tests : add non-cont unary tests (llama/7857)

* tests : add non-cont unary tests

* ggml : update unary asserts and "supports_op"

ggml-ci

src/ggml-cuda.cu

@@ -2740,7 +2740,7 @@ GGML_CALL static bool ggml_backend_cuda_supports_op(ggml_backend_t backend, cons
  case GGML_UNARY_OP_HARDSWISH:
  case GGML_UNARY_OP_GELU_QUICK:
  case GGML_UNARY_OP_TANH:
- return true;
+ return ggml_is_contiguous(op->src[0]);
  default:
  return false;
  }

src/ggml-cuda/unary.cu

@@ -148,6 +148,8 @@ void ggml_cuda_op_gelu(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
  float * dst_d = (float *)dst->data;
  cudaStream_t stream = ctx.stream();
 
+ GGML_ASSERT(ggml_is_contiguous(src0));
+
  GGML_ASSERT(src0->type == GGML_TYPE_F32);
  GGML_ASSERT( dst->type == GGML_TYPE_F32);
 
@@ -160,6 +162,8 @@ void ggml_cuda_op_silu(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
  float * dst_d = (float *)dst->data;
  cudaStream_t stream = ctx.stream();
 
+ GGML_ASSERT(ggml_is_contiguous(src0));
+
  GGML_ASSERT(src0->type == GGML_TYPE_F32);
  GGML_ASSERT( dst->type == GGML_TYPE_F32);
 
@@ -172,6 +176,8 @@ void ggml_cuda_op_gelu_quick(ggml_backend_cuda_context & ctx, ggml_tensor * dst)
  float * dst_d = (float *)dst->data;
  cudaStream_t stream = ctx.stream();
 
+ GGML_ASSERT(ggml_is_contiguous(src0));
+
  GGML_ASSERT(src0->type == GGML_TYPE_F32);
  GGML_ASSERT( dst->type == GGML_TYPE_F32);
 
@@ -184,6 +190,8 @@ void ggml_cuda_op_tanh(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
  float * dst_d = (float *)dst->data;
  cudaStream_t stream = ctx.stream();
 
+ GGML_ASSERT(ggml_is_contiguous(src0));
+
  GGML_ASSERT(src0->type == GGML_TYPE_F32);
  GGML_ASSERT( dst->type == GGML_TYPE_F32);
 
@@ -196,6 +204,8 @@ void ggml_cuda_op_relu(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
  float * dst_d = (float *)dst->data;
  cudaStream_t stream = ctx.stream();
 
+ GGML_ASSERT(ggml_is_contiguous(src0));
+
  GGML_ASSERT(src0->type == GGML_TYPE_F32);
  GGML_ASSERT( dst->type == GGML_TYPE_F32);
 
@@ -208,6 +218,8 @@ void ggml_cuda_op_sigmoid(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
  float * dst_d = (float *)dst->data;
  cudaStream_t stream = ctx.stream();
 
+ GGML_ASSERT(ggml_is_contiguous(src0));
+
  GGML_ASSERT(src0->type == GGML_TYPE_F32);
  GGML_ASSERT( dst->type == GGML_TYPE_F32);
 
@@ -220,6 +232,8 @@ void ggml_cuda_op_hardsigmoid(ggml_backend_cuda_context & ctx, ggml_tensor * dst
  float * dst_d = (float *)dst->data;
  cudaStream_t stream = ctx.stream();
 
+ GGML_ASSERT(ggml_is_contiguous(src0));
+
  GGML_ASSERT(src0->type == GGML_TYPE_F32);
  GGML_ASSERT( dst->type == GGML_TYPE_F32);
 
@@ -232,6 +246,8 @@ void ggml_cuda_op_hardswish(ggml_backend_cuda_context & ctx, ggml_tensor * dst)
  float * dst_d = (float *)dst->data;
  cudaStream_t stream = ctx.stream();
 
+ GGML_ASSERT(ggml_is_contiguous(src0));
+
  GGML_ASSERT(src0->type == GGML_TYPE_F32);
  GGML_ASSERT( dst->type == GGML_TYPE_F32);
 
@@ -244,6 +260,8 @@ void ggml_cuda_op_leaky_relu(ggml_backend_cuda_context & ctx, ggml_tensor * dst)
  float * dst_d = (float *)dst->data;
  cudaStream_t stream = ctx.stream();
 
+ GGML_ASSERT(ggml_is_contiguous(src0));
+
  GGML_ASSERT(src0->type == GGML_TYPE_F32);
  GGML_ASSERT( dst->type == GGML_TYPE_F32);
 
@@ -259,6 +277,8 @@ void ggml_cuda_op_sqr(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
  float * dst_d = (float *)dst->data;
  cudaStream_t stream = ctx.stream();
 
+ GGML_ASSERT(ggml_is_contiguous(src0));
+
  GGML_ASSERT(src0->type == GGML_TYPE_F32);
  GGML_ASSERT( dst->type == GGML_TYPE_F32);
 

src/ggml-kompute.cpp

@@ -1340,7 +1340,7 @@ static bool ggml_vk_supports_op(const struct ggml_tensor * op) {
  case GGML_UNARY_OP_RELU:
  case GGML_UNARY_OP_GELU:
  case GGML_UNARY_OP_SILU:
- return true;
+ return ggml_is_contiguous(op->src[0]);
  default:
  ;
  }

src/ggml-metal.m

@@ -744,7 +744,7 @@ static bool ggml_metal_supports_op(const struct ggml_metal_context * ctx, const
  case GGML_UNARY_OP_GELU:
  case GGML_UNARY_OP_GELU_QUICK:
  case GGML_UNARY_OP_SILU:
- return true;
+ return ggml_is_contiguous(op->src[0]);
  default:
  return false;
  }

src/ggml-sycl.cpp

@@ -17190,7 +17190,7 @@ GGML_CALL static bool ggml_backend_sycl_supports_op(ggml_backend_t backend, cons
  case GGML_UNARY_OP_HARDSWISH:
  case GGML_UNARY_OP_GELU_QUICK:
  case GGML_UNARY_OP_TANH:
- return true;
+ return ggml_is_contiguous(op->src[0]);
  default:
  return false;
  }

src/ggml-vulkan.cpp

@@ -6439,7 +6439,7 @@ GGML_CALL static bool ggml_backend_vk_supports_op(ggml_backend_t backend, const
  case GGML_UNARY_OP_GELU:
  case GGML_UNARY_OP_SILU:
  case GGML_UNARY_OP_RELU:
- return true;
+ return ggml_is_contiguous(op->src[0]);
  default:
  return false;
  }