llama_cpp 0.16.0 → 0.16.1

Sign up to get free protection for your applications and to get access to all the features.
Files changed (134) hide show
  1. checksums.yaml +4 -4
  2. data/CHANGELOG.md +6 -0
  3. data/ext/llama_cpp/extconf.rb +2 -0
  4. data/ext/llama_cpp/llama_cpp.cpp +2 -0
  5. data/lib/llama_cpp/version.rb +2 -2
  6. data/sig/llama_cpp.rbs +2 -0
  7. data/vendor/tmp/llama.cpp/Makefile +110 -53
  8. data/vendor/tmp/llama.cpp/ggml-alloc.c +78 -22
  9. data/vendor/tmp/llama.cpp/ggml-backend-impl.h +20 -8
  10. data/vendor/tmp/llama.cpp/ggml-backend.c +178 -64
  11. data/vendor/tmp/llama.cpp/ggml-backend.h +3 -3
  12. data/vendor/tmp/llama.cpp/ggml-blas.cpp +363 -0
  13. data/vendor/tmp/llama.cpp/ggml-blas.h +23 -0
  14. data/vendor/tmp/llama.cpp/ggml-common.h +6 -0
  15. data/vendor/tmp/llama.cpp/ggml-cuda/argsort.cu +1 -0
  16. data/vendor/tmp/llama.cpp/ggml-cuda/dmmv.cu +21 -9
  17. data/vendor/tmp/llama.cpp/ggml-cuda/fattn-tile-f16.cu +1 -1
  18. data/vendor/tmp/llama.cpp/ggml-cuda/mmq.cu +15 -1491
  19. data/vendor/tmp/llama.cpp/ggml-cuda/mmvq.cu +76 -61
  20. data/vendor/tmp/llama.cpp/ggml-cuda/quantize.cu +77 -10
  21. data/vendor/tmp/llama.cpp/ggml-cuda/softmax.cu +1 -0
  22. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-f16.cu +1 -1
  23. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q4_0.cu +1 -1
  24. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q4_1.cu +1 -1
  25. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q5_0.cu +1 -1
  26. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q5_1.cu +1 -1
  27. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q8_0.cu +1 -1
  28. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-f16.cu +1 -1
  29. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q4_0.cu +1 -1
  30. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q4_1.cu +1 -1
  31. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q5_0.cu +1 -1
  32. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q5_1.cu +1 -1
  33. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q8_0.cu +1 -1
  34. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-f16.cu +1 -1
  35. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q4_0.cu +1 -1
  36. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q4_1.cu +1 -1
  37. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q5_0.cu +1 -1
  38. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q5_1.cu +1 -1
  39. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q8_0.cu +1 -1
  40. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-f16.cu +1 -1
  41. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q4_0.cu +1 -1
  42. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q4_1.cu +1 -1
  43. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q5_0.cu +1 -1
  44. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q5_1.cu +1 -1
  45. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q8_0.cu +1 -1
  46. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-f16.cu +1 -1
  47. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q4_0.cu +1 -1
  48. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q4_1.cu +1 -1
  49. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q5_0.cu +1 -1
  50. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q5_1.cu +1 -1
  51. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q8_0.cu +1 -1
  52. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-f16.cu +1 -1
  53. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q4_0.cu +1 -1
  54. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q4_1.cu +1 -1
  55. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q5_0.cu +1 -1
  56. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q5_1.cu +1 -1
  57. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q8_0.cu +1 -1
  58. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs256-f16-f16.cu +1 -1
  59. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-f16.cu +1 -1
  60. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q4_0.cu +1 -1
  61. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q4_1.cu +1 -1
  62. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q5_0.cu +1 -1
  63. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q5_1.cu +1 -1
  64. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q8_0.cu +1 -1
  65. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-f16.cu +1 -1
  66. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q4_0.cu +1 -1
  67. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q4_1.cu +1 -1
  68. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q5_0.cu +1 -1
  69. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q5_1.cu +1 -1
  70. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q8_0.cu +1 -1
  71. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-f16.cu +1 -1
  72. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q4_0.cu +1 -1
  73. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q4_1.cu +1 -1
  74. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q5_0.cu +1 -1
