whispercpp 1.3.6 → 1.3.7

data/ext/sources/ggml/src/ggml-hexagon/htp/cpy-ops.c

@@ -11,7 +11,7 @@
 #define GGML_COMMON_DECL_C
 #include "ggml-common.h"
 #include "htp-ctx.h"
-#include "htp-msg.h"
+#include "htp-ops.h"
 #include "htp-ops.h"
 #include "hvx-utils.h"
 
@@ -28,135 +28,170 @@ struct htp_copy_context {
     uint32_t          dst_blocks_per_row;
 
     uint32_t          src0_nrows_per_thread;
-
-    void (*copy)(struct htp_copy_context * ct, struct htp_ops_context * octx, int nth, int ith);
 };
 
-#define cpy_preamble                       \
-    struct htp_tensor *src0 = &octx->src0; \
-    struct htp_tensor *dst  = &octx->dst;  \
-                                           \
-    const uint32_t ne00 = src0->ne[0];     \
-    const uint32_t ne01 = src0->ne[1];     \
-    const uint32_t ne02 = src0->ne[2];     \
-    const uint32_t ne03 = src0->ne[3];     \
-                                           \
-    const uint32_t nb00 = src0->nb[0];     \
-    const uint32_t nb01 = src0->nb[1];     \
-    const uint32_t nb02 = src0->nb[2];     \
-    const uint32_t nb03 = src0->nb[3];     \
-                                           \
-    const uint32_t  ne0 = dst->ne[0];      \
-    const uint32_t  ne1 = dst->ne[1];      \
-    const uint32_t  ne2 = dst->ne[2];      \
-    const uint32_t  ne3 = dst->ne[3];      \
-                                           \
-    const uint32_t  nb0 = dst->nb[0];      \
-    const uint32_t  nb1 = dst->nb[1];      \
-    const uint32_t  nb2 = dst->nb[2];      \
-    const uint32_t  nb3 = dst->nb[3];      \
-                                           \
+#define cpy_preamble                              \
+    const struct htp_tensor *src0 = octx->src[0]; \
+    const struct htp_tensor *dst  = octx->dst;    \
+                                                  \
+    const uint32_t ne00 = src0->ne[0];            \
+    const uint32_t ne01 = src0->ne[1];            \
+    const uint32_t ne02 = src0->ne[2];            \
+    const uint32_t ne03 = src0->ne[3];            \
+                                                  \
+    const uint32_t nb00 = src0->nb[0];            \
+    const uint32_t nb01 = src0->nb[1];            \
+    const uint32_t nb02 = src0->nb[2];            \
+    const uint32_t nb03 = src0->nb[3];            \
+                                                  \
+    const uint32_t  ne0 = dst->ne[0];             \
+    const uint32_t  ne1 = dst->ne[1];             \
+    const uint32_t  ne2 = dst->ne[2];             \
+    const uint32_t  ne3 = dst->ne[3];             \
+                                                  \
+    const uint32_t  nb0 = dst->nb[0];             \
+    const uint32_t  nb1 = dst->nb[1];             \
+    const uint32_t  nb2 = dst->nb[2];             \
+    const uint32_t  nb3 = dst->nb[3];             \
+                                                  \
     const uint32_t   nr = ne01;
 
