ggml : add GGML_TENSOR_LOCALS helper macros (#309)

* [WIP] ref #292

* Further code reduction

* ggml : minor style fixes

* ggml : hide op locals in source file

---------

Co-authored-by: Georgi Gerganov <redacted>

diff --git a/include/ggml/ggml.h b/include/ggml/ggml.h
index 11b51f8bd656a39b79317c16d8346d999389c182..fb3fec29cc6d6aa9f70b6f5ddca1c7d7ec0831e9 100644 (file)
--- a/include/ggml/ggml.h
+++ b/include/ggml/ggml.h
@@ -201,6 +201,8 @@
 #define GGML_MAX_NAME          48
 #define GGML_DEFAULT_N_THREADS 4
 
+#define GGML_UNUSED(x) (void)(x)
+
 #define GGML_ASSERT(x) \
     do { \
         if (!(x)) { \
@@ -209,6 +211,30 @@
         } \
     } while (0)
 
+// used to copy the number of elements and stride in bytes of tensors into local variables.
+// main purpose is to reduce code duplication and improve readability.
+//
+// example:
+//
+//    GGML_TENSOR_LOCALS(int64_t, ne1, src1, ne);
+//    GGML_TENSOR_LOCALS(size_t,  nb1, src1, nb);
+//
+#define GGML_TENSOR_LOCALS_1(type, prefix, pointer, array) \
+    const type prefix##0 = (pointer)->array[0]; \
+    GGML_UNUSED(prefix##0);
+#define GGML_TENSOR_LOCALS_2(type, prefix, pointer, array) \
+    GGML_TENSOR_LOCALS_1    (type, prefix, pointer, array) \
+    const type prefix##1 = (pointer)->array[1]; \
+    GGML_UNUSED(prefix##1);
+#define GGML_TENSOR_LOCALS_3(type, prefix, pointer, array) \
+    GGML_TENSOR_LOCALS_2    (type, prefix, pointer, array) \
+    const type prefix##2 = (pointer)->array[2]; \
+    GGML_UNUSED(prefix##2);
+#define GGML_TENSOR_LOCALS(type, prefix, pointer, array) \
+    GGML_TENSOR_LOCALS_3  (type, prefix, pointer, array) \
+    const type prefix##3 = (pointer)->array[3]; \
+    GGML_UNUSED(prefix##3);
+
 #ifdef  __cplusplus
 extern "C" {
 #endif

diff --git a/src/ggml.c b/src/ggml.c
index afeb72ff00ccfc3872849823d6c62e2f43d40be9..7eaff45cc626ccb692e1ab5db0145c834d11c6e6 100644 (file)
--- a/src/ggml.c
+++ b/src/ggml.c
@@ -220,9 +220,27 @@ inline static void* ggml_aligned_malloc(size_t size) {
 #define GGML_ALIGNED_FREE(ptr)     free(ptr)
 #endif
 
-#define UNUSED(x) (void)(x)
+#define UNUSED GGML_UNUSED
 #define SWAP(x, y, T) do { T SWAP = x; x = y; y = SWAP; } while (0)
 
+//
+// tensor access macros
+//
+
+#define GGML_TENSOR_UNARY_OP_LOCALS \
+    GGML_TENSOR_LOCALS(int64_t, ne0, src0, ne); \
+    GGML_TENSOR_LOCALS(size_t,  nb0, src0, nb); \
+    GGML_TENSOR_LOCALS(int64_t, ne,  dst,  ne); \
+    GGML_TENSOR_LOCALS(size_t,  nb,  dst,  nb);
+
+#define GGML_TENSOR_BINARY_OP_LOCALS \
+    GGML_TENSOR_LOCALS(int64_t, ne0, src0, ne); \
+    GGML_TENSOR_LOCALS(size_t,  nb0, src0, nb); \
+    GGML_TENSOR_LOCALS(int64_t, ne1, src1, ne); \
+    GGML_TENSOR_LOCALS(size_t,  nb1, src1, nb); \
+    GGML_TENSOR_LOCALS(int64_t, ne,  dst,  ne); \
+    GGML_TENSOR_LOCALS(size_t,  nb,  dst,  nb);
+
 #if defined(GGML_USE_ACCELERATE)
 #include <Accelerate/Accelerate.h>
 #if defined(GGML_USE_CLBLAST) // allow usage of CLBlast alongside Accelerate functions
@@ -7603,25 +7621,7 @@ static void ggml_compute_forward_dup_f16(
         return;
     }
 
-    const int64_t ne00 = src0->ne[0];
-    const int64_t ne01 = src0->ne[1];
-    const int64_t ne02 = src0->ne[2];
-    const int64_t ne03 = src0->ne[3];
-
-    const int64_t ne0 = dst->ne[0];
-    const int64_t ne1 = dst->ne[1];
-    const int64_t ne2 = dst->ne[2];
-    const int64_t ne3 = dst->ne[3];
-
-    const size_t nb00 = src0->nb[0];
-    const size_t nb01 = src0->nb[1];
-    const size_t nb02 = src0->nb[2];
-    const size_t nb03 = src0->nb[3];
-
-    const size_t nb0 = dst->nb[0];
-    const size_t nb1 = dst->nb[1];
-    const size_t nb2 = dst->nb[2];
-    const size_t nb3 = dst->nb[3];
+    GGML_TENSOR_UNARY_OP_LOCALS;
 
     const int ith = params->ith; // thread index
     const int nth = params->nth; // number of threads
@@ -7892,25 +7892,7 @@ static void ggml_compute_forward_dup_f32(
         return;
     }
 
-    const int64_t ne00 = src0->ne[0];
-    const int64_t ne01 = src0->ne[1];
-    const int64_t ne02 = src0->ne[2];
-    const int64_t ne03 = src0->ne[3];
-
-    const int64_t ne0 = dst->ne[0];
-    const int64_t ne1 = dst->ne[1];
-    const int64_t ne2 = dst->ne[2];
-    const int64_t ne3 = dst->ne[3];
-
-    const size_t nb00 = src0->nb[0];
-    const size_t nb01 = src0->nb[1];
-    const size_t nb02 = src0->nb[2];
-    const size_t nb03 = src0->nb[3];
-
-    const size_t nb0 = dst->nb[0];
-    const size_t nb1 = dst->nb[1];
-    const size_t nb2 = dst->nb[2];
-    const size_t nb3 = dst->nb[3];
+    GGML_TENSOR_UNARY_OP_LOCALS;
 
     const int ith = params->ith; // thread index
     const int nth = params->nth; // number of threads
@@ -8208,24 +8190,8 @@ static void ggml_compute_forward_add_f32(
     const int nth = params->nth;
 
     const int nr  = ggml_nrows(src0);
-    const int64_t ne0 = src0->ne[0];
-    const int64_t ne1 = src0->ne[1];
-    const int64_t ne2 = src0->ne[2];
-
-    const size_t nb00 = src0->nb[0];
-    const size_t nb01 = src0->nb[1];
-    const size_t nb02 = src0->nb[2];
-    const size_t nb03 = src0->nb[3];
 
-    const size_t nb10 = src1->nb[0];
-    const size_t nb11 = src1->nb[1];
-    const size_t nb12 = src1->nb[2];
-    const size_t nb13 = src1->nb[3];
-
-    const size_t nb0 = dst->nb[0];
-    const size_t nb1 = dst->nb[1];
-    const size_t nb2 = dst->nb[2];
-    const size_t nb3 = dst->nb[3];
+    GGML_TENSOR_BINARY_OP_LOCALS;
 
     GGML_ASSERT( nb0 == sizeof(float));
     GGML_ASSERT(nb00 == sizeof(float));
@@ -8294,28 +8260,12 @@ static void ggml_compute_forward_add_f16_f32(
     const int nth = params->nth;
 
     const int nr  = ggml_nrows(src0);
-    const int64_t ne0 = src0->ne[0];
-    const int64_t ne1 = src0->ne[1];
-    const int64_t ne2 = src0->ne[2];
 