  75. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q5_1.cu +1 -1
  76. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q8_0.cu +1 -1
  77. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-f16.cu +1 -1
  78. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q4_0.cu +1 -1
  79. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q4_1.cu +1 -1
  80. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q5_0.cu +1 -1
  81. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q5_1.cu +1 -1
  82. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q8_0.cu +1 -1
  83. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-f16.cu +1 -1
  84. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q4_0.cu +1 -1
  85. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q4_1.cu +1 -1
  86. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q5_0.cu +1 -1
  87. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q5_1.cu +1 -1
  88. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q8_0.cu +1 -1
  89. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-f16.cu +1 -1
  90. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q4_0.cu +1 -1
  91. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q4_1.cu +1 -1
  92. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q5_0.cu +1 -1
  93. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q5_1.cu +1 -1
  94. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q8_0.cu +1 -1
  95. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-f16.cu +1 -1
  96. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q4_0.cu +1 -1
  97. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q4_1.cu +1 -1
  98. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q5_0.cu +1 -1
  99. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q5_1.cu +1 -1
  100. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q8_0.cu +1 -1
  101. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs256-f16-f16.cu +1 -1
  102. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-f16.cu +1 -1
  103. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q4_0.cu +1 -1
  104. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q4_1.cu +1 -1
  105. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q5_0.cu +1 -1
  106. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q5_1.cu +1 -1
  107. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q8_0.cu +1 -1
  108. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-wmma-f16-instance-kqfloat-cpb16.cu +1 -1
  109. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-wmma-f16-instance-kqfloat-cpb32.cu +1 -1
  110. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-wmma-f16-instance-kqhalf-cpb16.cu +1 -1
  111. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-wmma-f16-instance-kqhalf-cpb32.cu +1 -1
  112. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/fattn-wmma-f16-instance-kqhalf-cpb8.cu +1 -1
  113. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/mmq-instance-q2_k.cu +5 -0
  114. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/mmq-instance-q3_k.cu +5 -0
  115. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/mmq-instance-q4_0.cu +5 -0
  116. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/mmq-instance-q4_1.cu +5 -0
  117. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/mmq-instance-q4_k.cu +5 -0
  118. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/mmq-instance-q5_0.cu +5 -0
  119. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/mmq-instance-q5_1.cu +5 -0
  120. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/mmq-instance-q5_k.cu +5 -0
  121. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/mmq-instance-q6_k.cu +5 -0
  122. data/vendor/tmp/llama.cpp/ggml-cuda/template-instances/mmq-instance-q8_0.cu +5 -0
  123. data/vendor/tmp/llama.cpp/ggml-cuda/unary.cu +20 -0
  124. data/vendor/tmp/llama.cpp/ggml-cuda.cu +95 -129
  125. data/vendor/tmp/llama.cpp/ggml-kompute.cpp +8 -7
  126. data/vendor/tmp/llama.cpp/ggml-metal.m +11 -9
  127. data/vendor/tmp/llama.cpp/ggml-rpc.cpp +13 -12
  128. data/vendor/tmp/llama.cpp/ggml-sycl.cpp +19 -23
  129. data/vendor/tmp/llama.cpp/ggml-vulkan-shaders.hpp +1230 -1129
  130. data/vendor/tmp/llama.cpp/ggml-vulkan.cpp +181 -148
  131. data/vendor/tmp/llama.cpp/ggml.c +102 -275
  132. data/vendor/tmp/llama.cpp/llama.cpp +103 -47
  133. data/vendor/tmp/llama.cpp/llama.h +4 -0
  134. metadata +15 -3
data/vendor/tmp/llama.cpp/ggml-sycl.cpp CHANGED
@@ -9108,6 +9108,7 @@ static void soft_max_f32(const float * x, const float * mask, float * dst, const
9108
9108
  // find the sum of exps in the block
9109
9109
  tmp = warp_reduce_sum(tmp, item_ct1);
9110
9110
  if (block_size > WARP_SIZE) {
9111
+ item_ct1.barrier(sycl::access::fence_space::local_space);
9111
9112
  if (warp_id == 0) {
9112
9113
  buf[lane_id] = 0.f;
9113
9114
  }
@@ -13088,10 +13089,12 @@ void *ggml_sycl_host_malloc(size_t size) try {
13088
13089
  return nullptr;
13089
13090
  }
13090
13091
 