-static void cpy_thread_sametype_sameshape(struct htp_copy_context * ct, struct htp_ops_context * octx, const int nth, const int ith) {
-    cpy_preamble;
-
-    // parallelize by src0 rows
-    const uint32_t dr  = ct->src0_nrows_per_thread;
-    const uint32_t ir0 = dr * ith;
-    const uint32_t ir1 = (ir0 + dr) < nr ? (ir0 + dr) : nr;
-
-    // copy by rows
-    for (uint32_t i03 = 0; i03 < ne03; i03++) {
-        for (uint32_t i02 = 0; i02 < ne02; i02++) {
-            #pragma unroll(2)
-            for (uint32_t i01 = ir0; i01 < ir1; i01++) {
-                uint8_t* dst_ptr  = (uint8_t*) dst->data  + i01*nb1  + i02*nb2  + i03*nb3;
-                uint8_t* src0_ptr = (uint8_t*) src0->data + i01*nb01 + i02*nb02 + i03*nb03;
-                hex_l2fetch(src0_ptr, ne00 * ct->src0_type_size, nb01, 2);
-                hvx_copy_uu(dst_ptr, src0_ptr, ne00, ct->src0_type_size);
-            }
-        }
-    }
+#define DEFINE_CPY_SAMESHAPE(NAME, ELEM_TYPE, ELEM_SIZE)                                                       \
+static void cpy_thread_##NAME##_sameshape(unsigned int nth, unsigned int ith, void * data) {                   \
+    struct htp_copy_context * ct = (struct htp_copy_context *) data;                                           \
+    struct htp_ops_context * octx = ct->octx;                                                                  \
+    cpy_preamble;                                                                                              \
+    const uint32_t dr  = ct->src0_nrows_per_thread;                                                            \
+    const uint32_t ir0 = dr * ith;                                                                             \
+    const uint32_t ir1 = (ir0 + dr) < nr ? (ir0 + dr) : nr;                                                    \
+    if (ir0 >= nr) return;                                                                                     \
+    for (uint32_t i03 = 0; i03 < ne03; i03++) {                                                                \
+        for (uint32_t i02 = 0; i02 < ne02; i02++) {                                                            \
+            _Pragma("unroll(4)")                                                                               \
+            for (uint32_t i01 = ir0; i01 < ir1; i01++) {                                                       \
+                uint8_t* dst_ptr  = (uint8_t*) dst->data  + i01*nb1  + i02*nb2  + i03*nb3;                     \
+                uint8_t* src0_ptr = (uint8_t*) src0->data + i01*nb01 + i02*nb02 + i03*nb03;                    \
+                hex_l2fetch(src0_ptr, ne00 * ELEM_SIZE, nb01, 2);                                              \
+                hvx_copy_uu(dst_ptr, src0_ptr, ne00, ELEM_SIZE);                                               \
+            }                                                                                                  \
+        }                                                                                                      \
+    }                                                                                                          \
 }
 