-    const size_t nb00 = src0->nb[0];
-    const size_t nb01 = src0->nb[1];
-    const size_t nb02 = src0->nb[2];
-    const size_t nb03 = src0->nb[3];
-
-    const size_t nb10 = src1->nb[0];
-    const size_t nb11 = src1->nb[1];
-    const size_t nb12 = src1->nb[2];
-    const size_t nb13 = src1->nb[3];
-
-    const size_t nb0 = dst->nb[0];
-    const size_t nb1 = dst->nb[1];
-    const size_t nb2 = dst->nb[2];
-    const size_t nb3 = dst->nb[3];
+    GGML_TENSOR_BINARY_OP_LOCALS;
 
     GGML_ASSERT(src0->type == GGML_TYPE_F16);
     GGML_ASSERT(src1->type == GGML_TYPE_F32);
-    GGML_ASSERT(dst->type == GGML_TYPE_F16);
+    GGML_ASSERT(dst->type  == GGML_TYPE_F16);
 
     GGML_ASSERT( nb0 == sizeof(ggml_fp16_t));
     GGML_ASSERT(nb00 == sizeof(ggml_fp16_t));
@@ -8364,24 +8314,8 @@ static void ggml_compute_forward_add_f16_f16(
     const int nth = params->nth;
 
     const int nr  = ggml_nrows(src0);
-    const int64_t ne0 = src0->ne[0];
-    const int64_t ne1 = src0->ne[1];
-    const int64_t ne2 = src0->ne[2];
 
-    const size_t nb00 = src0->nb[0];
-    const size_t nb01 = src0->nb[1];
-    const size_t nb02 = src0->nb[2];
-    const size_t nb03 = src0->nb[3];
-
-    const size_t nb10 = src1->nb[0];
-    const size_t nb11 = src1->nb[1];
-    const size_t nb12 = src1->nb[2];
-    const size_t nb13 = src1->nb[3];
-
-    const size_t nb0 = dst->nb[0];
-    const size_t nb1 = dst->nb[1];
-    const size_t nb2 = dst->nb[2];
-    const size_t nb3 = dst->nb[3];
+    GGML_TENSOR_BINARY_OP_LOCALS;
 
     GGML_ASSERT(src0->type == GGML_TYPE_F16);
     GGML_ASSERT(src1->type == GGML_TYPE_F16);
@@ -8431,25 +8365,8 @@ static void ggml_compute_forward_add_q_f32(
     }
 
     const int nr  = ggml_nrows(src0);
-    const int64_t ne00 = src0->ne[0];
-    const int64_t ne01 = src0->ne[1];
-    const int64_t ne02 = src0->ne[2];
-    //const int64_t ne03 = src0->ne[3];
-
-    const size_t nb00 = src0->nb[0];
-    const size_t nb01 = src0->nb[1];
-    const size_t nb02 = src0->nb[2];
-    const size_t nb03 = src0->nb[3];
 
-    const size_t nb10 = src1->nb[0];
-    const size_t nb11 = src1->nb[1];
-    const size_t nb12 = src1->nb[2];
-    const size_t nb13 = src1->nb[3];
-
-    const size_t nb0  = dst->nb[0];
-    const size_t nb1  = dst->nb[1];
-    const size_t nb2  = dst->nb[2];
-    const size_t nb3  = dst->nb[3];
+    GGML_TENSOR_BINARY_OP_LOCALS;
 
     const int ith = params->ith;
     const int nth = params->nth;
@@ -8570,19 +8487,8 @@ static void ggml_compute_forward_add1_f32(
     const int nth = params->nth;
 
     const int nr  = ggml_nrows(src0);
-    const int64_t ne0 = src0->ne[0];
-    const int64_t ne1 = src0->ne[1];
-    const int64_t ne2 = src0->ne[2];
 
-    const size_t nb00 = src0->nb[0];
-    const size_t nb01 = src0->nb[1];
-    const size_t nb02 = src0->nb[2];
-    const size_t nb03 = src0->nb[3];
-
-    const size_t nb0 = dst->nb[0];
-    const size_t nb1 = dst->nb[1];
-    const size_t nb2 = dst->nb[2];
-    const size_t nb3 = dst->nb[3];
+    GGML_TENSOR_UNARY_OP_LOCALS;
 
     GGML_ASSERT( nb0 == sizeof(float));
     GGML_ASSERT(nb00 == sizeof(float));
@@ -8636,23 +8542,12 @@ static void ggml_compute_forward_add1_f16_f32(
     const int nth = params->nth;
 
     const int nr  = ggml_nrows(src0);
-    const int64_t ne0 = src0->ne[0];
-    const int64_t ne1 = src0->ne[1];
-    const int64_t ne2 = src0->ne[2];
 
-    const size_t nb00 = src0->nb[0];
-    const size_t nb01 = src0->nb[1];
-    const size_t nb02 = src0->nb[2];
-    const size_t nb03 = src0->nb[3];
-
-    const size_t nb0 = dst->nb[0];
-    const size_t nb1 = dst->nb[1];
-    const size_t nb2 = dst->nb[2];
-    const size_t nb3 = dst->nb[3];
+    GGML_TENSOR_UNARY_OP_LOCALS;
 
     GGML_ASSERT(src0->type == GGML_TYPE_F16);
     GGML_ASSERT(src1->type == GGML_TYPE_F32);
-    GGML_ASSERT(dst->type == GGML_TYPE_F16);
+    GGML_ASSERT(dst->type  == GGML_TYPE_F16);
 
     GGML_ASSERT( nb0 == sizeof(ggml_fp16_t));
     GGML_ASSERT(nb00 == sizeof(ggml_fp16_t));
@@ -8697,23 +8592,12 @@ static void ggml_compute_forward_add1_f16_f16(
     const int nth = params->nth;
 
     const int nr  = ggml_nrows(src0);
-    const int64_t ne0 = src0->ne[0];
-    const int64_t ne1 = src0->ne[1];
-    const int64_t ne2 = src0->ne[2];
 
-    const size_t nb00 = src0->nb[0];
-    const size_t nb01 = src0->nb[1];
-    const size_t nb02 = src0->nb[2];
-    const size_t nb03 = src0->nb[3];
-
-    const size_t nb0 = dst->nb[0];
-    const size_t nb1 = dst->nb[1];
-    const size_t nb2 = dst->nb[2];
-    const size_t nb3 = dst->nb[3];
+    GGML_TENSOR_UNARY_OP_LOCALS;
 
     GGML_ASSERT(src0->type == GGML_TYPE_F16);
     GGML_ASSERT(src1->type == GGML_TYPE_F16);
-    GGML_ASSERT(dst->type == GGML_TYPE_F16);
+    GGML_ASSERT(dst->type  == GGML_TYPE_F16);
 
     GGML_ASSERT( nb0 == sizeof(ggml_fp16_t));
     GGML_ASSERT(nb00 == sizeof(ggml_fp16_t));
@@ -8758,19 +8642,8 @@ static void ggml_compute_forward_add1_q_f32(
     const int nth = params->nth;
 
     const int nr  = ggml_nrows(src0);
-    const int64_t ne0 = src0->ne[0];
-    const int64_t ne1 = src0->ne[1];
-    const int64_t ne2 = src0->ne[2];
-
-    const size_t nb00 = src0->nb[0];
-    const size_t nb01 = src0->nb[1];
-    const size_t nb02 = src0->nb[2];
-    const size_t nb03 = src0->nb[3];
 
-    const size_t nb0 = dst->nb[0];
-    const size_t nb1 = dst->nb[1];
-    const size_t nb2 = dst->nb[2];
-    const size_t nb3 = dst->nb[3];
+    GGML_TENSOR_UNARY_OP_LOCALS;
 
     const enum ggml_type type = src0->type;
     dequantize_row_q_t const dequantize_row_q = quantize_fns[type].dequantize_row_q;
@@ -8902,15 +8775,8 @@ static void ggml_compute_forward_acc_f32(
     const int nr = ggml_nrows(src1);
     const int nc = src1->ne[0];
 
