Fix performence performance (#61246)

co63oc · web-flow · commit c080ac2f13f3 · 2024-01-29T16:01:56.000+08:00
diff --git a/paddle/phi/kernels/fusion/cutlass/conv2d/conv2d_common.py b/paddle/phi/kernels/fusion/cutlass/conv2d/conv2d_common.py
@@ -111,7 +111,7 @@
 
 # CommonConvFunction is a wrapper for many kernels
 # a func_name is like conv2d_bias_silu_sm75
-# it has many kernels, we should pick up a performence-best
+# it has many kernels, we should pick up a performance-best
 # ${func_name} is like conv2d_bias_silu_sm75
 # ${enum_op_name} is like CONV2D_BIAS_SILU
 
diff --git a/paddle/phi/kernels/fusion/cutlass/memory_efficient_attention/iterators/predicated_tile_access_iterator_residual_last.h b/paddle/phi/kernels/fusion/cutlass/memory_efficient_attention/iterators/predicated_tile_access_iterator_residual_last.h
@@ -195,7 +195,7 @@ class PredicatedTileAccessIteratorResidualLast<Shape_,
   BytePointer pointer_;
 
   /// Below is used when Gather is turned on.  We need to record strided_offset
-  /// and contiguous_offset seperated to compute the offset by using
+  /// and contiguous_offset separated to compute the offset by using
   ///
   /// offset = contiguous_offset + indices[strided_offset]
   ///
diff --git a/paddle/phi/kernels/fusion/gpu/fused_fc_elementwise_layernorm_kernel.cu b/paddle/phi/kernels/fusion/gpu/fused_fc_elementwise_layernorm_kernel.cu
@@ -85,7 +85,7 @@ __global__ void InplaceAddReluAddLayerNormKernel(const T* y,
   for (int i = blockIdx.x; i < M; i += gridDim.x) {
     int index = i * N + threadIdx.x;
 
-    // The fisrt BlockDim elements will be saved to shared memory.
+    // The first BlockDim elements will be saved to shared memory.
     int save_index = threadIdx.x;
     T* save_ptr = shared_mem;
 
@@ -189,7 +189,7 @@ __global__ void InplaceAddReluAddLayerNormKernel(const float16* y_data,
   for (int i = blockIdx.x; i < M; i += gridDim.x) {
     int index = i * N + threadIdx.x;
 
-    // The fisrt BlockDim elements will be saved to shared memory.
+    // The first BlockDim elements will be saved to shared memory.
     int save_index = threadIdx.x;
 #if defined(PADDLE_WITH_CUDA)
     half* save_ptr = shared_mem;
diff --git a/paddle/phi/kernels/fusion/gpu/fused_rope_kernel.cu b/paddle/phi/kernels/fusion/gpu/fused_rope_kernel.cu
@@ -96,7 +96,7 @@ void FusedRopeKernel(const Context& dev_ctx,
                       cos.get_ptr()->dims(),
                       phi::errors::InvalidArgument(
                           "The dims of sin and cos must be the same. But "
-                          "recieved sin's dims is {%s}, cos's dims is {%s}.",
+                          "received sin's dims is {%s}, cos's dims is {%s}.",
                           sin.get_ptr()->dims(),
                           cos.get_ptr()->dims()));
 
@@ -106,7 +106,7 @@ void FusedRopeKernel(const Context& dev_ctx,
         (dims_size == 2 || dims_size == 4),
         true,
         phi::errors::InvalidArgument("The dims of sin and cos is expected to "
-                                     "be 2 or 4, but recieved %d.",
+                                     "be 2 or 4, but received %d.",
                                      dims_size));
     if (dims_size == 4) {
       // sin.shape: [1, seq_len, 1, head_dim]
@@ -126,7 +126,7 @@ void FusedRopeKernel(const Context& dev_ctx,
           phi::errors::InvalidArgument(
               "The seq_len of sin and cos must be greater than or equal to "
               "this of q. The head_dim of sin and cos must be the same as this "
-              "of q. But recieved sin's "
+              "of q. But received sin's "
               "shape is {%s}, q's shape is {%s}.",
               sin_dims,
               q.dims()));
@@ -136,7 +136,7 @@ void FusedRopeKernel(const Context& dev_ctx,
                         2,
                         phi::errors::InvalidArgument(
                             "The dims of position_ids is expected to "
-                            "be 2, but recieved %d.",
+                            "be 2, but received %d.",
                             position_ids_dims.size()));
 