-static void cpy_thread_sametype_reshape(struct htp_copy_context * ct, struct htp_ops_context * octx, int nth, int ith) {
-    cpy_preamble;
-
-    // parallelize by src0 rows
-    const uint32_t dr  = ct->src0_nrows_per_thread;
-    const uint32_t ir0 = dr * ith;
-    const uint32_t ir1 = (ir0 + dr) < nr ? (ir0 + dr) : nr;
-
-    // dst counters
-    int64_t k10 = 0;
-    int64_t i11 = 0;
-    int64_t i12 = 0;
-    int64_t i13 = 0;
-
-    // number of blocks in a row
-    const int64_t nk00 = ct->src0_blocks_per_row;
-    const int64_t nk0  = ct->dst_blocks_per_row;
-
-    for (int64_t i03 = 0; i03 < ne03; i03++) {
-        for (int64_t i02 = 0; i02 < ne02; i02++) {
-            k10 += nk00 * ir0;
-            while (k10 >= nk0) {
-                k10 -= nk0;
-                if (++i11 == ne1) {
-                    i11 = 0;
-                    if (++i12 == ne2) {
-                        i12 = 0;
-                        if (++i13 == ne3) {
-                            i13 = 0;
-                        }
-                    }
-                }
-            }
-            for (int64_t i01 = ir0; i01 < ir1; i01++) {
-                for (int64_t k00 = 0; k00 < nk00; k00++) {
-                    const char * src0_ptr = ((char *) src0->data + k00*nb00 + i01*nb01 + i02*nb02 + i03*nb03);
-                          char * dst_ptr  = ((char *)  dst->data + k10*nb0  + i11*nb1  + i12*nb2  + i13*nb3);
-                    memcpy(dst_ptr, src0_ptr, ct->dst_type_size);
-
-                    if (++k10 == nk0) {
-                        k10 = 0;
-                        if (++i11 == ne1) {
-                            i11 = 0;
-                            if (++i12 == ne2) {
-                                i12 = 0;
-                                if (++i13 == ne3) {
-                                    i13 = 0;
-                                }
-                            }
-                        }
-                    }
-                }
-            }
-            k10 += nk00 * (ne01 - ir1);
-            while (k10 >= nk0) {
-                k10 -= nk0;
-                if (++i11 == ne1) {
-                    i11 = 0;
-                    if (++i12 == ne2) {
-                        i12 = 0;
-                        if (++i13 == ne3) {
-                            i13 = 0;
-                        }
-                    }
-                }
-            }
-        }
-    }
+DEFINE_CPY_SAMESHAPE(f32, float, 4)
+DEFINE_CPY_SAMESHAPE(f16, __fp16, 2)
+
+#define DEFINE_CPY_RESHAPE(NAME, ELEM_TYPE, ELEM_SIZE)                                                         \
+static void cpy_thread_##NAME##_reshape(unsigned int nth, unsigned int ith, void * data) {                     \
+    struct htp_copy_context * ct = (struct htp_copy_context *) data;                                           \
+    struct htp_ops_context * octx = ct->octx;                                                                  \
+    cpy_preamble;                                                                                              \
+    const uint32_t dr  = ct->src0_nrows_per_thread;                                                            \
+    const uint32_t ir0 = dr * ith;                                                                             \
+    const uint32_t ir1 = (ir0 + dr) < nr ? (ir0 + dr) : nr;                                                    \
+    if (ir0 >= nr) return;                                                                                     \
+    const bool src0_contig = (nb00 == ELEM_SIZE)   &&                                                          \
+                             (nb01 == ne00 * nb00) &&                                                          \
+                             (nb02 == ne01 * nb01) &&                                                          \
+                             (nb03 == ne02 * nb02);                                                            \
+    const bool dst_contig  = (nb0  == ELEM_SIZE)   &&                                                          \
+                             (nb1  == ne0  * nb0)  &&                                                          \
+                             (nb2  == ne1  * nb1)  &&                                                          \
+                             (nb3  == ne2  * nb2);                                                             \
+    if (src0_contig && dst_contig) {                                                                           \
+        for (int64_t i03 = 0; i03 < ne03; i03++) {                                                             \
+            for (int64_t i02 = 0; i02 < ne02; i02++) {                                                         \
+                uint8_t * src_ptr = (uint8_t *) src0->data + i03*nb03 + i02*nb02 + ir0*nb01;                   \
+                uint32_t  flat    = ((i03*ne02 + i02)*ne01 + ir0) * ne00;                                      \
+                uint8_t * dst_ptr = (uint8_t *) dst->data  + flat * ELEM_SIZE;                                 \
+                hvx_copy_uu(dst_ptr, src_ptr, (ir1 - ir0) * ne00, ELEM_SIZE);                                  \
+            }                                                                                                  \
+        }                                                                                                      \
+        return;                                                                                                \
+    }                                                                                                          \
+    const bool reshape_flat_fast = (ne03 == 1 && ne2 == 1 && ne3 == 1) &&                                      \
+                                   (ne0 == ne00 * ne01) && (ne1 == ne02) &&                                    \
+                                   (nb00 == ELEM_SIZE) && (nb0 == ELEM_SIZE);                                  \
+    if (reshape_flat_fast) {                                                                                   \
+        for (uint32_t i02 = 0; i02 < ne02; i02++) {                                                            \
+            for (uint32_t i01 = ir0; i01 < ir1; i01++) {                                                       \
+                uint8_t * src0_ptr = (uint8_t *) src0->data + i01 * nb01 + i02 * nb02;                         \
+                uint8_t * dst_ptr  = (uint8_t *) dst->data  + i01 * ne00 * ELEM_SIZE + i02 * nb1;              \
+                hvx_copy_uu(dst_ptr, src0_ptr, ne00, ELEM_SIZE);                                               \