-    const int64_t ne10 = src1->ne[0];
-    const int64_t ne11 = src1->ne[1];
-    const int64_t ne12 = src1->ne[2];
-    const int64_t ne13 = src1->ne[3];
-
-    const size_t nb10 = src1->nb[0];
-    const size_t nb11 = src1->nb[1];
-    const size_t nb12 = src1->nb[2];
-    const size_t nb13 = src1->nb[3];
+    GGML_TENSOR_LOCALS(int64_t, ne1, src1, ne);
+    GGML_TENSOR_LOCALS(size_t,  nb1, src1, nb);
 
     // src0 and dst as viewed during acc
     const size_t nb0 = ggml_element_size(src0);
@@ -8999,24 +8865,8 @@ static void ggml_compute_forward_sub_f32(
     }
 
     const int nr  = ggml_nrows(src0);
-    const int64_t ne0 = src0->ne[0];
-    const int64_t ne1 = src0->ne[1];
-    const int64_t ne2 = src0->ne[2];
-
-    const size_t nb00 = src0->nb[0];
-    const size_t nb01 = src0->nb[1];
-    const size_t nb02 = src0->nb[2];
-    const size_t nb03 = src0->nb[3];
 
-    const size_t nb10 = src1->nb[0];
-    const size_t nb11 = src1->nb[1];
-    const size_t nb12 = src1->nb[2];
-    const size_t nb13 = src1->nb[3];
-
-    const size_t nb0 = dst->nb[0];
-    const size_t nb1 = dst->nb[1];
-    const size_t nb2 = dst->nb[2];
-    const size_t nb3 = dst->nb[3];
+    GGML_TENSOR_BINARY_OP_LOCALS;
 
     GGML_ASSERT( nb0 == sizeof(float));
     GGML_ASSERT(nb00 == sizeof(float));
@@ -9106,29 +8956,7 @@ static void ggml_compute_forward_mul_f32(
 
     const int64_t nr = ggml_nrows(src0);
 
-    const int64_t ne00 = src0->ne[0];
-    const int64_t ne01 = src0->ne[1];
-    const int64_t ne02 = src0->ne[2];
-
-    const int64_t ne10 = src1->ne[0];
-    const int64_t ne11 = src1->ne[1];
-    const int64_t ne12 = src1->ne[2];
-    const int64_t ne13 = src1->ne[3];
-
-    const size_t nb00 = src0->nb[0];
-    const size_t nb01 = src0->nb[1];
-    const size_t nb02 = src0->nb[2];
-    const size_t nb03 = src0->nb[3];
-
-    const size_t nb10 = src1->nb[0];
-    const size_t nb11 = src1->nb[1];
-    const size_t nb12 = src1->nb[2];
-    const size_t nb13 = src1->nb[3];
-
-    const size_t nb0 = dst->nb[0];
-    const size_t nb1 = dst->nb[1];
-    const size_t nb2 = dst->nb[2];
-    const size_t nb3 = dst->nb[3];
+    GGML_TENSOR_BINARY_OP_LOCALS;
 
     GGML_ASSERT( nb0 == sizeof(float));
     GGML_ASSERT(nb00 == sizeof(float));
@@ -9216,24 +9044,8 @@ static void ggml_compute_forward_div_f32(
     }
 
     const int nr  = ggml_nrows(src0);
-    const int64_t ne0 = src0->ne[0];
-    const int64_t ne1 = src0->ne[1];
-    const int64_t ne2 = src0->ne[2];
 
-    const size_t nb00 = src0->nb[0];
-    const size_t nb01 = src0->nb[1];
-    const size_t nb02 = src0->nb[2];
-    const size_t nb03 = src0->nb[3];
-
-    const size_t nb10 = src1->nb[0];
-    const size_t nb11 = src1->nb[1];
-    const size_t nb12 = src1->nb[2];
-    const size_t nb13 = src1->nb[3];
-
-    const size_t nb0 = dst->nb[0];
-    const size_t nb1 = dst->nb[1];
-    const size_t nb2 = dst->nb[2];
-    const size_t nb3 = dst->nb[3];
+    GGML_TENSOR_BINARY_OP_LOCALS;
 
     GGML_ASSERT( nb0 == sizeof(float));
     GGML_ASSERT(nb00 == sizeof(float));
@@ -9440,14 +9252,8 @@ static void ggml_compute_forward_sum_f32(
     assert(ggml_is_scalar(dst));
     assert(src0->nb[0] == sizeof(float));
 
-    const int64_t ne00 = src0->ne[0];
-    const int64_t ne01 = src0->ne[1];
-    const int64_t ne02 = src0->ne[2];
-    const int64_t ne03 = src0->ne[3];
-
-    const size_t nb01 = src0->nb[1];
-    const size_t nb02 = src0->nb[2];
-    const size_t nb03 = src0->nb[3];
+    GGML_TENSOR_LOCALS(int64_t, ne0, src0, ne);
+    GGML_TENSOR_LOCALS(size_t,  nb0, src0, nb);
 
     ggml_float sum     = 0;
     ggml_float row_sum = 0;
@@ -9496,29 +9302,13 @@ static void ggml_compute_forward_sum_rows_f32(
     GGML_ASSERT(src0->nb[0] == sizeof(float));
     GGML_ASSERT(dst->nb[0] == sizeof(float));
 
-    const int64_t ne00 = src0->ne[0];
-    const int64_t ne01 = src0->ne[1];
-    const int64_t ne02 = src0->ne[2];
-    const int64_t ne03 = src0->ne[3];
-
-    const int64_t ne0 = dst->ne[0];
-    const int64_t ne1 = dst->ne[1];
-    const int64_t ne2 = dst->ne[2];
-    const int64_t ne3 = dst->ne[3];
+    GGML_TENSOR_UNARY_OP_LOCALS;
 
     GGML_ASSERT(ne0 == 1);
     GGML_ASSERT(ne1 == ne01);
     GGML_ASSERT(ne2 == ne02);
     GGML_ASSERT(ne3 == ne03);
 
-    const size_t nb01 = src0->nb[1];
-    const size_t nb02 = src0->nb[2];
-    const size_t nb03 = src0->nb[3];
-
-    const size_t nb1 = dst->nb[1];
-    const size_t nb2 = dst->nb[2];
-    const size_t nb3 = dst->nb[3];
-
     for (int64_t i3 = 0; i3 < ne03; i3++) {
         for (int64_t i2 = 0; i2 < ne02; i2++) {
             for (int64_t i1 = 0; i1 < ne01; i1++) {
@@ -9562,19 +9352,7 @@ static void ggml_compute_forward_mean_f32(
 
     assert(src0->nb[0] == sizeof(float));
 
-    const int64_t ne00 = src0->ne[0];
-    const int64_t ne01 = src0->ne[1];
-    const int64_t ne02 = src0->ne[2];
-    const int64_t ne03 = src0->ne[3];
-
-    const size_t nb01 = src0->nb[1];
-    const size_t nb02 = src0->nb[2];
-    const size_t nb03 = src0->nb[3];
-
-    const int64_t ne0 = dst->ne[0];
-    const int64_t ne1 = dst->ne[1];
-    const int64_t ne2 = dst->ne[2];
-    const int64_t ne3 = dst->ne[3];
+    GGML_TENSOR_UNARY_OP_LOCALS;
 
     assert(ne0 == 1);
     assert(ne1 == ne01);
@@ -9586,10 +9364,6 @@ static void ggml_compute_forward_mean_f32(
     UNUSED(ne2);
     UNUSED(ne3);
 
-    const size_t nb1 = dst->nb[1];
-    const size_t nb2 = dst->nb[2];
-    const size_t nb3 = dst->nb[3];
-
     for (int64_t i03 = 0; i03 < ne03; i03++) {
         for (int64_t i02 = 0; i02 < ne02; i02++) {
             for (int64_t i01 = 0; i01 < ne01; i01++) {
@@ -9632,25 +9406,7 @@ static void ggml_compute_forward_repeat_f32(
         return;
     }
 