13092
+ ggml_sycl_set_device(g_main_device);
13093
+ dpct::queue_ptr main_stream = g_syclStreams[g_main_device][0];
13094
+
13091
13095
  void * ptr = nullptr;
13092
- //allow to use dpct::get_in_order_queue() for host malloc
13093
13096
  dpct::err0 err = CHECK_TRY_ERROR(
13094
- ptr = (void *)sycl::malloc_host(size, dpct::get_in_order_queue()));
13097
+ ptr = (void *)sycl::malloc_host(size, *main_stream));
13095
13098
 
13096
13099
  if (err != 0) {
13097
13100
  // clear the error
@@ -13112,8 +13115,9 @@ catch (sycl::exception const &exc) {
13112
13115
  }
13113
13116
 
13114
13117
  void ggml_sycl_host_free(void *ptr) try {
13115
- //allow to use dpct::get_in_order_queue() for host malloc
13116
- SYCL_CHECK(CHECK_TRY_ERROR(sycl::free(ptr, dpct::get_in_order_queue())));
13118
+ ggml_sycl_set_device(g_main_device);
13119
+ dpct::queue_ptr main_stream = g_syclStreams[g_main_device][0];
13120
+ SYCL_CHECK(CHECK_TRY_ERROR(sycl::free(ptr, *main_stream)));
13117
13121
  }
13118
13122
  catch (sycl::exception const &exc) {
13119
13123
  std::cerr << exc.what() << "Exception caught at file:" << __FILE__
@@ -16571,22 +16575,12 @@ GGML_CALL static size_t ggml_backend_sycl_buffer_type_get_alloc_size(ggml_backen
16571
16575
  UNUSED(buft);
16572
16576
  }
16573
16577
 
16574
- GGML_CALL static bool ggml_backend_sycl_buffer_type_supports_backend(ggml_backend_buffer_type_t buft, ggml_backend_t backend) {
16575
- if (!ggml_backend_is_sycl(backend)) {
16576
- return false;
16577
- }
16578
- ggml_backend_sycl_buffer_type_context * buft_ctx = (ggml_backend_sycl_buffer_type_context *)buft->context;
16579
- ggml_backend_sycl_context * sycl_ctx = (ggml_backend_sycl_context *)backend->context;
16580
- return buft_ctx->device == sycl_ctx->device;
16581
- }
16582
-
16583
16578
  static ggml_backend_buffer_type_i ggml_backend_sycl_buffer_type_interface = {
16584
16579
  /* .get_name = */ ggml_backend_sycl_buffer_type_name,
16585
16580
  /* .alloc_buffer = */ ggml_backend_sycl_buffer_type_alloc_buffer,
16586
16581
  /* .get_alignment = */ ggml_backend_sycl_buffer_type_get_alignment,
16587
16582
  /* .get_max_size = */ ggml_backend_sycl_buffer_type_get_max_size,
16588
16583
  /* .get_alloc_size = */ ggml_backend_sycl_buffer_type_get_alloc_size,
16589
- /* .supports_backend = */ ggml_backend_sycl_buffer_type_supports_backend,
16590
16584
  /* .is_host = */ nullptr,
16591
16585
  };
16592
16586
 
@@ -16938,12 +16932,6 @@ GGML_CALL static size_t ggml_backend_sycl_split_buffer_type_get_alloc_size(ggml_
16938
16932
  return total_size;
16939
16933
  }
16940
16934
 
16941
- GGML_CALL static bool ggml_backend_sycl_split_buffer_type_supports_backend(ggml_backend_buffer_type_t buft, ggml_backend_t backend) {
16942
- return ggml_backend_is_sycl(backend);
16943
-
16944
- UNUSED(buft);
16945
- }
16946
-
16947
16935
  GGML_CALL static bool ggml_backend_sycl_split_buffer_type_is_host(ggml_backend_buffer_type_t buft) {
16948
16936
  return false;
16949
16937
 
