Fix set_value bug and chang kernel (#1835)

Author: duqimeng

backends/metax_gpu/kernels/cuda_kernels/c_embedding_grad_kernel_register.cu

@@ -23,6 +23,7 @@ PD_CUSTOM_KERNEL_REGISTER(c_embedding_grad,
                           phi::CEmbeddingGradKernel,
                           float,
                           double,
+                          phi::dtype::bfloat16,
                           phi::dtype::float16,
                           phi::dtype::complex<float>,
                           phi::dtype::complex<double>) {}

backends/metax_gpu/kernels/cuda_kernels/c_embedding_kernel_register.cu

@@ -23,6 +23,7 @@ PD_CUSTOM_KERNEL_REGISTER(c_embedding,
                           phi::CEmbeddingKernel,
                           float,
                           double,
+                          phi::dtype::bfloat16,
                           phi::dtype::float16,
                           phi::dtype::complex<float>,
                           phi::dtype::complex<double>) {}

backends/metax_gpu/kernels/cuda_kernels/cast_kernel_register.cu

@@ -16,7 +16,7 @@
 #include "paddle/phi/kernels/cast_kernel.h"
 
 PD_CUSTOM_KERNEL_REGISTER(cast,
-                          iluvatar_gpu,
+                          metax_gpu,
                           ALL_LAYOUT,
                           phi::CastKernel,
                           float,

backends/metax_gpu/kernels/cuda_kernels/set_value_kernel_register.cu

@@ -20,7 +20,7 @@
 PD_CUSTOM_KERNEL_REGISTER(set_value,
                           metax_gpu,
                           ALL_LAYOUT,
-                          phi::SetValueKernelV2,
+                          phi::SetValueKernel,
                           float,
                           double,
                           int,
@@ -36,7 +36,7 @@ PD_CUSTOM_KERNEL_REGISTER(set_value,
 PD_CUSTOM_KERNEL_REGISTER(set_value_with_tensor,
                           metax_gpu,
                           ALL_LAYOUT,
-                          phi::SetTensorValueKernelV2,
+                          phi::SetTensorValueKernel,
                           float,
                           double,
                           int,

…etax_kerenl/blha_get_max_len_register.cu …etax_kernel/blha_get_max_len_register.cubackends/metax_gpu/kernels/metax_kerenl/blha_get_max_len_register.cu renamed to backends/metax_gpu/kernels/metax_kernel/blha_get_max_len_register.cu backends/metax_gpu/kernels/metax_kerenl/blha_get_max_len_register.cu renamed to backends/metax_gpu/kernels/metax_kernel/blha_get_max_len_register.cu

@@ -13,7 +13,7 @@
 // limitations under the License.
 
 #include "kernels/flash_attn_utils.h"
-#include "kernels/metax_kerenl/block_attn.h"
+#include "kernels/metax_kernel/block_attn.h"
 #include "paddle/phi/backends/context_pool.h"
 #include "paddle/phi/core/dense_tensor.h"
 #include "paddle/phi/core/kernel_registry.h"

…ax_gpu/kernels/metax_kerenl/block_attn.h …ax_gpu/kernels/metax_kernel/block_attn.hbackends/metax_gpu/kernels/metax_kerenl/block_attn.h renamed to backends/metax_gpu/kernels/metax_kernel/block_attn.h backends/metax_gpu/kernels/metax_kerenl/block_attn.h renamed to backends/metax_gpu/kernels/metax_kernel/block_attn.h

@@ -15,7 +15,7 @@
 #pragma once
 
 #include "kernels/funcs/quant_dequant.h"
-#include "kernels/metax_kerenl/mmha_util.cu.h"
+#include "kernels/metax_kernel/mmha_util.cu.h"
 #include "paddle/common/flags.h"
 #include "paddle/phi/backends/gpu/gpu_context.h"
 #include "paddle/phi/common/memory_utils.h"

…_gpu/kernels/metax_kerenl/mmha_util.cu.h …_gpu/kernels/metax_kernel/mmha_util.cu.hbackends/metax_gpu/kernels/metax_kerenl/mmha_util.cu.h renamed to backends/metax_gpu/kernels/metax_kernel/mmha_util.cu.h backends/metax_gpu/kernels/metax_kerenl/mmha_util.cu.h renamed to backends/metax_gpu/kernels/metax_kernel/mmha_util.cu.h

@@ -30,7 +30,7 @@ def test_api_fp16(self):
             limit = paddle.to_tensor([3, 2], dtype="float16")
             out = paddle.greater_equal(x=label, y=limit)
             if core.is_compiled_with_cuda():
-                place = paddle.CustomPlace("iluvatar_gpu", 0)
+                place = paddle.CustomPlace("metax_gpu", 0)
                 exe = static.Executor(place)
                 (res,) = exe.run(fetch_list=[out])
                 self.assertEqual((res == np.array([True, True])).all(), True)