ggml webgpu: add CEIL operation support (#18605)

* ggml-webgpu: add CEIL operation support

      Add support for the CEIL unary operation in the WebGPU backend:
      - Add CEIL_FUNC shader template in unary_op.wgsl
      - Add 4 shader variants (f32, f16, inplace versions)
      - Initialize CEIL pipelines in ggml-webgpu.cpp
      - Register CEIL in supports_op function

* docs: update WebGPU ops support for CEIL

docs/ops.md

@@ -22,7 +22,7 @@ Legend:
 |                           ARANGE | ❌ | ✅ | ✅ | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
 |                           ARGMAX | ❌ | ✅ | ✅ | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
 |                          ARGSORT | ❌ | ✅ | ✅ | ✅ | ✅ | 🟡 | 🟡 | ✅ | ❌ | ❌ | ❌ |
-|                             CEIL | ❌ | ❌ | ✅ | 🟡 | ❌ | ❌ | 🟡 | 🟡 | ❌ | ❌ | ❌ |
+|                             CEIL | ❌ | ❌ | ✅ | 🟡 | ❌ | ❌ | 🟡 | 🟡 | ✅ | ❌ | ❌ |
 |                            CLAMP | ❌ | ✅ | ✅ | ✅ | 🟡 | 🟡 | ✅ | 🟡 | ❌ | ❌ | ❌ |
 |                           CONCAT | ❌ | ✅ | ✅ | 🟡 | ✅ | 🟡 | ✅ | ✅ | ❌ | ❌ | ❌ |
 |                             CONT | ❌ | 🟡 | ✅ | ✅ | ✅ | 🟡 | 🟡 | ✅ | 🟡 | ❌ | ❌ |

ggml/src/ggml-webgpu/ggml-webgpu.cpp

@@ -2273,6 +2273,16 @@ static void ggml_webgpu_init_unary_pipeline(webgpu_context & webgpu_ctx) {
         ggml_webgpu_create_pipeline(webgpu_ctx->device, wgsl_xielu_inplace_f32, "xielu_inplace_f32", constants);
     webgpu_ctx->unary_pipelines[GGML_UNARY_OP_XIELU][GGML_TYPE_F16][1] =
         ggml_webgpu_create_pipeline(webgpu_ctx->device, wgsl_xielu_inplace_f16, "xielu_inplace_f16", constants);
+
+    // CEIL
+    webgpu_ctx->unary_pipelines[GGML_UNARY_OP_CEIL][GGML_TYPE_F32][0] =
+        ggml_webgpu_create_pipeline(webgpu_ctx->device, wgsl_ceil_f32, "ceil_f32", constants);
+    webgpu_ctx->unary_pipelines[GGML_UNARY_OP_CEIL][GGML_TYPE_F16][0] =
+        ggml_webgpu_create_pipeline(webgpu_ctx->device, wgsl_ceil_f16, "ceil_f16", constants);
+    webgpu_ctx->unary_pipelines[GGML_UNARY_OP_CEIL][GGML_TYPE_F32][1] =
+        ggml_webgpu_create_pipeline(webgpu_ctx->device, wgsl_ceil_inplace_f32, "ceil_inplace_f32", constants);
+    webgpu_ctx->unary_pipelines[GGML_UNARY_OP_CEIL][GGML_TYPE_F16][1] =
+        ggml_webgpu_create_pipeline(webgpu_ctx->device, wgsl_ceil_inplace_f16, "ceil_inplace_f16", constants);
 }
 
 static void ggml_webgpu_init_scale_pipeline(webgpu_context & webgpu_ctx) {
@@ -2528,6 +2538,7 @@ static bool ggml_backend_webgpu_device_supports_op(ggml_backend_dev_t dev, const
                     case GGML_UNARY_OP_EXP:
                     case GGML_UNARY_OP_GELU_ERF:
                     case GGML_UNARY_OP_XIELU:
+                    case GGML_UNARY_OP_CEIL:
                         supports_op = supports_op =
                             (op->type == GGML_TYPE_F32 || op->type == GGML_TYPE_F16) && (src0->type == op->type);
                         break;

ggml/src/ggml-webgpu/wgsl-shaders/unary_op.wgsl

@@ -16,7 +16,8 @@
     "HARDSWISH_FUNC": "{{MUTATE}}[dst_i] = src[src_i] * min(1.0, max(0.0, (src[src_i] + 3.0) / 6.0));",
     "GELU_FUNC": "{{MUTATE}}[dst_i] = 0.5 * src[src_i] * (1.0 + tanh(clamp(sqrt(2.0 / 3.14159265) * (src[src_i] + 0.044715 * pow(src[src_i], 3.0)), -9.010913, 9.010913))); // Regarding tanh() domain restrictions in wgsl https://github.com/gpuweb/gpuweb/issues/4458",
     "GELU_QUICK_FUNC": "{{MUTATE}}[dst_i] = src[src_i] * 0.5 * (1.0 + tanh(clamp(0.79788456 * (src[src_i] + 0.044715 * src[src_i] * src[src_i] * src[src_i]), -9.010913, 9.010913))); // Regarding tanh() domain restrictions in wgsl https://github.com/gpuweb/gpuweb/issues/4458",
-    "GELU_ERF_FUNC": "{{MUTATE}}[dst_i] = 0.5 * src[src_i] * (1.0 + tanh(clamp(0.79788456 * (src[src_i] + 0.044715 * src[src_i] * src[src_i] * src[src_i]), -9.010913, 9.010913))); // Regarding tanh() domain restrictions in wgsl https://github.com/gpuweb/gpuweb/issues/4458"
+    "GELU_ERF_FUNC": "{{MUTATE}}[dst_i] = 0.5 * src[src_i] * (1.0 + tanh(clamp(0.79788456 * (src[src_i] + 0.044715 * src[src_i] * src[src_i] * src[src_i]), -9.010913, 9.010913))); // Regarding tanh() domain restrictions in wgsl https://github.com/gpuweb/gpuweb/issues/4458",
+    "CEIL_FUNC": "{{MUTATE}}[dst_i] = ceil(src[src_i]);"
 }
 
 #end(REPL_TEMPLATES)
@@ -357,6 +358,27 @@
         "SHADER_NAME": "gelu_erf_inplace_f16",
         "REPLS": { "TYPE": "f16", "FUNC": "GELU_ERF_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
         "DECLS": ["INPLACE"]
+    },
+
+    {
+        "SHADER_NAME": "ceil_f32",
+        "REPLS": { "TYPE": "f32", "FUNC": "CEIL_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
+        "DECLS": ["NOT_INPLACE"]
+    },
+    {
+        "SHADER_NAME": "ceil_f16",
+        "REPLS": { "TYPE": "f16", "FUNC": "CEIL_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
+        "DECLS": ["NOT_INPLACE"]
+    },
+    {
+        "SHADER_NAME": "ceil_inplace_f32",
+        "REPLS": { "TYPE": "f32", "FUNC": "CEIL_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
+        "DECLS": ["INPLACE"]
+    },
+    {
+        "SHADER_NAME": "ceil_inplace_f16",
+        "REPLS": { "TYPE": "f16", "FUNC": "CEIL_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
+        "DECLS": ["INPLACE"]
     }
 ]