-    const int64_t ne0  = dst->ne[0];
-    const int64_t ne1  = dst->ne[1];
-    const int64_t ne2  = dst->ne[2];
-    const int64_t ne3  = dst->ne[3];
-
-    const int64_t ne00 = src0->ne[0];
-    const int64_t ne01 = src0->ne[1];
-    const int64_t ne02 = src0->ne[2];
-    const int64_t ne03 = src0->ne[3];
-
-    const size_t nb0  = dst->nb[0];
-    const size_t nb1  = dst->nb[1];
-    const size_t nb2  = dst->nb[2];
-    const size_t nb3  = dst->nb[3];
-
-    const size_t nb00 = src0->nb[0];
-    const size_t nb01 = src0->nb[1];
-    const size_t nb02 = src0->nb[2];
-    const size_t nb03 = src0->nb[3];
+    GGML_TENSOR_UNARY_OP_LOCALS;
 
     // guaranteed to be an integer due to the check in ggml_can_repeat
     const int nr0 = (int)(ne0/ne00);
@@ -9711,25 +9467,7 @@ static void ggml_compute_forward_repeat_back_f32(
         return;
     }
 
-    const int64_t ne0  = dst->ne[0];
-    const int64_t ne1  = dst->ne[1];
-    const int64_t ne2  = dst->ne[2];
-    const int64_t ne3  = dst->ne[3];
-
-    const int64_t ne00 = src0->ne[0];
-    const int64_t ne01 = src0->ne[1];
-    const int64_t ne02 = src0->ne[2];
-    const int64_t ne03 = src0->ne[3];
-
-    const size_t nb0  = dst->nb[0];
-    const size_t nb1  = dst->nb[1];
-    const size_t nb2  = dst->nb[2];
-    const size_t nb3  = dst->nb[3];
-
-    const size_t nb00 = src0->nb[0];
-    const size_t nb01 = src0->nb[1];
-    const size_t nb02 = src0->nb[2];
-    const size_t nb03 = src0->nb[3];
+    GGML_TENSOR_UNARY_OP_LOCALS;
 
     // guaranteed to be an integer due to the check in ggml_can_repeat
     const int nr0 = (int)(ne00/ne0);
@@ -10260,18 +9998,7 @@ static void ggml_compute_forward_norm_f32(
     const int ith = params->ith;
     const int nth = params->nth;
 
-    const int64_t ne00 = src0->ne[0];
-    const int64_t ne01 = src0->ne[1];
-    const int64_t ne02 = src0->ne[2];
-    const int64_t ne03 = src0->ne[3];
-
-    const size_t nb01 = src0->nb[1];
-    const size_t nb02 = src0->nb[2];
-    const size_t nb03 = src0->nb[3];
-
-    const size_t nb1 = dst->nb[1];
-    const size_t nb2 = dst->nb[2];
-    const size_t nb3 = dst->nb[3];
+    GGML_TENSOR_UNARY_OP_LOCALS;
 
     const float eps = 1e-5f; // TODO: make this a parameter
 
@@ -10337,18 +10064,7 @@ static void ggml_compute_forward_rms_norm_f32(
     const int ith = params->ith;
     const int nth = params->nth;
 
-    const int64_t ne00 = src0->ne[0];
-    const int64_t ne01 = src0->ne[1];
-    const int64_t ne02 = src0->ne[2];
-    const int64_t ne03 = src0->ne[3];
-
-    const size_t nb01 = src0->nb[1];
-    const size_t nb02 = src0->nb[2];
-    const size_t nb03 = src0->nb[3];
-
-    const size_t nb1 = dst->nb[1];
-    const size_t nb2 = dst->nb[2];
-    const size_t nb3 = dst->nb[3];
+    GGML_TENSOR_UNARY_OP_LOCALS;
 
     const float eps = 1e-6f; // TODO: make this a parameter
 
@@ -10413,22 +10129,7 @@ static void ggml_compute_forward_rms_norm_back_f32(
     const int ith = params->ith;
     const int nth = params->nth;
 
-    const int64_t ne00 = src0->ne[0];
-    const int64_t ne01 = src0->ne[1];
-    const int64_t ne02 = src0->ne[2];
-    const int64_t ne03 = src0->ne[3];
-
-    const size_t nb01 = src0->nb[1];
-    const size_t nb02 = src0->nb[2];
-    const size_t nb03 = src0->nb[3];
-
-    const size_t nb11 = src1->nb[1];
-    const size_t nb12 = src1->nb[2];
-    const size_t nb13 = src1->nb[3];
-
-    const size_t nb1 = dst->nb[1];
-    const size_t nb2 = dst->nb[2];
-    const size_t nb3 = dst->nb[3];
+    GGML_TENSOR_BINARY_OP_LOCALS;
 
     const float eps = 1e-6f; // TODO: make this a parameter
 
@@ -10624,41 +10325,7 @@ static void ggml_compute_forward_mul_mat_f32(
     int64_t t0 = ggml_perf_time_us();
     UNUSED(t0);
 
-    const int64_t ne00 = src0->ne[0];
-    const int64_t ne01 = src0->ne[1];
-    const int64_t ne02 = src0->ne[2];
-    const int64_t ne03 = src0->ne[3];
-
-#if defined(GGML_USE_ACCELERATE) || defined(GGML_USE_OPENBLAS)
-    const int64_t ne10 = src1->ne[0];
-#endif
-    const int64_t ne11 = src1->ne[1];
-#ifndef NDEBUG
-    const int64_t ne12 = src1->ne[2];
-    const int64_t ne13 = src1->ne[3];
-
-    const int64_t ne0  = dst->ne[0];
-    const int64_t ne1  = dst->ne[1];
-    const int64_t ne2  = dst->ne[2];
-    const int64_t ne3  = dst->ne[3];
-
-    const int nb00 = src0->nb[0];
-#endif
-    const int nb01 = src0->nb[1];
-    const int nb02 = src0->nb[2];
-    const int nb03 = src0->nb[3];
-
-#ifndef NDEBUG
-    const int nb10 = src1->nb[0];
-#endif
-    const int nb11 = src1->nb[1];
-    const int nb12 = src1->nb[2];
-    const int nb13 = src1->nb[3];
-
-    const int nb0  = dst->nb[0];
-    const int nb1  = dst->nb[1];
-    const int nb2  = dst->nb[2];
-    const int nb3  = dst->nb[3];
+    GGML_TENSOR_BINARY_OP_LOCALS;
 
     const int ith = params->ith;
     const int nth = params->nth;
@@ -10795,37 +10462,10 @@ static void ggml_compute_forward_mul_mat_f16_f32(
     int64_t t0 = ggml_perf_time_us();
     UNUSED(t0);
 
-    const int64_t ne00 = src0->ne[0];
-    const int64_t ne01 = src0->ne[1];
-    const int64_t ne02 = src0->ne[2];
-    const int64_t ne03 = src0->ne[3];
+    GGML_TENSOR_BINARY_OP_LOCALS;
 
-    const int64_t ne10 = src1->ne[0];
-    const int64_t ne11 = src1->ne[1];
-    const int64_t ne12 = src1->ne[2];
-    const int64_t ne13 = src1->ne[3];
-
-    const int64_t ne0  = dst->ne[0];
-    const int64_t ne1  = dst->ne[1];
-    const int64_t ne2  = dst->ne[2];
-    const int64_t ne3  = dst->ne[3];
     //const int64_t ne   = ne0*ne1*ne2*ne3;
 
