ggml: fix build break for the vulkan-debug.

[cyzero-kim]

• windows build : Ok.

• linux build : Ok.

• I have read the contributing guidelines

• Self-reported review complexity:

  • Low
  • Medium
  • High

This is a minor patch that fixes an error occurring in the Vulkan debug build.

Case : cmake .. -DGGML_VULKAN:BOOL=ON -DGGML_VULKAN_DEBUG:BOOL=ON

[ Windows ]

[ Linux ]

[  9%] Building C object ggml/src/CMakeFiles/ggml.dir/ggml-aarch64.c.o
/mnt/hdd_500gb/work/llama/llama.cpp/ggml/src/ggml-vulkan.cpp: In function ‘void ggml_vk_dispatch_pipeline(ggml_backend_vk_context*, vk_context&, vk_pipeline&, const std::initializer_list<vk::DescriptorBufferInfo>&, size_t, const void*, std::array<unsigned int, 3>)’:
/mnt/hdd_500gb/work/llama/llama.cpp/ggml/src/ggml-vulkan.cpp:2483:26: error: no match for ‘operator<<’ (operand types are ‘std::basic_ostream<char>’ and ‘const vk::DescriptorBufferInfo’)
 2483 |         std::cerr << "(" << buffer << ", " << buffer.offset << ", " << buffer.size << "), ";
      |         ~~~~~~~~~~~~~~~~ ^~ ~~~~~~
      |                   |         |
      |                   |         const vk::DescriptorBufferInfo
      |                   std::basic_ostream<char>
/mnt/hdd_500gb/work/llama/llama.cpp/ggml/src/ggml-vulkan.cpp:61:40: note: in definition of macro ‘VK_LOG_DEBUG’
   61 | #define VK_LOG_DEBUG(msg) std::cerr << msg << std::endl
      |                                        ^~~
In file included from /usr/include/c++/11/istream:39,
                 from /usr/include/c++/11/sstream:38,
                 from /usr/include/vulkan/vulkan_to_string.hpp:16,
                 from /usr/include/vulkan/vulkan.hpp:6110,
                 from /mnt/hdd_500gb/work/llama/llama.cpp/ggml/src/ggml-vulkan.cpp:7:

...

/mnt/hdd_500gb/work/llama/llama.cpp/ggml/src/ggml-vulkan.cpp:2483:79: error: ‘const struct vk::DescriptorBufferInfo’ has no member named ‘size’
 2483 |         std::cerr << "(" << buffer << ", " << buffer.offset << ", " << buffer.size << "), ";
      |                                                                               ^~~~
/mnt/hdd_500gb/work/llama/llama.cpp/ggml/src/ggml-vulkan.cpp:61:40: note: in definition of macro ‘VK_LOG_DEBUG’
   61 | #define VK_LOG_DEBUG(msg) std::cerr << msg << std::endl

[ VulkanSDK 1.3.283.0 ]
C:\VulkanSDK\1.3.283.0\Include\vulkan\vulkan_structs.hpp

  struct DescriptorBufferInfo
  {
    using NativeType = VkDescriptorBufferInfo;

#if !defined( VULKAN_HPP_NO_STRUCT_CONSTRUCTORS )
    VULKAN_HPP_CONSTEXPR DescriptorBufferInfo( VULKAN_HPP_NAMESPACE::Buffer     buffer_ = {},
                                               VULKAN_HPP_NAMESPACE::DeviceSize offset_ = {},
                                               VULKAN_HPP_NAMESPACE::DeviceSize range_  = {} ) VULKAN_HPP_NOEXCEPT
      : buffer( buffer_ )
      , offset( offset_ )
      , range( range_ )
    {
    }
...

[ logs ]

ggml_vk_dispatch_pipeline(mul_mat_vec_q6_k_f32_f32, {(0000027EF0332248, 1889308416, 48168960), (0000027EF11B37C8, 86144, 57344), (0000027EF11B37C8, 0,
 16384), }, (4096,1,1))
ggml_vk_build_graph(0000027EE5273270, ADD)
ggml_vk_op_f32((0000027EE5273100, name=ffn_out-31, type=0, ne0=4096, ne1=1, ne2=1, ne3=1, nb0=4, nb1=16384, nb2=16384, nb3=16384), (0000027EE52726F0,
name=ffn_inp-31, type=0, ne0=4096, ne1=1, ne2=1, ne3=1, nb0=4, nb1=16384, nb2=16384, nb3=16384), (0000027EE5273270, name=l_out-31, type=0, ne0=4096, n
e1=1, ne2=1, ne3=1, nb0=4, nb1=16384, nb2=16384, nb3=16384), ADD, )
ggml_vk_sync_buffers()
ggml_vk_dispatch_pipeline(add_f32, {(0000027EF11B37C8, 0, 16384), (0000027EF11B37C8, 69760, 16384), (0000027EF11B37C8, 0, 16384), }, (1,8,1))
ggml_vk_build_graph(0000027EE52733E0, RMS_NORM)
ggml_vk_op_f32((0000027EE5273270, name=l_out-31, type=0, ne0=4096, ne1=1, ne2=1, ne3=1, nb0=4, nb1=16384, nb2=16384, nb3=16384), (0000027EE52733E0, na
me=norm, type=0, ne0=4096, ne1=1, ne2=1, ne3=1, nb0=4, nb1=16384, nb2=16384, nb3=16384), RMS_NORM, )
ggml_vk_sync_buffers()
ggml_vk_dispatch_pipeline(rms_norm_f32, {(0000027EF11B37C8, 0, 16384), (0000027EF11B37C8, 0, 16384), }, (1,1,1))

[slaren]

The same fix was merged in ggml-org/ggml#948, so there may be some conflicts when synchronizing the repositories.