+            }                                                                                                  \
+        }                                                                                                      \
+        return;                                                                                                \
+    }                                                                                                          \
+    int64_t k10 = 0;                                                                                           \
+    int64_t i11 = 0;                                                                                           \
+    int64_t i12 = 0;                                                                                           \
+    int64_t i13 = 0;                                                                                           \
+    const int64_t nk00 = ct->src0_blocks_per_row;                                                              \
+    const int64_t nk0  = ct->dst_blocks_per_row;                                                               \
+    for (int64_t i03 = 0; i03 < ne03; i03++) {                                                                 \
+        for (int64_t i02 = 0; i02 < ne02; i02++) {                                                             \
+            k10 += nk00 * ir0;                                                                                 \
+            while (k10 >= nk0) {                                                                               \
+                k10 -= nk0;                                                                                    \
+                if (++i11 == ne1) {                                                                            \
+                    i11 = 0;                                                                                   \
+                    if (++i12 == ne2) {                                                                        \
+                        i12 = 0;                                                                               \
+                        if (++i13 == ne3) {                                                                    \
+                            i13 = 0;                                                                           \
+                        }                                                                                      \
+                    }                                                                                          \
+                }                                                                                              \
+            }                                                                                                  \
+            for (int64_t i01 = ir0; i01 < ir1; i01++) {                                                        \
+                for (int64_t k00 = 0; k00 < nk00; k00++) {                                                     \
+                    const char * src0_ptr = ((char *) src0->data + k00*nb00 + i01*nb01 + i02*nb02 + i03*nb03); \
+                          char * dst_ptr  = ((char *)  dst->data + k10*nb0  + i11*nb1  + i12*nb2  + i13*nb3);  \
+                    memcpy(dst_ptr, src0_ptr, ELEM_SIZE);                                                      \
+                    if (++k10 == nk0) {                                                                        \
+                        k10 = 0;                                                                               \
+                        if (++i11 == ne1) {                                                                    \
+                            i11 = 0;                                                                           \
+                            if (++i12 == ne2) {                                                                \
+                                i12 = 0;                                                                       \
+                                if (++i13 == ne3) {                                                            \
+                                    i13 = 0;                                                                   \
+                                }                                                                              \
+                            }                                                                                  \
+                        }                                                                                      \
+                    }                                                                                          \
+                }                                                                                              \
+            }                                                                                                  \
+            k10 += nk00 * (ne01 - ir1);                                                                        \
+            while (k10 >= nk0) {                                                                               \
+                k10 -= nk0;                                                                                    \
+                if (++i11 == ne1) {                                                                            \
+                    i11 = 0;                                                                                   \
+                    if (++i12 == ne2) {                                                                        \
+                        i12 = 0;                                                                               \
+                        if (++i13 == ne3) {                                                                    \
+                            i13 = 0;                                                                           \
+                        }                                                                                      \
+                    }                                                                                          \
+                }                                                                                              \
+            }                                                                                                  \
+        }                                                                                                      \
+    }                                                                                                          \
 }
 