-    const int nb00 = src0->nb[0];
-    const int nb01 = src0->nb[1];
-    const int nb02 = src0->nb[2];
-    const int nb03 = src0->nb[3];
-
-    const int nb10 = src1->nb[0];
-    const int nb11 = src1->nb[1];
-    const int nb12 = src1->nb[2];
-    const int nb13 = src1->nb[3];
-
-    const int nb0  = dst->nb[0];
-    const int nb1  = dst->nb[1];
-    const int nb2  = dst->nb[2];
-    const int nb3  = dst->nb[3];
-
     const int ith = params->ith;
     const int nth = params->nth;
 
@@ -10995,35 +10635,7 @@ static void ggml_compute_forward_mul_mat_q_f32(
     int64_t t0 = ggml_perf_time_us();
     UNUSED(t0);
 
-    const int64_t ne00 = src0->ne[0];
-    const int64_t ne01 = src0->ne[1];
-    const int64_t ne02 = src0->ne[2];
-    const int64_t ne03 = src0->ne[3];
-
-    const int64_t ne10 = src1->ne[0];
-    const int64_t ne11 = src1->ne[1];
-    const int64_t ne12 = src1->ne[2];
-    const int64_t ne13 = src1->ne[3];
-
-    const int64_t ne0  = dst->ne[0];
-    const int64_t ne1  = dst->ne[1];
-    const int64_t ne2  = dst->ne[2];
-    const int64_t ne3  = dst->ne[3];
-
-    const int nb00 = src0->nb[0];
-    const int nb01 = src0->nb[1];
-    const int nb02 = src0->nb[2];
-    const int nb03 = src0->nb[3];
-
-    const int nb10 = src1->nb[0];
-    const int nb11 = src1->nb[1];
-    const int nb12 = src1->nb[2];
-    const int nb13 = src1->nb[3];
-
-    const int nb0  = dst->nb[0];
-    const int nb1  = dst->nb[1];
-    const int nb2  = dst->nb[2];
-    const int nb3  = dst->nb[3];
+    GGML_TENSOR_BINARY_OP_LOCALS;
 
     const int ith = params->ith;
     const int nth = params->nth;
@@ -11039,7 +10651,7 @@ static void ggml_compute_forward_mul_mat_q_f32(
     enum ggml_type   const vec_dot_type       = quantize_fns[type].vec_dot_type;
 
     // we don't support permuted src0 or src1
-    GGML_ASSERT(nb00 == (int) GGML_TYPE_SIZE[type]);
+    GGML_ASSERT(nb00 == GGML_TYPE_SIZE[type]);
     GGML_ASSERT(nb10 == sizeof(float));
 
     // dst cannot be transposed or permuted
@@ -11233,35 +10845,7 @@ static void ggml_compute_forward_out_prod_f32(
     int64_t t0 = ggml_perf_time_us();
     UNUSED(t0);
 
-    const int64_t ne00 = src0->ne[0];
-    const int64_t ne01 = src0->ne[1];
-    const int64_t ne02 = src0->ne[2];
-    const int64_t ne03 = src0->ne[3];
-
-    const int64_t ne10 = src1->ne[0];
-    //const int64_t ne11 = src1->ne[1];
-    const int64_t ne12 = src1->ne[2];
-    const int64_t ne13 = src1->ne[3];
-
-    const int64_t ne0  = dst->ne[0];
-    const int64_t ne1  = dst->ne[1];
-    const int64_t ne2  = dst->ne[2];
-    const int64_t ne3  = dst->ne[3];
-
-    const int nb00 = src0->nb[0];
-    const int nb01 = src0->nb[1];
-    const int nb02 = src0->nb[2];
-    const int nb03 = src0->nb[3];
-
-    const int nb10 = src1->nb[0];
-    const int nb11 = src1->nb[1];
-    const int nb12 = src1->nb[2];
-    const int nb13 = src1->nb[3];
-
-    const int nb0  = dst->nb[0];
-    const int nb1  = dst->nb[1];
-    const int nb2  = dst->nb[2];
-    const int nb3  = dst->nb[3];
+    GGML_TENSOR_BINARY_OP_LOCALS;
 
     const int ith = params->ith;
     const int nth = params->nth;
@@ -11496,15 +11080,8 @@ static void ggml_compute_forward_set_f32(
     const int nr = ggml_nrows(src1);
     const int nc = src1->ne[0];
 
-    const int64_t ne10 = src1->ne[0];
-    const int64_t ne11 = src1->ne[1];
-    const int64_t ne12 = src1->ne[2];
-    const int64_t ne13 = src1->ne[3];
-
-    const size_t nb10 = src1->nb[0];
-    const size_t nb11 = src1->nb[1];
-    const size_t nb12 = src1->nb[2];
-    const size_t nb13 = src1->nb[3];
+    GGML_TENSOR_LOCALS(int64_t, ne1, src1, ne);
+    GGML_TENSOR_LOCALS(size_t,  nb1, src1, nb);
 
     // src0 and dst as viewed during set
     const size_t nb0 = ggml_element_size(src0);
@@ -11895,29 +11472,14 @@ static void ggml_compute_forward_diag_f32(
 
     // TODO: handle transposed/permuted matrices
 
-    const int ne00 = src0->ne[0];
-    const int ne01 = src0->ne[1];
-    const int ne02 = src0->ne[2];
-    const int ne03 = src0->ne[3];
-    const int ne0 = dst->ne[0];
-    const int ne1 = dst->ne[1];
-    const int ne2 = dst->ne[2];
-    const int ne3 = dst->ne[3];
+    GGML_TENSOR_UNARY_OP_LOCALS;
+
     GGML_ASSERT(ne00 == ne0);
     GGML_ASSERT(ne00 == ne1);
     GGML_ASSERT(ne01 == 1);
     GGML_ASSERT(ne02 == ne2);
     GGML_ASSERT(ne03 == ne3);
 
-    const int nb00 = src0->nb[0];
-    //const int nb01 = src0->nb[1];
-    const int nb02 = src0->nb[2];
-    const int nb03 = src0->nb[3];
-    const int nb0 = dst->nb[0];
-    const int nb1 = dst->nb[1];
-    const int nb2 = dst->nb[2];
-    const int nb3 = dst->nb[3];
-
     GGML_ASSERT(nb00 == sizeof(float));
     GGML_ASSERT(nb0  == sizeof(float));
 
@@ -12494,20 +12056,7 @@ static void ggml_compute_forward_rope_f32(
 
     assert(n_past >= 0);
 
-    const size_t nb00 = src0->nb[0];
-    const size_t nb01 = src0->nb[1];
-    const size_t nb02 = src0->nb[2];
-    const size_t nb03 = src0->nb[3];
-
-    const int64_t ne0 = dst->ne[0];
-    const int64_t ne1 = dst->ne[1];
-    const int64_t ne2 = dst->ne[2];
-    const int64_t ne3 = dst->ne[3];
-
-    const size_t nb0 = dst->nb[0];
-    const size_t nb1 = dst->nb[1];
-    const size_t nb2 = dst->nb[2];
-    const size_t nb3 = dst->nb[3];
+    GGML_TENSOR_UNARY_OP_LOCALS;
 
     //printf("ne0: %d, ne1: %d, ne2: %d, ne3: %d\n", ne0, ne1, ne2, ne3);
     //printf("n_past = %d, ne2 = %d\n", n_past, ne2);
@@ -12634,20 +12183,7 @@ static void ggml_compute_forward_rope_f16(
 
     assert(n_past >= 0);
 
-    const size_t nb00 = src0->nb[0];
-    const size_t nb01 = src0->nb[1];
-    const size_t nb02 = src0->nb[2];
-    const size_t nb03 = src0->nb[3];
-
-    const int64_t ne0 = dst->ne[0];
-    const int64_t ne1 = dst->ne[1];
-    const int64_t ne2 = dst->ne[2];
-    const int64_t ne3 = dst->ne[3];
-
-    const size_t nb0 = dst->nb[0];
-    const size_t nb1 = dst->nb[1];
-    const size_t nb2 = dst->nb[2];
-    const size_t nb3 = dst->nb[3];
+    GGML_TENSOR_UNARY_OP_LOCALS;
 
