support OP EXPM1, support all UT cases of FLOOR, TRUNC, ROUND

Author: arthw

ggml/src/ggml-sycl/element_wise.cpp

@@ -124,6 +124,11 @@ static __dpct_inline__ T op_exp(T x) {
     return sycl::exp(x);
 }
 
+template<typename T>
+static __dpct_inline__ T op_expm1(T x) {
+    return sycl::expm1(x);
+}
+
 template<typename T>
 static __dpct_inline__ T op_log(T x) {
     if (x <= static_cast<T>(0)) {
@@ -605,6 +610,12 @@ static inline void ggml_sycl_op_exp(ggml_backend_sycl_context & ctx, ggml_tensor
     });
 }
 
+static inline void ggml_sycl_op_expm1(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
+    ggml_sycl_detail::ggml_sycl_op_unary(ctx, dst, [](auto x) {
+        return op_expm1(x);
+    });
+}
+
 static inline void ggml_sycl_op_log(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
     ggml_sycl_detail::dispatch_ggml_sycl_op_unary(ctx, dst,
         [](const auto* src, auto* dst_ptr, int k_elements, queue_ptr stream) {
@@ -728,16 +739,9 @@ static inline void ggml_sycl_op_clamp(ggml_backend_sycl_context & ctx, ggml_tens
 }
 
 static inline void ggml_sycl_op_floor(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
-    ggml_sycl_detail::dispatch_ggml_sycl_op_unary(ctx, dst,
-        [](const auto* src, auto* dst_ptr, int k_elements, queue_ptr stream) {
-            const int num_blocks = ceil_div(k_elements, 256);
-            stream->parallel_for(
-                sycl::nd_range<1>(sycl::range<1>(num_blocks) * sycl::range<1>(256),
-                                  sycl::range<1>(256)),
-                [=](sycl::nd_item<1> item_ct1) {
-                    unary_op_floor_kernel(src, dst_ptr, k_elements, item_ct1);
-                });
-        });
+    ggml_sycl_detail::ggml_sycl_op_unary(ctx, dst, [](auto x) {
+        return op_floor(x);
+    });
 }
 
 static inline void ggml_sycl_op_ceil(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
@@ -747,29 +751,15 @@ static inline void ggml_sycl_op_ceil(ggml_backend_sycl_context & ctx, ggml_tenso
 }
 
 static inline void ggml_sycl_op_round(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
-    ggml_sycl_detail::dispatch_ggml_sycl_op_unary(ctx, dst,
-        [](const auto* src, auto* dst_ptr, int k_elements, queue_ptr stream) {
-            const int num_blocks = ceil_div(k_elements, 256);
-            stream->parallel_for(
-                sycl::nd_range<1>(sycl::range<1>(num_blocks) * sycl::range<1>(256),
-                                  sycl::range<1>(256)),
-                [=](sycl::nd_item<1> item_ct1) {
-                    unary_op_round_kernel(src, dst_ptr, k_elements, item_ct1);
-                });
-        });
+    ggml_sycl_detail::ggml_sycl_op_unary(ctx, dst, [](auto x) {
+        return op_round(x);
+    });
 }
 
 static inline void ggml_sycl_op_trunc(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
-    ggml_sycl_detail::dispatch_ggml_sycl_op_unary(ctx, dst,
-        [](const auto* src, auto* dst_ptr, int k_elements, queue_ptr stream) {
-            const int num_blocks = ceil_div(k_elements, 256);
-            stream->parallel_for(
-                sycl::nd_range<1>(sycl::range<1>(num_blocks) * sycl::range<1>(256),
-                                  sycl::range<1>(256)),
-                [=](sycl::nd_item<1> item_ct1) {
-                    unary_op_trunc_kernel(src, dst_ptr, k_elements, item_ct1);
-                });
-        });
+    ggml_sycl_detail::ggml_sycl_op_unary(ctx, dst, [](auto x) {
+        return op_trunc(x);
+    });
 }
 
 static inline void ggml_sycl_op_acc(ggml_backend_sycl_context & ctx, ggml_tensor *dst) {
@@ -1018,6 +1008,11 @@ void ggml_sycl_exp(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
     ggml_sycl_op_exp(ctx, dst);
 }
 
+void ggml_sycl_expm1(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
+    scope_op_debug_print scope_dbg_print(__func__, dst, /*num_src=*/1);
+    ggml_sycl_op_expm1(ctx, dst);
+}
+
 void ggml_sycl_log(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
     scope_op_debug_print scope_dbg_print(__func__, dst, /*num_src=*/1);
     ggml_sycl_op_log(ctx, dst);

ggml/src/ggml-sycl/element_wise.hpp

@@ -59,6 +59,8 @@ void ggml_sycl_hardswish(ggml_backend_sycl_context & ctx, ggml_tensor * dst);
 
 void ggml_sycl_exp(ggml_backend_sycl_context & ctx, ggml_tensor * dst);
 
+void ggml_sycl_expm1(ggml_backend_sycl_context & ctx, ggml_tensor * dst);
+
 void ggml_sycl_log(ggml_backend_sycl_context & ctx, ggml_tensor * dst);
 
 void ggml_sycl_softplus(ggml_backend_sycl_context & ctx, ggml_tensor * dst);

ggml/src/ggml-sycl/ggml-sycl.cpp

@@ -4592,6 +4592,9 @@ static bool ggml_sycl_compute_forward(ggml_backend_sycl_context & ctx, struct gg
                 case GGML_UNARY_OP_EXP:
                     ggml_sycl_exp(ctx, dst);
                     break;
+                case GGML_UNARY_OP_EXPM1:
+                    ggml_sycl_expm1(ctx, dst);
+                    break;
                 case GGML_UNARY_OP_SOFTPLUS:
                     ggml_sycl_softplus(ctx, dst);
                     break;
@@ -5238,18 +5241,15 @@ static bool ggml_backend_sycl_device_supports_op(ggml_backend_dev_t dev, const g
                 case GGML_UNARY_OP_GELU_QUICK:
                 case GGML_UNARY_OP_GELU_ERF:
                 case GGML_UNARY_OP_EXP:
+                case GGML_UNARY_OP_EXPM1:
                 case GGML_UNARY_OP_SOFTPLUS:
                 case GGML_UNARY_OP_ELU:
                 case GGML_UNARY_OP_CEIL:
                     return true;
                 case GGML_UNARY_OP_FLOOR:
                 case GGML_UNARY_OP_ROUND:
                 case GGML_UNARY_OP_TRUNC:
-#if defined (GGML_SYCL_F16)
-                    return ggml_is_contiguous(op->src[0]) && (op->type == op->src[0]->type);
-#else
-                    return ggml_is_contiguous(op->src[0]) && (op->src[0]->type == GGML_TYPE_F32 && op->type == GGML_TYPE_F32) && (op->type == op->src[0]->type);
-#endif
+                    return true;
                 default:
                     return false;
             }