-static void cpy_thread_f16_f32_sameshape(struct htp_copy_context * ct, struct htp_ops_context * octx, const int nth, const int ith) {
+DEFINE_CPY_RESHAPE(f32, float, 4)
+DEFINE_CPY_RESHAPE(f16, __fp16, 2)
+
+static void cpy_thread_f16_f32_sameshape(unsigned int nth, unsigned int ith, void * data) {
+    struct htp_copy_context * ct = (struct htp_copy_context *) data;
+    struct htp_ops_context * octx = ct->octx;
     cpy_preamble;
 
     // parallelize by src0 rows
     const uint32_t dr  = ct->src0_nrows_per_thread;
     const uint32_t ir0 = dr * ith;
     const uint32_t ir1 = (ir0 + dr) < nr ? (ir0 + dr) : nr;
+    if (ir0 >= nr) return;
 
     // copy by rows
     for (uint32_t i03 = 0; i03 < ne03; i03++) {
@@ -172,13 +207,16 @@ static void cpy_thread_f16_f32_sameshape(struct htp_copy_context * ct, struct ht
     }
 }
 
-static void cpy_thread_f32_f16_sameshape(struct htp_copy_context * ct, struct htp_ops_context * octx, const int nth, const int ith) {
+static void cpy_thread_f32_f16_sameshape(unsigned int nth, unsigned int ith, void * data) {
+    struct htp_copy_context * ct = (struct htp_copy_context *) data;
+    struct htp_ops_context * octx = ct->octx;
     cpy_preamble;
 
     // parallelize by src0 rows
     const uint32_t dr  = ct->src0_nrows_per_thread;
     const uint32_t ir0 = dr * ith;
     const uint32_t ir1 = (ir0 + dr) < nr ? (ir0 + dr) : nr;
+    if (ir0 >= nr) return;
 
     // copy by rows
     for (uint32_t i03 = 0; i03 < ne03; i03++) {
@@ -194,11 +232,6 @@ static void cpy_thread_f32_f16_sameshape(struct htp_copy_context * ct, struct ht
     }
 }
 
-static void cpy_work_func(unsigned int n, unsigned int i, void *data) {
-    struct htp_copy_context *ct = (struct htp_copy_context *) data;
-    ct->copy(ct, ct->octx, n, i);
-}
-
 int op_cpy(struct htp_ops_context * octx) {
     cpy_preamble;
 
@@ -231,22 +264,32 @@ int op_cpy(struct htp_ops_context * octx) {
 
     ct.src0_nrows_per_thread = (nr + n_threads - 1) / n_threads;
 
+    worker_callback_t copy_fun;
+
     if (sametype && sameshape) {
-        ct.copy = cpy_thread_sametype_sameshape;
+        if (src0->type == HTP_TYPE_F32) {
+            copy_fun = cpy_thread_f32_sameshape;
+        } else {
+            copy_fun = cpy_thread_f16_sameshape;
+        }
     } else if (sameshape) {
         /**/ if (dst->type == HTP_TYPE_F16 && src0->type == HTP_TYPE_F32)
-            ct.copy = cpy_thread_f16_f32_sameshape;
+            copy_fun = cpy_thread_f16_f32_sameshape;
         else if (dst->type == HTP_TYPE_F32 && src0->type == HTP_TYPE_F16)
-            ct.copy = cpy_thread_f32_f16_sameshape;
+            copy_fun = cpy_thread_f32_f16_sameshape;
         else
             return HTP_STATUS_NO_SUPPORT;
     } else if (sametype) {
-        ct.copy = cpy_thread_sametype_reshape;
+        if (src0->type == HTP_TYPE_F32) {
+            copy_fun = cpy_thread_f32_reshape;
+        } else {
+            copy_fun = cpy_thread_f16_reshape;
+        }
     } else {
         return HTP_STATUS_NO_SUPPORT;
     }
 
-    worker_pool_run_func(octx->ctx->worker_pool, cpy_work_func, &ct, n_threads);
+    worker_pool_run_func(octx->ctx->worker_pool, copy_fun, &ct, n_threads);
 