     //printf("ne0: %d, ne1: %d, ne2: %d, ne3: %d\n", ne0, ne1, ne2, ne3);
     //printf("n_past = %d, ne2 = %d\n", n_past, ne2);
@@ -12800,21 +12336,7 @@ static void ggml_compute_forward_rope_back_f32(
 
     assert(n_past >= 0);
 
-    const size_t nb00 = src0->nb[0];
-    const size_t nb01 = src0->nb[1];
-    const size_t nb02 = src0->nb[2];
-    const size_t nb03 = src0->nb[3];
-
-    const int64_t ne0 = dst->ne[0];
-    const int64_t ne1 = dst->ne[1];
-    const int64_t ne2 = dst->ne[2];
-    const int64_t ne3 = dst->ne[3];
-
-    const size_t nb0 = dst->nb[0];
-    const size_t nb1 = dst->nb[1];
-    const size_t nb2 = dst->nb[2];
-    const size_t nb3 = dst->nb[3];
-
+    GGML_TENSOR_UNARY_OP_LOCALS;
 
     //printf("ne0: %d, ne1: %d, ne2: %d, ne3: %d\n", ne0, ne1, ne2, ne3);
     //printf("n_past = %d, ne2 = %d\n", n_past, ne2);
@@ -12913,21 +12435,7 @@ static void ggml_compute_forward_rope_back_f16(
 
     assert(n_past >= 0);
 
-    const size_t nb00 = src0->nb[0];
-    const size_t nb01 = src0->nb[1];
-    const size_t nb02 = src0->nb[2];
-    const size_t nb03 = src0->nb[3];
-
-    const int64_t ne0 = dst->ne[0];
-    const int64_t ne1 = dst->ne[1];
-    const int64_t ne2 = dst->ne[2];
-    const int64_t ne3 = dst->ne[3];
-
-    const size_t nb0 = dst->nb[0];
-    const size_t nb1 = dst->nb[1];
-    const size_t nb2 = dst->nb[2];
-    const size_t nb3 = dst->nb[3];
-
+    GGML_TENSOR_UNARY_OP_LOCALS;
 
     //printf("ne0: %d, ne1: %d, ne2: %d, ne3: %d\n", ne0, ne1, ne2, ne3);
     //printf("n_past = %d, ne2 = %d\n", n_past, ne2);
@@ -13039,36 +12547,7 @@ static void ggml_compute_forward_conv_1d_s1_ph_f16_f32(
     int64_t t0 = ggml_perf_time_us();
     UNUSED(t0);
 
-    const int64_t ne00 = src0->ne[0];
-    const int64_t ne01 = src0->ne[1];
-    const int64_t ne02 = src0->ne[2];
-    //const int64_t ne03 = src0->ne[3];
-
-    const int64_t ne10 = src1->ne[0];
-    const int64_t ne11 = src1->ne[1];
-    //const int64_t ne12 = src1->ne[2];
-    //const int64_t ne13 = src1->ne[3];
-
-    //const int64_t ne0  = dst->ne[0];
-    //const int64_t ne1  = dst->ne[1];
-    //const int64_t ne2  = dst->ne[2];
-    //const int64_t ne3  = dst->ne[3];
-    //const int64_t ne   = ne0*ne1*ne2*ne3;
-
-    const int nb00 = src0->nb[0];
-    const int nb01 = src0->nb[1];
-    const int nb02 = src0->nb[2];
-    //const int nb03 = src0->nb[3];
-
-    const int nb10 = src1->nb[0];
-    const int nb11 = src1->nb[1];
-    //const int nb12 = src1->nb[2];
-    //const int nb13 = src1->nb[3];
-
-    //const int nb0  = dst->nb[0];
-    const int nb1  = dst->nb[1];
-    //const int nb2  = dst->nb[2];
-    //const int nb3  = dst->nb[3];
+    GGML_TENSOR_BINARY_OP_LOCALS;
 
     const int ith = params->ith;
     const int nth = params->nth;
@@ -13159,36 +12638,7 @@ static void ggml_compute_forward_conv_1d_s1_ph_f32(
     int64_t t0 = ggml_perf_time_us();
     UNUSED(t0);
 
-    const int64_t ne00 = src0->ne[0];
-    const int64_t ne01 = src0->ne[1];
-    const int64_t ne02 = src0->ne[2];
-    //const int64_t ne03 = src0->ne[3];
-
-    const int64_t ne10 = src1->ne[0];
-    const int64_t ne11 = src1->ne[1];
-    //const int64_t ne12 = src1->ne[2];
-    //const int64_t ne13 = src1->ne[3];
-
-    //const int64_t ne0  = dst->ne[0];
-    //const int64_t ne1  = dst->ne[1];
-    //const int64_t ne2  = dst->ne[2];
-    //const int64_t ne3  = dst->ne[3];
-    //const int64_t ne   = ne0*ne1*ne2*ne3;
-
-    const int nb00 = src0->nb[0];
-    const int nb01 = src0->nb[1];
-    const int nb02 = src0->nb[2];
-    //const int nb03 = src0->nb[3];
-
-    const int nb10 = src1->nb[0];
-    const int nb11 = src1->nb[1];
-    //const int nb12 = src1->nb[2];
-    //const int nb13 = src1->nb[3];
-
-    //const int nb0  = dst->nb[0];
-    const int nb1  = dst->nb[1];
-    //const int nb2  = dst->nb[2];
-    //const int nb3  = dst->nb[3];
+    GGML_TENSOR_BINARY_OP_LOCALS;
 
     const int ith = params->ith;
     const int nth = params->nth;
@@ -13302,36 +12752,7 @@ static void ggml_compute_forward_conv_1d_s2_ph_f16_f32(
     int64_t t0 = ggml_perf_time_us();
     UNUSED(t0);
 
-    const int64_t ne00 = src0->ne[0];
-    const int64_t ne01 = src0->ne[1];
-    const int64_t ne02 = src0->ne[2];
-    //const int64_t ne03 = src0->ne[3];
-
-    const int64_t ne10 = src1->ne[0];
-    const int64_t ne11 = src1->ne[1];
-    //const int64_t ne12 = src1->ne[2];
-    //const int64_t ne13 = src1->ne[3];
-
-    //const int64_t ne0  = dst->ne[0];
-    //const int64_t ne1  = dst->ne[1];
-    //const int64_t ne2  = dst->ne[2];
-    //const int64_t ne3  = dst->ne[3];
-    //const int64_t ne   = ne0*ne1*ne2*ne3;
-
-    const int nb00 = src0->nb[0];
-    const int nb01 = src0->nb[1];
-    const int nb02 = src0->nb[2];
-    //const int nb03 = src0->nb[3];
-
-    const int nb10 = src1->nb[0];
-    const int nb11 = src1->nb[1];
-    //const int nb12 = src1->nb[2];
-    //const int nb13 = src1->nb[3];
-
-    //const int nb0  = dst->nb[0];
-    const int nb1  = dst->nb[1];
-    //const int nb2  = dst->nb[2];
-    //const int nb3  = dst->nb[3];
+    GGML_TENSOR_BINARY_OP_LOCALS;
 
     const int ith = params->ith;
     const int nth = params->nth;
@@ -13422,36 +12843,7 @@ static void ggml_compute_forward_conv_1d_s2_ph_f32(
     int64_t t0 = ggml_perf_time_us();
     UNUSED(t0);
 