       PADDLE_ENFORCE_EQ(
@@ -145,7 +145,7 @@ void FusedRopeKernel(const Context& dev_ctx,
           true,
           phi::errors::InvalidArgument(
               "The batch_size and seq_len of position_ids must be the same as "
-              "those of q. But recieved position_ids's "
+              "those of q. But received position_ids's "
               "shape is {%s}, q's shape is {%s}.",
               position_ids_dims,
               q.dims()));
@@ -158,7 +158,7 @@ void FusedRopeKernel(const Context& dev_ctx,
           true,
           phi::errors::InvalidArgument(
               "The seq_len and head_dim of sin and cos "
-              "must be the same as those of q. But recieved sin's "
+              "must be the same as those of q. But received sin's "
               "shape is {%s}, q's shape is {%s}.",
               sin_dims,
               q.dims()));
diff --git a/paddle/phi/kernels/fusion/xpu/conv2d_xpu_kernel.cc b/paddle/phi/kernels/fusion/xpu/conv2d_xpu_kernel.cc
@@ -51,7 +51,7 @@ void Conv2dXPUKernelImpl(const Context& ctx,
   using XPUTypeOut = typename XPUTypeTrait<T_OUT>::Type;
   auto input_dims = x.dims();
   auto filter_dims = filter.dims();
-  // update paddings and dilations accoring to padding_algorithm
+  // update paddings and dilations according to padding_algorithm
   std::vector<int> paddings_vec = paddings;
   std::vector<int> dilations_vec = dilations;
   DDim in_data_dims = common::slice_ddim(input_dims, 2, input_dims.size());
diff --git a/paddle/phi/kernels/fusion/xpu/fused_rope_grad_kernel.cc b/paddle/phi/kernels/fusion/xpu/fused_rope_grad_kernel.cc
@@ -55,7 +55,7 @@ void FusedRopeGradKernel(const Context& dev_ctx,
                       cos.get_ptr()->dims(),
                       phi::errors::InvalidArgument(
                           "The dims of sin and cos must be the same. But "
-                          "recieved sin's dims is {%s}, cos's dims is {%s}.",
+                          "received sin's dims is {%s}, cos's dims is {%s}.",
                           sin.get_ptr()->dims(),
                           cos.get_ptr()->dims()));
   }
diff --git a/paddle/phi/kernels/fusion/xpu/fused_rope_kernel.cc b/paddle/phi/kernels/fusion/xpu/fused_rope_kernel.cc
@@ -57,7 +57,7 @@ void FusedRopeKernel(const Context& dev_ctx,
                       cos.get_ptr()->dims(),
                       phi::errors::InvalidArgument(
                           "The dims of sin and cos must be the same. But "
-                          "recieved sin's dims is {%s}, cos's dims is {%s}.",
+                          "received sin's dims is {%s}, cos's dims is {%s}.",
                           sin.get_ptr()->dims(),
                           cos.get_ptr()->dims()));
   }
diff --git a/paddle/phi/kernels/fusion/xpu/fused_rope_utils.h b/paddle/phi/kernels/fusion/xpu/fused_rope_utils.h
@@ -32,7 +32,7 @@ void XPUGetSinCosData(const Context& dev_ctx,
         (dims_size == 2 || dims_size == 4),
         true,
         phi::errors::InvalidArgument("The dims of sin and cos is expected to "
-                                     "be 2 or 4, but recieved %d.",
+                                     "be 2 or 4, but received %d.",
                                      dims_size));
     if (dims_size == 4) {
       // sin.shape: [1, seq_len, 1, head_dim]
@@ -58,7 +58,7 @@ void XPUGetSinCosData(const Context& dev_ctx,
                         2,
                         phi::errors::InvalidArgument(
                             "The dims of position_ids is expected to "
-                            "be 2, but recieved %d.",
+                            "be 2, but received %d.",
                             position_ids_dims.size()));
 
       PADDLE_ENFORCE_EQ(
diff --git a/paddle/phi/kernels/gpu/flash_attn_utils.h b/paddle/phi/kernels/gpu/flash_attn_utils.h
@@ -62,7 +62,7 @@ static std::vector<int64_t> GetAttnMaskDims(const DenseTensor* attn_mask) {
         4,
         phi::errors::InvalidArgument(
             "The number of dimensions of attn_mask is expected to be greater "
-            "or equal to 4, but recieved %d. The shape of attn_mask is {%s}",
+            "or equal to 4, but received %d. The shape of attn_mask is {%s}",
             rank,
             origin_dims));
 