     return HTP_STATUS_OK;
 }

data/ext/sources/ggml/src/ggml-hexagon/htp/cumsum-ops.c

@@ -0,0 +1,270 @@
+#pragma clang diagnostic ignored "-Wunused-variable"
+#pragma clang diagnostic ignored "-Wunused-function"
+#pragma clang diagnostic ignored "-Wunused-but-set-variable"
+
+#include <HAP_farf.h>
+#include <HAP_perf.h>
+
+#define GGML_COMMON_DECL_C
+#include "ggml-common.h"
+#include "htp-ctx.h"
+#include "htp-ops.h"
+#include "hvx-types.h"
+#include "hvx-utils.h"
+#include "hex-dma.h"
+
+#define htp_cumsum_tensors_preamble                         \
+    const struct htp_tensor * restrict src0 = octx->src[0]; \
+    const struct htp_tensor * restrict dst  = octx->dst;    \
+                                                     \
+    const uint32_t ne00 = src0->ne[0];               \
+    const uint32_t ne01 = src0->ne[1];               \
+    const uint32_t ne02 = src0->ne[2];               \
+    const uint32_t ne03 = src0->ne[3];               \
+                                                     \
+    const uint32_t ne0 = dst->ne[0];                 \
+    const uint32_t ne1 = dst->ne[1];                 \
+    const uint32_t ne2 = dst->ne[2];                 \
+    const uint32_t ne3 = dst->ne[3];                 \
+                                                     \
+    const uint32_t nb00 = src0->nb[0];               \
+    const uint32_t nb01 = src0->nb[1];               \
+    const uint32_t nb02 = src0->nb[2];               \
+    const uint32_t nb03 = src0->nb[3];               \
+                                                     \
+    const uint32_t nb0 = dst->nb[0];                 \
+    const uint32_t nb1 = dst->nb[1];                 \
+    const uint32_t nb2 = dst->nb[2];                 \
+    const uint32_t nb3 = dst->nb[3];
+
+struct htp_cumsum_context {
+    struct htp_ops_context * octx;
+    size_t          src_row_size;
+    size_t          dst_row_size;
+    size_t          src_row_size_aligned;
+    size_t          dst_row_size_aligned;
+    uint32_t        rows_per_thread;
+    uint32_t        total_rows;
+};
+
+#define htp_cumsum_preamble                                                \
+    struct htp_cumsum_context * cctx = (struct htp_cumsum_context *) data; \
+    struct htp_ops_context *    octx = cctx->octx;                         \
+    htp_cumsum_tensors_preamble;                                           \
+    dma_queue * dma_queue = octx->ctx->dma[ith];
+
+// ---------------------------------------------------------------------------
+// HVX prefix scan helpers
+// ---------------------------------------------------------------------------
+
+#if __HVX_ARCH__ > 75
+static inline HVX_Vector hvx_cumsum_vadd(HVX_Vector a, HVX_Vector b) {
+    return Q6_Vsf_vadd_VsfVsf(a, b);
+}
+#else
+static inline HVX_Vector hvx_cumsum_vadd(HVX_Vector a, HVX_Vector b) {
+    return Q6_Vsf_equals_Vqf32(Q6_Vqf32_vadd_VsfVsf(a, b));
+}
+#endif  // __HVX_ARCH__ > 75
+
+static inline HVX_Vector hvx_prefix_scan_f32(HVX_Vector v, HVX_Vector carry_in) {
+    const HVX_Vector zero = Q6_V_vsplat_R(0);
+
+    v = hvx_cumsum_vadd(v, Q6_V_vlalign_VVR(v, zero,  4));
+    v = hvx_cumsum_vadd(v, Q6_V_vlalign_VVR(v, zero,  8));
+    v = hvx_cumsum_vadd(v, Q6_V_vlalign_VVR(v, zero, 16));
+    v = hvx_cumsum_vadd(v, Q6_V_vlalign_VVR(v, zero, 32));
+    v = hvx_cumsum_vadd(v, Q6_V_vlalign_VVR(v, zero, 64));
+    v = hvx_cumsum_vadd(v, carry_in);
+
+    return v;
+}
+
+static inline HVX_Vector hvx_splat_last_f32(HVX_Vector v) {
+    return hvx_vec_repl4(Q6_V_vror_VR(v, 124));
+}
+
+static inline void hvx_cumsum_row_f32(const float * restrict src, float * restrict dst, uint32_t n) {