-    const int64_t ne00 = src0->ne[0];
-    const int64_t ne01 = src0->ne[1];
-    const int64_t ne02 = src0->ne[2];
-    //const int64_t ne03 = src0->ne[3];
-
-    const int64_t ne10 = src1->ne[0];
-    const int64_t ne11 = src1->ne[1];
-    //const int64_t ne12 = src1->ne[2];
-    //const int64_t ne13 = src1->ne[3];
-
-    //const int64_t ne0  = dst->ne[0];
-    //const int64_t ne1  = dst->ne[1];
-    //const int64_t ne2  = dst->ne[2];
-    //const int64_t ne3  = dst->ne[3];
-    //const int64_t ne   = ne0*ne1*ne2*ne3;
-
-    const int nb00 = src0->nb[0];
-    const int nb01 = src0->nb[1];
-    const int nb02 = src0->nb[2];
-    //const int nb03 = src0->nb[3];
-
-    const int nb10 = src1->nb[0];
-    const int nb11 = src1->nb[1];
-    //const int nb12 = src1->nb[2];
-    //const int nb13 = src1->nb[3];
-
-    //const int nb0  = dst->nb[0];
-    const int nb1  = dst->nb[1];
-    //const int nb2  = dst->nb[2];
-    //const int nb3  = dst->nb[3];
+    GGML_TENSOR_BINARY_OP_LOCALS;
 
     const int ith = params->ith;
     const int nth = params->nth;
@@ -13565,36 +12957,7 @@ static void ggml_compute_forward_conv_2d_sk_p0_f16_f32(
     int64_t t0 = ggml_perf_time_us();
     UNUSED(t0);
 
-    const int ne00 = src0->ne[0];
-    const int ne01 = src0->ne[1];
-    const int ne02 = src0->ne[2];
-    //const int ne03 = src0->ne[3];
-
-    const int ne10 = src1->ne[0];
-    //const int ne11 = src1->ne[1];
-    const int ne12 = src1->ne[2];
-    //const int ne13 = src1->ne[3];
-
-    const int ne0  = dst->ne[0];
-    const int ne1  = dst->ne[1];
-    const int ne2  = dst->ne[2];
-    //const int ne3  = dst->ne[3];
-    //const int ne   = ne0*ne1*ne2*ne3;
-
-    const int nb00 = src0->nb[0];
-    //const int nb01 = src0->nb[1];
-    //const int nb02 = src0->nb[2];
-    const int nb03 = src0->nb[3];
-
-    const int nb10 = src1->nb[0];
-    //const int nb11 = src1->nb[1];
-    const int nb12 = src1->nb[2];
-    //const int nb13 = src1->nb[3];
-
-    //const int nb0  = dst->nb[0];
-    //const int nb1  = dst->nb[1];
-    const int nb2  = dst->nb[2];
-    //const int nb3  = dst->nb[3];
+    GGML_TENSOR_BINARY_OP_LOCALS;
 
     const int ith = params->ith;
     const int nth = params->nth;
@@ -13699,45 +13062,14 @@ static void ggml_compute_forward_flash_attn_f32(
     int64_t t0 = ggml_perf_time_us();
     UNUSED(t0);
 
-    const int64_t neq0 = q->ne[0];
-    const int64_t neq1 = q->ne[1];
-    const int64_t neq2 = q->ne[2];
-    const int64_t neq3 = q->ne[3];
-
-    const int64_t nek0 = k->ne[0];
-    const int64_t nek1 = k->ne[1];
-    //const int64_t nek2 = k->ne[2];
-    //const int64_t nek3 = k->ne[3];
-
-    //const int64_t nev0 = v->ne[0];
-    const int64_t nev1 = v->ne[1];
-    //const int64_t nev2 = v->ne[2];
-    //const int64_t nev3 = v->ne[3];
-
-    const int64_t ne0  = dst->ne[0];
-    const int64_t ne1  = dst->ne[1];
-    //const int64_t ne2  = dst->ne[2];
-    //const int64_t ne3  = dst->ne[3];
-
-    const int nbk0 = k->nb[0];
-    const int nbk1 = k->nb[1];
-    const int nbk2 = k->nb[2];
-    const int nbk3 = k->nb[3];
-
-    const int nbq0 = q->nb[0];
-    const int nbq1 = q->nb[1];
-    const int nbq2 = q->nb[2];
-    const int nbq3 = q->nb[3];
-
-    const int nbv0 = v->nb[0];
-    const int nbv1 = v->nb[1];
-    const int nbv2 = v->nb[2];
-    const int nbv3 = v->nb[3];
-
-    const int nb0  = dst->nb[0];
-    const int nb1  = dst->nb[1];
-    const int nb2  = dst->nb[2];
-    const int nb3  = dst->nb[3];
+    GGML_TENSOR_LOCALS(int64_t, neq, q,   ne);
+    GGML_TENSOR_LOCALS(size_t,  nbq, q,   nb);
+    GGML_TENSOR_LOCALS(int64_t, nek, k,   ne);
+    GGML_TENSOR_LOCALS(size_t,  nbk, k,   nb);
+    GGML_TENSOR_LOCALS(int64_t, nev, v,   ne);
+    GGML_TENSOR_LOCALS(size_t,  nbv, v,   nb);
+    GGML_TENSOR_LOCALS(int64_t, ne,  dst, ne);
+    GGML_TENSOR_LOCALS(size_t,  nb,  dst, nb);
 
     const int ith = params->ith;
     const int nth = params->nth;
@@ -13908,45 +13240,14 @@ static void ggml_compute_forward_flash_attn_f16(
     int64_t t0 = ggml_perf_time_us();
     UNUSED(t0);
 
-    const int64_t neq0 = q->ne[0];
-    const int64_t neq1 = q->ne[1];
-    const int64_t neq2 = q->ne[2];
-    const int64_t neq3 = q->ne[3];
-
-    const int64_t nek0 = k->ne[0];
-    const int64_t nek1 = k->ne[1];
-    //const int64_t nek2 = k->ne[2];
-    //const int64_t nek3 = k->ne[3];
-
-    //const int64_t nev0 = v->ne[0];
-    const int64_t nev1 = v->ne[1];
-    //const int64_t nev2 = v->ne[2];
-    //const int64_t nev3 = v->ne[3];
-
-    const int64_t ne0  = dst->ne[0];
-    const int64_t ne1  = dst->ne[1];
-    //const int64_t ne2  = dst->ne[2];
-    //const int64_t ne3  = dst->ne[3];
-
-    const int nbk0 = k->nb[0];
-    const int nbk1 = k->nb[1];
-    const int nbk2 = k->nb[2];
-    const int nbk3 = k->nb[3];
-
-    const int nbq0 = q->nb[0];
-    const int nbq1 = q->nb[1];
-    const int nbq2 = q->nb[2];
-    const int nbq3 = q->nb[3];
-
-    const int nbv0 = v->nb[0];
-    const int nbv1 = v->nb[1];
-    const int nbv2 = v->nb[2];
-    const int nbv3 = v->nb[3];
-
-    const int nb0  = dst->nb[0];
-    const int nb1  = dst->nb[1];
-    const int nb2  = dst->nb[2];
-    const int nb3  = dst->nb[3];
+    GGML_TENSOR_LOCALS(int64_t, neq, q,   ne);
+    GGML_TENSOR_LOCALS(size_t,  nbq, q,   nb);
+    GGML_TENSOR_LOCALS(int64_t, nek, k,   ne);
+    GGML_TENSOR_LOCALS(size_t,  nbk, k,   nb);
+    GGML_TENSOR_LOCALS(int64_t, nev, v,   ne);
+    GGML_TENSOR_LOCALS(size_t,  nbv, v,   nb);
+    GGML_TENSOR_LOCALS(int64_t, ne,  dst, ne);
+    GGML_TENSOR_LOCALS(size_t,  nb,  dst, nb);
 
     const int ith = params->ith;
     const int nth = params->nth;
@@ -14180,65 +13481,18 @@ static void ggml_compute_forward_flash_ff_f16(
     int64_t t0 = ggml_perf_time_us();
     UNUSED(t0);
 
