fix temporary bug (#752)

Author: awni

mlx/backend/metal/conv.cpp

@@ -31,7 +31,8 @@ void explicit_gemm_conv_1D_gpu(
   array in_padded(padded_shape, in.dtype(), nullptr, {});
 
   // Fill with zeros
-  copy_gpu(array(0, in.dtype()), in_padded, CopyType::Scalar, s);
+  auto zero = array(0, in.dtype());
+  copy_gpu(zero, in_padded, CopyType::Scalar, s);
 
   // Pick input slice from padded
   size_t data_offset = conv_params.pad[0] * in_padded.strides()[1];
@@ -68,7 +69,7 @@ void explicit_gemm_conv_1D_gpu(
   copy_gpu(in_strided_view, in_strided, CopyType::General, s);
 
   // Perform gemm
-  std::vector<array> copies = {in_padded, in_strided};
+  std::vector<array> copies = {zero, in_padded, in_strided};
   return steel_matmul(
       s,
       d,
@@ -213,6 +214,7 @@ void explicit_gemm_conv_2D_gpu(
   array in_padded(padded_shape, in.dtype(), nullptr, {});
 
   // Fill with zeros
+  auto zero = array(0, in.dtype());
   copy_gpu(array(0, in.dtype()), in_padded, CopyType::Scalar, s);
 
   // Pick input slice from padded
@@ -259,7 +261,7 @@ void explicit_gemm_conv_2D_gpu(
   copy_gpu(in_strided_view, in_strided, CopyType::General, s);
 
   // Perform gemm
-  std::vector<array> copies = {in_padded, in_strided};
+  std::vector<array> copies = {zero, in_padded, in_strided};
   return steel_matmul(
       s,
       d,