+    const uint32_t nvec = n / VLEN_FP32;
+    const uint32_t nloe = n % VLEN_FP32;
+
+    HVX_Vector carry = Q6_V_vsplat_R(0);
+
+    for (uint32_t i = 0; i < nvec; i++) {
+        HVX_Vector v = *((const HVX_UVector *) (src + i * VLEN_FP32));
+        v = hvx_prefix_scan_f32(v, carry);
+        hvx_vec_store_u(dst + i * VLEN_FP32, VLEN, v);
+        carry = hvx_splat_last_f32(v);
+    }
+
+    if (nloe) {
+        float acc = hvx_vec_get_f32(carry);
+        const float * src_tail = src + nvec * VLEN_FP32;
+        float       * dst_tail = dst + nvec * VLEN_FP32;
+        for (uint32_t i = 0; i < nloe; i++) {
+            acc        += src_tail[i];
+            dst_tail[i] = acc;
+        }
+    }
+}
+
+// ---------------------------------------------------------------------------
+// Per thread worker: Double-buffered DMA
+// ---------------------------------------------------------------------------
+
+static void cumsum_thread_f32_dma(unsigned int nth, unsigned int ith, void * data) {
+    htp_cumsum_preamble;
+
+    uint64_t t1, t2;
+    t1 = HAP_perf_get_qtimer_count();
+
+    const uint32_t ir0 = cctx->rows_per_thread * ith;
+    const uint32_t ir1 = MIN(ir0 + cctx->rows_per_thread, cctx->total_rows);
+
+    if (ir0 >= ir1) {
+        return;
+    }
+
+    const size_t src_row_size         = cctx->src_row_size;
+    const size_t dst_row_size         = cctx->dst_row_size;
+    const size_t src_row_size_aligned = cctx->src_row_size_aligned;
+    const size_t dst_row_size_aligned = cctx->dst_row_size_aligned;
+
+    const uint8_t * src_data = (const uint8_t *) src0->data;
+    uint8_t *       dst_data = (uint8_t *) dst->data;
+
+    uint8_t * src_spad = octx->src0_spad.data + (ith * src_row_size_aligned * 2);
+    uint8_t * dst_spad = octx->dst_spad.data  + (ith * dst_row_size_aligned * 2);
+
+    for (uint32_t ir = ir0, spad_idx = 0; ir < ir1 && spad_idx < 2; ir++, spad_idx++) {
+        // Dummy dst writeback to establish queue ordering
+        dma_queue_push_vtcm_to_ddr(dma_queue,
+                                   dma_make_ptr(dst_data, dst_spad + (spad_idx * dst_row_size_aligned)),
+                                   dst_row_size, dst_row_size_aligned, 0);
+
+        dma_queue_push_ddr_to_vtcm(dma_queue,
+                                   dma_make_ptr(src_spad + (spad_idx * src_row_size_aligned),
+                                                src_data + (ir * src_row_size)),
+                                   src_row_size_aligned, src_row_size, 1);
+    }
+
+    for (uint32_t ir = ir0; ir < ir1; ir++) {
+        float * dst_spad_row = (float *) dma_queue_pop(dma_queue).src;
+        float * src_spad_row = (float *) dma_queue_pop(dma_queue).dst;
+
+        hvx_cumsum_row_f32(src_spad_row, dst_spad_row, ne00);
+
+        dma_queue_push_vtcm_to_ddr(dma_queue,
+                                   dma_make_ptr(dst_data + (ir * dst_row_size), (uint8_t *) dst_spad_row),
+                                   dst_row_size, dst_row_size_aligned, 1);
+
+        const uint32_t next_row = ir + 2;
+        if (next_row < ir1) {
+            dma_queue_push_ddr_to_vtcm(dma_queue,
+                                       dma_make_ptr((uint8_t *) src_spad_row, src_data + (next_row * src_row_size)),
+                                       src_row_size_aligned, src_row_size, 1);
+        }
+    }
+
+    dma_queue_flush(dma_queue);
+    t2 = HAP_perf_get_qtimer_count();
+
+    FARF(HIGH, "cumsum-f32-dma %d/%d: %ux%ux%ux%u (%u:%u) -> %ux%ux%ux%u usec %u\n",
+         ith, nth, src0->ne[0], src0->ne[1], src0->ne[2], src0->ne[3], ir0, ir1,
+         dst->ne[0], dst->ne[1], dst->ne[2], dst->ne[3],