-    const int64_t nea0 = a->ne[0];
-    const int64_t nea1 = a->ne[1];
-    const int64_t nea2 = a->ne[2];
-    const int64_t nea3 = a->ne[3];
-
-    const int64_t neb00 = b0->ne[0];
-    const int64_t neb01 = b0->ne[1];
-    //const int64_t neb02 = b0->ne[2];
-    //const int64_t neb03 = b0->ne[3];
-
-    const int64_t neb10 = b1->ne[0];
-    const int64_t neb11 = b1->ne[1];
-    //const int64_t neb12 = b1->ne[2];
-    //const int64_t neb13 = b1->ne[3];
-
-    const int64_t nec00 = c0->ne[0];
-    const int64_t nec01 = c0->ne[1];
-    //const int64_t nec02 = c0->ne[2];
-    //const int64_t nec03 = c0->ne[3];
-
-    const int64_t nec10 = c1->ne[0];
-    const int64_t nec11 = c1->ne[1];
-    //const int64_t nec12 = c1->ne[2];
-    //const int64_t nec13 = c1->ne[3];
-
-    const int64_t ne0 = dst->ne[0];
-    const int64_t ne1 = dst->ne[1];
-    const int64_t ne2 = dst->ne[2];
-    //const int64_t ne3 = dst->ne[3];
-
-    const int nba0 = a->nb[0];
-    const int nba1 = a->nb[1];
-    const int nba2 = a->nb[2];
-    const int nba3 = a->nb[3];
-
-    const int nbb00 = b0->nb[0];
-    const int nbb01 = b0->nb[1];
-    const int nbb02 = b0->nb[2];
-    const int nbb03 = b0->nb[3];
-
-    const int nbb10 = b1->nb[0];
-    //const int nbb11 = b1->nb[1];
-    //const int nbb12 = b1->nb[2];
-    //const int nbb13 = b1->nb[3];
-
-    const int nbc00 = c0->nb[0];
-    const int nbc01 = c0->nb[1];
-    const int nbc02 = c0->nb[2];
-    const int nbc03 = c0->nb[3];
-
-    const int nbc10 = c1->nb[0];
-    //const int nbc11 = c1->nb[1];
-    //const int nbc12 = c1->nb[2];
-    //const int nbc13 = c1->nb[3];
-
-    const int nb0 = dst->nb[0];
-    const int nb1 = dst->nb[1];
-    const int nb2 = dst->nb[2];
-    const int nb3 = dst->nb[3];
+    GGML_TENSOR_LOCALS(int64_t, nea,  a,   ne);
+    GGML_TENSOR_LOCALS(size_t,  nba,  a,   nb);
+    GGML_TENSOR_LOCALS(int64_t, neb0, b0,  ne);
+    GGML_TENSOR_LOCALS(size_t,  nbb0, b0,  nb);
+    GGML_TENSOR_LOCALS(int64_t, neb1, b1,  ne);
+    GGML_TENSOR_LOCALS(size_t,  nbb1, b1,  nb);
+    GGML_TENSOR_LOCALS(int64_t, nec0, c0,  ne);
+    GGML_TENSOR_LOCALS(size_t,  nbc0, c0,  nb);
+    GGML_TENSOR_LOCALS(int64_t, nec1, c1,  ne);
+    GGML_TENSOR_LOCALS(size_t,  nbc1, c1,  nb);
+    GGML_TENSOR_LOCALS(int64_t, ne,   dst, ne);
+    GGML_TENSOR_LOCALS(size_t,  nb,   dst, nb);
 
     const int ith = params->ith;
     const int nth = params->nth;
@@ -14386,55 +13640,16 @@ static void ggml_compute_forward_flash_attn_back_f32(
     int64_t t0 = ggml_perf_time_us();
     UNUSED(t0);
 
-    const int64_t neq0 = q->ne[0];
-    const int64_t neq1 = q->ne[1];
-    const int64_t neq2 = q->ne[2];
-    const int64_t neq3 = q->ne[3];
-
-    const int64_t nek0 = k->ne[0];
-    const int64_t nek1 = k->ne[1];
-    //const int64_t nek2 = k->ne[2];
-    //const int64_t nek3 = k->ne[3];
-
-    const int64_t nev0 = v->ne[0];
-    const int64_t nev1 = v->ne[1];
-    //const int64_t nev2 = v->ne[2];
-    //const int64_t nev3 = v->ne[3];
-
-    const int64_t ned0 = d->ne[0];
-    const int64_t ned1 = d->ne[1];
-    //const int64_t ned2 = d->ne[2];
-    //const int64_t ned3 = d->ne[3];
-
-    const int64_t ne0  = dst->ne[0];
-    const int64_t ne1  = dst->ne[1];
-    const int64_t ne2  = dst->ne[2];
-    const int64_t ne3  = dst->ne[3];
-
-    const int nbk0 = k->nb[0];
-    const int nbk1 = k->nb[1];
-    const int nbk2 = k->nb[2];
-    const int nbk3 = k->nb[3];
-
-    const int nbq0 = q->nb[0];
-    const int nbq1 = q->nb[1];
-    const int nbq2 = q->nb[2];
-    const int nbq3 = q->nb[3];
-
-    const int nbv0 = v->nb[0];
-    const int nbv1 = v->nb[1];
-    const int nbv2 = v->nb[2];
-    const int nbv3 = v->nb[3];
-
-    const int nbd0 = d->nb[0];
-    const int nbd1 = d->nb[1];
-    const int nbd2 = d->nb[2];
-    const int nbd3 = d->nb[3];
-
-    const int nb0  = dst->nb[0];
-    const int nb1  = dst->nb[1];
-    const int nb2  = dst->nb[2];
-    const int nb3  = dst->nb[3];
+    GGML_TENSOR_LOCALS(int64_t, neq, q,   ne);
+    GGML_TENSOR_LOCALS(size_t,  nbq, q,   nb);
+    GGML_TENSOR_LOCALS(int64_t, nek, k,   ne);
+    GGML_TENSOR_LOCALS(size_t,  nbk, k,   nb);
+    GGML_TENSOR_LOCALS(int64_t, nev, v,   ne);
+    GGML_TENSOR_LOCALS(size_t,  nbv, v,   nb);
+    GGML_TENSOR_LOCALS(int64_t, ned, d,   ne);
+    GGML_TENSOR_LOCALS(size_t,  nbd, d,   nb);
+    GGML_TENSOR_LOCALS(int64_t, ne,  dst, ne);
+    GGML_TENSOR_LOCALS(size_t,  nb,  dst, nb);
 
     const int ith = params->ith;
     const int nth = params->nth;
@@ -14792,15 +14007,8 @@ static void ggml_compute_forward_win_part_f32(
         return;
     }
 
-    const int64_t ne00 = src0->ne[0]; UNUSED(ne00);
-    const int64_t ne01 = src0->ne[1];
-    const int64_t ne02 = src0->ne[2];
-    const int64_t ne03 = src0->ne[3]; UNUSED(ne03);
-
-    const int64_t ne0 = dst->ne[0];
-    const int64_t ne1 = dst->ne[1];
-    const int64_t ne2 = dst->ne[2];
-    const int64_t ne3 = dst->ne[3]; UNUSED(ne3);
+    GGML_TENSOR_LOCALS(int64_t, ne0, src0, ne);
+    GGML_TENSOR_LOCALS(int64_t, ne,  dst,  ne);
 
     const int32_t nep0 = ((const int32_t *)(opt0->data))[0];
     const int32_t nep1 = ((const int32_t *)(opt0->data))[1];
@@ -14863,14 +14071,8 @@ static void ggml_compute_forward_win_unpart_f32(
         return;
     }
 
-    const int64_t ne00 = src0->ne[0];
-    const int64_t ne01 = src0->ne[1];
-    const int64_t ne02 = src0->ne[2];
-    //const int64_t ne03 = src0->ne[3];
-
-    const int64_t ne0 = dst->ne[0];
-    const int64_t ne1 = dst->ne[1];
-    const int64_t ne2 = dst->ne[2];
+    GGML_TENSOR_LOCALS(int64_t, ne0, src0, ne);
+    GGML_TENSOR_LOCALS(int64_t, ne,  dst,  ne);
 
     const int32_t w = ((const int32_t *)(opt0->data))[0];