Hexagon add support for f16/f32 flash attention, scale, set-rows and improve f16/32 matmul (#18611)

* hexagon: improve fp16 matmul and add fp32/fp16 flash-attention

* hexagon: add support for set-rows fp32 -> fp16 with i32/i64 row-idx

* hexagon: add support for SCALE fp32

* hexagon: replace scalar fp32 -> fp16 copy with HVX

* hexagon: optimize flash_atten_ext with aligned VTCM buffers and DMA

- Implements double-buffered DMA prefetching for K, V, and Mask tensors.
- Ensures K and V rows in VTCM are padded to 128 bytes to support aligned HVX operations.
- Correctly synchronizes DMA transfers to prevent race conditions.
- Uses `FLASH_ATTN_BLOCK_SIZE` of 128 for efficient chunking.

* hexagon: use aligned mad_f16

* hexagon: flash_atten more aligned ops

* hexagon: optimize scale_f32 hvx helpers

* hexagon: unroll fa loops

* hexagon: remove unused set-rows log

* hexagon: flash_attn_ext add support for DMAing Q

- Update `op_flash_attn_ext` to include Q row size in scratchpad allocation.
- Pad Q row size to 128 bytes for alignment.
- Implement DMA transfer for Q tensor in `flash_attn_ext_f16_thread`.
- Update dot product computations to use VTCM-buffered Q data.

* hexagon: fix handling of NANs hvx dotproducts

* hexagon: cleanup spad allocation in flash-atten

* hexagon: improve fp16/fp32 matmul

- Introduced `vec_dot_f16_f16` and `vec_dot_f16_f16_rx2` kernels using efficient HVX dot product intrinsics.
- Added `quantize_fp32_f16` to copy/convert weights from DDR to VTCM
- Updated `op_matmul` to use the optimized path when VTCM capacity allows and broadcasting requirements are compatible.
- Implemented fallback logic to the original implementation for complex broadcasting scenarios.

* hexagon: fix HVX_ARCH check

* hexagon: matmul cleanup and fp16 fixes

Use aligned vec_dot_f16 for 2d matmuls and unaligned version for 4d.

* hexagon: fix fp16 x fp16 matmuls and some minor refactoring

* hexagon: add support for GET_ROWS f32 -> f32

Also optimize SET_ROWS threading a bit when we have just a few rows to process.

* hexagon: optimize set-rows threading

* hexagon: update adb/run-bench.sh to properly support experimental and verbose options

* hexagon: flash_atten use aligned vectors for dot products

ggml/src/ggml-hexagon/htp/unary-ops.c

@@ -83,6 +83,31 @@ static void hvx_fast_rms_norm_f32(const uint8_t * restrict src,
     }
 }
 
+static void scale_htp_f32(const float * restrict src,
+                          float * restrict dst,
+                          uint8_t * restrict spad,
+                          const uint32_t num_rows,
+                          const uint32_t row_elems,
+                          const size_t   row_size,
+                          int32_t *      op_params,
+                          int            opt_path) {
+    float scale = 0.f;
+    float bias  = 0.f;
+    memcpy(&scale, &op_params[0], sizeof(float));
+    memcpy(&bias,  &op_params[1], sizeof(float));
+
+    for (uint32_t ir = 0; ir < num_rows; ir++) {
+        const float * restrict src_local = src + (ir * row_elems);
+        float * restrict dst_local       = dst + (ir * row_elems);
+
+        if (ir + 1 < num_rows) {
+            htp_l2fetch(src_local + row_elems, 1, row_size, row_size);
+        }
+
+        hvx_scale_offset_f32((uint8_t *) dst_local, (const uint8_t *) src_local, row_elems, scale, bias);
+    }
+}
+
 static void rms_norm_htp_f32(const float * restrict src,
                              float * restrict dst,
                              uint8_t * restrict spad,
@@ -110,7 +135,7 @@ static void rms_norm_htp_f32(const float * restrict src,
             const float mean  = sum / row_elems;
             const float scale = 1.0f / sqrtf(mean + epsilon);
 
-            hvx_scale_f32((const uint8_t *) src_local, (uint8_t *) dst_local, row_elems, scale);
+            hvx_scale_f32((uint8_t *) dst_local, (const uint8_t *) src_local, row_elems, scale);
         }
     }
 }
@@ -162,6 +187,9 @@ static void unary_job_f32_per_thread(const struct htp_tensor * src,
         case HTP_OP_RMS_NORM:
             rms_norm_htp_f32(src_th, dst_th, spad_th, src0_end_row - src0_start_row, ne0, nb1, op_params, opt_path);
             break;
+        case HTP_OP_SCALE:
+            scale_htp_f32(src_th, dst_th, spad_th, src0_end_row - src0_start_row, ne0, nb1, op_params, opt_path);
+            break;
 
         default:
             break;
@@ -195,6 +223,10 @@ static int execute_op_unary_f32(struct htp_ops_context * octx) {
             unary_op_func = unary_job_dispatcher_f32;
             op_type       = "rmsnorm-f32";
             break;
+        case HTP_OP_SCALE:
+            unary_op_func = unary_job_dispatcher_f32;
+            op_type       = "scale-f32";
+            break;
 
         default:
             FARF(ERROR, "Unsupported unary Op %u\n", octx->op);