Fixes for build failures seen if Kleidi is disabled

Author: Jayakrishna T N

src/pf32-gemm/gen/pf32-gemm-32x32-minmax-asm-aarch64-neonsme.S

@@ -6,7 +6,7 @@
 // This source code is licensed under the BSD-style license found in the
 // LICENSE file in the root directory of this source tree.
 
-#include "xnnpack/assembly.h"
+#include "src/xnnpack/assembly.h"
 
 #define fpack_lhs_packed_in      x0
 #define fpack_rhs_packed_in      x1
@@ -551,4 +551,4 @@ END_FUNCTION xnn_pf32_gemm_minmax__asm_aarch64_neonsme
 
 #ifdef __ELF__
 .section ".note.GNU-stack","",%progbits
-#endif
+#endif

src/pf32-gemm/gen/pf32-gemm-minmax-utils.h

@@ -7,7 +7,7 @@
 // LICENSE file in the root directory of this source tree.
 
 #include <stdint.h>
-#include "xnnpack/common.h"
+#include "src/xnnpack/common.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -47,4 +47,4 @@ XNN_INTERNAL void xnn_pf32_gemm_minmax__asm_aarch64_neonsme(
 
 #ifdef __cplusplus
 }  // extern "C"
-#endif
+#endif

src/pf32-gemm/gen/pf32-gemm-pack-utils-asm-aarch64-neonsme.S

@@ -6,7 +6,7 @@
 // This source code is licensed under the BSD-style license found in the
 // LICENSE file in the root directory of this source tree.
  
-#include "xnnpack/assembly.h"
+#include "src/xnnpack/assembly.h"
  
 #define fpack_rhs_head_rows                    4
 #define fpack_rhs_tail_rows                    1
@@ -408,4 +408,4 @@ END_FUNCTION xnn_pf32_pack_rhs__asm_aarch64_neonsme
 
 #ifdef __ELF__
 .section ".note.GNU-stack","",%progbits
-#endif
+#endif

src/pf32-gemm/pf32-gemm-32x32-minmax-neonsme.c

@@ -7,8 +7,8 @@
 // LICENSE file in the root directory of this source tree.
 
 #include <stddef.h>
-#include "xnnpack/microparams.h"
-#include "pf32-gemm/gen/pf32-gemm-minmax-utils.h"
+#include "src/xnnpack/microparams.h"
+#include "src/pf32-gemm/gen/pf32-gemm-minmax-utils.h"
 
 size_t xnn_pf32_gemm_minmax_ukernel_32x32__neonsme_get_mr() {
     return xnn_pf32_gemm_minmax_mr * xnn_pf32_get_word_sme_vl__asm_aarch64_neonsme();
@@ -24,7 +24,7 @@ void xnn_pf32_gemm_minmax_ukernel_32x32__neonsme(
     size_t m, size_t n, size_t k, const void* lhs_packed,
     const void* rhs_packed, float* dst, size_t dst_stride_row,
     size_t dst_stride_col,
-    union xnn_f32_minmax_params
+    struct xnn_f32_minmax_params
         minmax_params[XNN_RESTRICT XNN_MIN_ELEMENTS(1)]) {
 
     xnn_pf32_gemm_minmax__asm_aarch64_neonsme(lhs_packed, rhs_packed, dst, (k/sizeof(float)), &minmax_params->scalar.max,

src/reference/packing.cc

@@ -1880,6 +1880,15 @@ void xnn_pack_qu8_weights_and_biases(
       packed_weights_ptr, extra_bytes, params);
 }
 
+void transpose_weights(const float* in, float* out, size_t height,
+                       size_t width) {
+  for (size_t i = 0; i < height; ++i) {
+    for (size_t j = 0; j < width; ++j) {
+      out[j * height + i] = in[i * width + j];
+    }
+  }
+}
+
 #if XNN_ENABLE_KLEIDIAI
 size_t xnn_packed_stride_kai_qs4_weights_and_biases_sme(
     const struct xnn_gemm_config* gemm_config, size_t k, size_t unused_k_stride,
@@ -2357,15 +2366,6 @@ void xnn_pack_kai_qb4_weights_and_biases(
 }
 #endif  // XNN_ENABLE_KLEIDIAI
 
-void transpose_weights(const float* in, float* out, size_t height,
-                       size_t width) {
-  for (size_t i = 0; i < height; ++i) {
-    for (size_t j = 0; j < width; ++j) {
-      out[j * height + i] = in[i * width + j];
-    }
-  }
-}
-
 void xnn_pack_f32_run_pack_rhs(size_t num_groups, size_t n, size_t k, size_t nr, size_t kr,
                   size_t sr, size_t rhs_stride, const void *rhs,
                   const void *bias, const void *scale, void *rhs_packed,

src/x32-pack-lh/x32-packlh-neonsme.c

@@ -14,10 +14,10 @@
 #include <stdint.h>
 #include <stdlib.h>
 
-#include "xnnpack/common.h"
-#include "xnnpack/math.h"
-#include "xnnpack/pack-lh.h"
-#include "pf32-gemm/gen/pf32-gemm-minmax-utils.h"
+#include "src/xnnpack/common.h"
+#include "src/xnnpack/math.h"
+#include "src/xnnpack/pack-lh.h"
+#include "src/pf32-gemm/gen/pf32-gemm-minmax-utils.h"
 
 #define GET_MIN(a, b) ((a) < (b) ? (a) : (b))
 #define roundup(x, y)   ((((x) + ((y) - 1)) / (y)) * (y))
@@ -87,4 +87,4 @@ size_t xnn_x32_pack_lhs_offset__neonsme(size_t m, size_t k, size_t mr, size_t kr
   (void)sr;
 
   return m * k;
-}
+}