@@ -16956,7 +16944,6 @@ static ggml_backend_buffer_type_i ggml_backend_sycl_split_buffer_type_interface
16956
16944
  /* .get_alignment = */ ggml_backend_sycl_split_buffer_type_get_alignment,
16957
16945
  /* .get_max_size = */ NULL, // defaults to SIZE_MAX
16958
16946
  /* .get_alloc_size = */ ggml_backend_sycl_split_buffer_type_get_alloc_size,
16959
- /* .supports_backend = */ ggml_backend_sycl_split_buffer_type_supports_backend,
16960
16947
  /* .is_host = */ ggml_backend_sycl_split_buffer_type_is_host,
16961
16948
  };
16962
16949
 
@@ -17042,7 +17029,6 @@ ggml_backend_buffer_type_t ggml_backend_sycl_host_buffer_type() {
17042
17029
  /* .get_alignment = */ ggml_backend_cpu_buffer_type()->iface.get_alignment,
17043
17030
  /* .get_max_size = */ NULL, // TODO: return device.maxBufferLength
17044
17031
  /* .get_alloc_size = */ ggml_backend_cpu_buffer_type()->iface.get_alloc_size,
17045
- /* .supports_backend = */ ggml_backend_cpu_buffer_type()->iface.supports_backend,
17046
17032
  /* .is_host = */ ggml_backend_cpu_buffer_type()->iface.is_host,
17047
17033
  },
17048
17034
  /* .context = */ nullptr,
@@ -17186,7 +17172,7 @@ GGML_CALL static bool ggml_backend_sycl_supports_op(ggml_backend_t backend, cons
17186
17172
  case GGML_UNARY_OP_HARDSWISH:
17187
17173
  case GGML_UNARY_OP_GELU_QUICK:
17188
17174
  case GGML_UNARY_OP_TANH:
17189
- return true;
17175
+ return ggml_is_contiguous(op->src[0]);
17190
17176
  default:
17191
17177
  return false;
17192
17178
  }
@@ -17307,6 +17293,14 @@ GGML_CALL static bool ggml_backend_sycl_offload_op(ggml_backend_t backend, const
17307
17293
  GGML_UNUSED(backend);
17308
17294
  }
17309
17295
 
17296
+ GGML_CALL static bool ggml_backend_sycl_supports_buft(ggml_backend_t backend, ggml_backend_buffer_type_t buft) {
17297
+ if (buft->iface.get_name != ggml_backend_sycl_buffer_type_name) {
17298
+ return false;
17299
+ }
17300
+ ggml_backend_sycl_buffer_type_context * buft_ctx = (ggml_backend_sycl_buffer_type_context *)buft->context;
17301
+ ggml_backend_sycl_context * sycl_ctx = (ggml_backend_sycl_context *)backend->context;
17302
+ return buft_ctx->device == sycl_ctx->device;
17303
+ }
17310
17304
 
17311
17305
  static ggml_backend_i ggml_backend_sycl_interface = {
17312
17306
  /* .get_name = */ ggml_backend_sycl_name,
@@ -17318,9 +17312,11 @@ static ggml_backend_i ggml_backend_sycl_interface = {
17318
17312
  /* .synchronize = */ ggml_backend_sycl_synchronize,
17319
17313
  /* .graph_plan_create = */ NULL,
17320
17314
  /* .graph_plan_free = */ NULL,
17315
+ /* .graph_plan_update = */ NULL,
17321
17316
  /* .graph_plan_compute = */ NULL,
17322
17317
  /* .graph_compute = */ ggml_backend_sycl_graph_compute,
17323
17318
  /* .supports_op = */ ggml_backend_sycl_supports_op,
17319
+ /* .supports_buft = */ ggml_backend_sycl_supports_buft,
17324
17320
  /* .offload_op = */ ggml_backend_sycl_offload_op,
17325
17321
  /* .event_new = */ NULL,
17326
17322
  /* .event_free = */ NULL,