diff --git a/paddle/phi/kernels/gpu/take_along_axis_grad_kernel.cu b/paddle/phi/kernels/gpu/take_along_axis_grad_kernel.cu
@@ -54,7 +54,7 @@ void TakeAlongAxisGradKernel(const Context& dev_ctx,
   } else {
     PADDLE_THROW(
         phi::errors::InvalidArgument("The data type of input index is expected "
-                                     "to be int32 or int64, but recieved %s.",
+                                     "to be int32 or int64, but received %s.",
                                      phi::DataTypeToString(index_type)));
   }
 }
diff --git a/paddle/phi/kernels/gpu/take_along_axis_kernel.cu b/paddle/phi/kernels/gpu/take_along_axis_kernel.cu
@@ -41,7 +41,7 @@ void TakeAlongAxisKernel(const Context& dev_ctx,
   } else {
     PADDLE_THROW(
         phi::errors::InvalidArgument("The data type of input index is expected "
-                                     "to be int32 or int64, but recieved %s.",
+                                     "to be int32 or int64, but received %s.",
                                      phi::DataTypeToString(index_type)));
   }
 }
diff --git a/paddle/phi/kernels/gpudnn/conv_cudnn_v7.h b/paddle/phi/kernels/gpudnn/conv_cudnn_v7.h
@@ -698,7 +698,7 @@ struct SearchAlgorithm : public SearchAlgorithmBase<CK> {
         result.exhaustive_search = t.exhaustive_search;
       }
       if (!result.exhaustive_search) {
-        // In conv2d_tranpose, enable_autotune is set to false because some
+        // In conv2d_transpose, enable_autotune is set to false because some
         // algorithm picked by exhaustive search method produce wrong result.
         use_autotune = enable_autotune &&
                        phi::autotune::AutoTuneStatus::Instance().UseAutoTune();
diff --git a/paddle/phi/kernels/gpudnn/conv_gpudnn_base.h b/paddle/phi/kernels/gpudnn/conv_gpudnn_base.h
@@ -54,11 +54,11 @@ static size_t CalcWorkspaceLimitInBytes(bool use_fixed_workspace) {
         memory_utils::DeviceMemoryStatCurrentValue("Allocated", device_id);
     int64_t reserved =
         memory_utils::DeviceMemoryStatCurrentValue("Reserved", device_id);
-    int64_t availble = phi::backends::gpu::GpuAvailableMemToAlloc();
+    int64_t available = phi::backends::gpu::GpuAvailableMemToAlloc();
     VLOG(3) << "[memory] allocated=" << ToMegaBytes(allocated)
             << " MB, reserved=" << ToMegaBytes(reserved)
-            << " MB, available_to_alloc=" << ToMegaBytes(availble) << " MB.";
-    return std::max(availble, reserved - allocated);
+            << " MB, available_to_alloc=" << ToMegaBytes(available) << " MB.";
+    return std::max(available, reserved - allocated);
   } else {
     return FLAGS_conv_workspace_size_limit * 1024 * 1024;
   }
diff --git a/paddle/phi/kernels/gpudnn/softmax_gpudnn.h b/paddle/phi/kernels/gpudnn/softmax_gpudnn.h
@@ -829,7 +829,7 @@ void SwitchWarpSoftmaxBackward(const int blocks,
 
 /**
  * <NormalSoftmaxKernel>
- * Better performence when axis != -1
+ * Better performance when axis != -1
  */
 
 static void GetGridDim(

Original file line number	Diff line number	Diff line change
`@@ -55,7 +55,7 @@ void FusedRopeGradKernel(const Context& dev_ctx,`
`55`	`55`	`cos.get_ptr()->dims(),`
`56`	`56`	`phi::errors::InvalidArgument(`
`57`	`57`	`"The dims of sin and cos must be the same. But "`
`58`		`- "recieved sin's dims is {%s}, cos's dims is {%s}.",`
	`58`	`+ "received sin's dims is {%s}, cos's dims is {%s}.",`
`59`	`59`	`sin.get_ptr()->dims(),`
`60`	`60`	`cos.get_ptr()->dims()));`
`61`	`61`	`}`
Original file line number	Diff line number	Diff line change
`@@ -57,7 +57,7 @@ void FusedRopeKernel(const Context& dev_ctx,`
`57`	`57`	`cos.get_ptr()->dims(),`
`58`	`58`	`phi::errors::InvalidArgument(`
`59`	`59`	`"The dims of sin and cos must be the same. But "`
`60`		`- "recieved sin's dims is {%s}, cos's dims is {%s}.",`
	`60`	`+ "received sin's dims is {%s}, cos's dims is {%s}.",`
`61`	`61`	`sin.get_ptr()->dims(),`
`62`	`62`	`cos.get_ptr()->dims()));`
`63`	`63`	`}`
Original file line number	Diff line number	Diff line change
`@@ -54,7 +54,7 @@ void TakeAlongAxisGradKernel(const Context& dev_ctx,`
`54`	`54`	`} else {`
`55`	`55`	`PADDLE_THROW(`
`56`	`56`	`phi::errors::InvalidArgument("The data type of input index is expected "`
`57`		`- "to be int32 or int64, but recieved %s.",`
	`57`	`+ "to be int32 or int64, but received %s.",`
`58`	`58`	`phi::DataTypeToString(index_type)));`
`59`	`59`	`}`
`60`	`60`	`}`