xpu (#54802)

e9f8baa6 · lijin23 · GitHub · 3371f98b · e9f8baa6
隐藏空白更改
内联并排

Showing with 59 addition and 27 deletion

paddle/phi/kernels/xpu/conv_transpose_grad_kernel.cc paddle/phi/kernels/xpu/conv_transpose_grad_kernel.cc +59 -27

未找到文件。
--- a/paddle/phi/kernels/xpu/conv_transpose_grad_kernel.cc
+++ b/paddle/phi/kernels/xpu/conv_transpose_grad_kernel.cc
@@ -17,6 +17,7 @@
 #include "paddle/phi/backends/xpu/enforce_xpu.h"
 #include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/kernels/cpu/conv_util.h"
+#include "paddle/phi/kernels/xpu/xpu_api_wrapper.h"
 namespace phi {
 template <typename T, typename Context>
@@ -68,33 +69,64 @@ void Conv2dTransposeGradKernel(const Context& ctx,
  if (dfilter) {
    ctx.template Alloc<T>(dfilter);
  }
+  int fccal_type = FCCalcType<T>();
-  int r = xpu::conv2d_transpose_grad<float, float, float, int16_t>(
+  if (fccal_type == XPUFCCalcType::FC_INT32 ||
-      ctx.x_context(),
+      fccal_type == XPUFCCalcType::FC_INT32_WITH_LL) {
-      x.data<T>(),
+    // xpu api do not support int31 quantization now.
-      filter_.data<T>(),
+    int r = xpu::conv2d_transpose_grad<float, float, float, int_with_ll_t>(
-      dout.data<T>(),
+        ctx.x_context(),
-      dx ? dx->data<T>() : nullptr,
+        x.data<T>(),
-      dfilter ? dfilter->data<T>() : nullptr,
+        filter_.data<T>(),
-      batch_size,
+        dout.data<T>(),
-      img_yc,
+        dx ? dx->data<T>() : nullptr,
-      img_yh,
+        dfilter ? dfilter->data<T>() : nullptr,
-      img_yw,
+        batch_size,
-      img_xc,
+        img_yc,
-      img_xh,
+        img_yh,
-      img_xw,
+        img_yw,
-      ksize,
+        img_xc,
-      strides,
+        img_xh,
-      paddings_,
+        img_xw,
-      dilations_,
+        ksize,
-      groups,
+        strides,
-      nullptr,
+        paddings_,
-      nullptr,
+        dilations_,
-      nullptr,
+        groups,
-      nullptr,
+        nullptr,
-      nullptr,
+        nullptr,
-      true);
+        nullptr,
-  PADDLE_ENFORCE_XDNN_SUCCESS(r, "conv2d_transpose_grad");
+        nullptr,
+        nullptr,
+        true);
+    PADDLE_ENFORCE_XDNN_SUCCESS(r, "conv2d_transpose_grad");
+  } else {
+    int r = xpu::conv2d_transpose_grad<float, float, float, int16_t>(
+        ctx.x_context(),
+        x.data<T>(),
+        filter_.data<T>(),
+        dout.data<T>(),
+        dx ? dx->data<T>() : nullptr,
+        dfilter ? dfilter->data<T>() : nullptr,
+        batch_size,
+        img_yc,
+        img_yh,
+        img_yw,
+        img_xc,
+        img_xh,
+        img_xw,
+        ksize,
+        strides,
+        paddings_,
+        dilations_,
+        groups,
+        nullptr,
+        nullptr,
+        nullptr,
+        nullptr,
+        nullptr,
+        true);
+    PADDLE_ENFORCE_XDNN_SUCCESS(r, "conv2d_transpose_grad");
+  }
 }
 template <typename T, typename Context>