+         (unsigned) HAP_perf_qtimer_count_to_us(t2 - t1));
+}
+
+// ---------------------------------------------------------------------------
+// Per thread worker: Direct HVX (no DMA)
+// ---------------------------------------------------------------------------
+
+static void cumsum_thread_f32(unsigned int nth, unsigned int ith, void * data) {
+    htp_cumsum_preamble;
+
+    uint64_t t1, t2;
+    t1 = HAP_perf_get_qtimer_count();
+
+    const uint8_t * src_data = (const uint8_t *) src0->data;
+    uint8_t *       dst_data = (uint8_t *) dst->data;
+
+    const uint32_t ir0 = cctx->rows_per_thread * ith;
+    const uint32_t ir1 = MIN(ir0 + cctx->rows_per_thread, cctx->total_rows);
+
+    for (uint32_t ir = ir0; ir < ir1; ir++) {
+        const float * restrict src_row = (const float *) (src_data + ir * cctx->src_row_size);
+        float       * restrict dst_row = (float *)       (dst_data + ir * cctx->dst_row_size);
+        hvx_cumsum_row_f32(src_row, dst_row, ne00);
+    }
+
+    t2 = HAP_perf_get_qtimer_count();
+
+    FARF(HIGH, "cumsum-f32 %d/%d: %ux%ux%ux%u (%u:%u) -> %ux%ux%ux%u usec %u\n",
+         ith, nth, src0->ne[0], src0->ne[1], src0->ne[2], src0->ne[3], ir0, ir1,
+         dst->ne[0], dst->ne[1], dst->ne[2], dst->ne[3],
+         (unsigned) HAP_perf_qtimer_count_to_us(t2 - t1));
+}
+
+int op_cumsum_f32(struct htp_ops_context * octx) {
+    const struct htp_tensor * src0 = octx->src[0];
+    const struct htp_tensor * dst  = octx->dst;
+
+    if (octx->flags & HTP_OPFLAGS_SKIP_COMPUTE) {
+        return HTP_STATUS_OK;
+    }
+
+    const uint32_t total_rows = src0->ne[1] * src0->ne[2] * src0->ne[3];
+    const uint32_t n_threads  = MIN(octx->n_threads, total_rows);
+
+    const size_t src_row_size         = src0->nb[1];
+    const size_t dst_row_size         = dst->nb[1];
+    const size_t src_row_size_aligned = hex_round_up(src_row_size, VLEN);
+    const size_t dst_row_size_aligned = hex_round_up(dst_row_size, VLEN);
+
+    // 2 ping-pong buffers per thread for src and dst
+    const size_t spad_per_thread = 2 * (src_row_size_aligned + dst_row_size_aligned);
+
+    octx->src0_spad.size_per_thread = src_row_size_aligned * 2;
+    octx->dst_spad.size_per_thread  = dst_row_size_aligned * 2;
+
+    octx->src0_spad.size  = n_threads * octx->src0_spad.size_per_thread;
+    octx->dst_spad.size   = n_threads * octx->dst_spad.size_per_thread;
+
+    octx->src0_spad.data  = octx->ctx->vtcm_base;                        octx->src0_spad.src = NULL;
+    octx->dst_spad.data   = octx->src0_spad.data + octx->src0_spad.size; octx->dst_spad.src  = NULL;
+
+    struct htp_cumsum_context cctx = {
+        .octx                 = octx,
+        .src_row_size         = src_row_size,
+        .dst_row_size         = dst_row_size,
+        .src_row_size_aligned = src_row_size_aligned,
+        .dst_row_size_aligned = dst_row_size_aligned,
+        .rows_per_thread      = (total_rows + n_threads - 1) / n_threads,
+        .total_rows           = total_rows,
+    };
+
+    if (octx->ctx->vtcm_size < spad_per_thread * n_threads) {
+        worker_pool_run_func(octx->ctx->worker_pool, cumsum_thread_f32, &cctx, n_threads);
+    } else {
+        worker_pool_run_func(octx->ctx->worker_pool, cumsum_thread_f32_dma, &cctx, n_threads);
+    }
+
+    return HTP_STATUS_OK;
+}
+
+int op_cumsum(struct htp_ops_context * octx) {
+    const struct htp_tensor * dst = octx->dst;
+
+    int err = HTP_STATUS_OK;
+
+    switch (dst->type) {
+        case HTP_TYPE_F32:
+            err = op_cumsum_f32(octx);
+            break;
+        default:
+            err = HTP_STATUS_NO_SUPPORT;
+            break;
+    }
+
+    return err;
+}