Disable neondot microkernels on iOS 32 bit

- select -mcpu=cyclone which is ARMv8 and supports AArch32.
- Disable dot product source code with #ifndef __APPLE__

PiperOrigin-RevId: 426318242

13 files changed, 40 insertions, 0 deletions

diff --git a/BUILD.bazel b/BUILD.bazel
index b27844dd8..33d191ee1 100644
--- a/BUILD.bazel
+++ b/BUILD.bazel
@@ -8711,6 +8711,10 @@ xnnpack_cc_library(
     aarch32_srcs = AARCH32_ASM_MICROKERNEL_SRCS,
     aarch64_copts = ["-march=armv8.2-a+fp16+dotprod"],
     aarch64_srcs = AARCH64_ASM_MICROKERNEL_SRCS,
+    apple_aarch32_copts = [
+        "-mcpu=cyclone",
+        "-mtune=generic",
+    ],
     wasm_srcs = WASM32_ASM_MICROKERNEL_SRCS,
     wasmrelaxedsimd_srcs = WASM32_ASM_MICROKERNEL_SRCS,
     wasmsimd_srcs = WASM32_ASM_MICROKERNEL_SRCS,
diff --git a/src/qc8-gemm/gen/4x8c4-minmax-fp32-aarch32-neondot-cortex-a55.S b/src/qc8-gemm/gen/4x8c4-minmax-fp32-aarch32-neondot-cortex-a55.S
index 0a7864899..f1ef0fb03 100644
--- a/src/qc8-gemm/gen/4x8c4-minmax-fp32-aarch32-neondot-cortex-a55.S
+++ b/src/qc8-gemm/gen/4x8c4-minmax-fp32-aarch32-neondot-cortex-a55.S
@@ -48,6 +48,8 @@
 //    int8_t output_max;          d13[7]
 //  } xnn_qs8_minmax_params.neonv8;
 
+// iOS does not support 32 bit ARM with Neon DotProduct.
+#ifndef __APPLE__
 BEGIN_FUNCTION xnn_qc8_gemm_minmax_fp32_ukernel_4x8c4__aarch32_neondot_cortex_a55
         # Push 80 bytes
         PUSH    {r4, r5, r6, r7, r8, r9, r10, r11}  // 32
@@ -315,6 +317,7 @@ BEGIN_FUNCTION xnn_qc8_gemm_minmax_fp32_ukernel_4x8c4__aarch32_neondot_cortex_a5
         BX      lr
 
 END_FUNCTION xnn_qc8_gemm_minmax_fp32_ukernel_4x8c4__aarch32_neondot_cortex_a55
+#endif  // __APPLE__
 
 #ifdef __ELF__
 .section ".note.GNU-stack","",%progbits
diff --git a/src/qc8-gemm/gen/4x8c4-minmax-fp32-aarch32-neondot-ld64.S b/src/qc8-gemm/gen/4x8c4-minmax-fp32-aarch32-neondot-ld64.S
index ad9af11b2..ae0f58014 100644
--- a/src/qc8-gemm/gen/4x8c4-minmax-fp32-aarch32-neondot-ld64.S
+++ b/src/qc8-gemm/gen/4x8c4-minmax-fp32-aarch32-neondot-ld64.S
@@ -48,6 +48,8 @@
 //    int8_t output_max;          d13[7]
 //  } xnn_qs8_minmax_params.neonv8;
 
+// iOS does not support 32 bit ARM with Neon DotProduct.
+#ifndef __APPLE__
 BEGIN_FUNCTION xnn_qc8_gemm_minmax_fp32_ukernel_4x8c4__aarch32_neondot_ld64
         # Push 80 bytes
         PUSH    {r4, r5, r6, r7, r8, r9, r10, r11}  // 32
@@ -268,6 +270,7 @@ BEGIN_FUNCTION xnn_qc8_gemm_minmax_fp32_ukernel_4x8c4__aarch32_neondot_ld64
         BX      lr
 
 END_FUNCTION xnn_qc8_gemm_minmax_fp32_ukernel_4x8c4__aarch32_neondot_ld64
+#endif  // __APPLE__
 
 #ifdef __ELF__
 .section ".note.GNU-stack","",%progbits
diff --git a/src/qc8-igemm/gen/4x8c4-minmax-fp32-aarch32-neondot-cortex-a55.S b/src/qc8-igemm/gen/4x8c4-minmax-fp32-aarch32-neondot-cortex-a55.S
index 17bf66564..c3605bdd5 100644
--- a/src/qc8-igemm/gen/4x8c4-minmax-fp32-aarch32-neondot-cortex-a55.S
+++ b/src/qc8-igemm/gen/4x8c4-minmax-fp32-aarch32-neondot-cortex-a55.S
@@ -50,6 +50,8 @@
 //    int8_t output_max;          d13[7]
 //  } xnn_qs8_minmax_params.neonv8;
 
+// iOS does not support 32 bit ARM with Neon DotProduct.
+#ifndef __APPLE__
 BEGIN_FUNCTION xnn_qc8_igemm_minmax_fp32_ukernel_4x8c4__aarch32_neondot_cortex_a55
         ADD     r2, r2, 3               // kc = (kc + 3) & ~3
         BIC     r2, r2, 3
@@ -339,6 +341,7 @@ BEGIN_FUNCTION xnn_qc8_igemm_minmax_fp32_ukernel_4x8c4__aarch32_neondot_cortex_a
         POP     {r4, r5, r6, r7, r8, r9, r10, r11, pc}
 
 END_FUNCTION xnn_qc8_igemm_minmax_fp32_ukernel_4x8c4__aarch32_neondot_cortex_a55
+#endif  // __APPLE__
 
 #ifdef __ELF__
 .section ".note.GNU-stack","",%progbits
diff --git a/src/qc8-igemm/gen/4x8c4-minmax-fp32-aarch32-neondot-ld64.S b/src/qc8-igemm/gen/4x8c4-minmax-fp32-aarch32-neondot-ld64.S
index 9c1d9e310..c6f402513 100644
--- a/src/qc8-igemm/gen/4x8c4-minmax-fp32-aarch32-neondot-ld64.S
+++ b/src/qc8-igemm/gen/4x8c4-minmax-fp32-aarch32-neondot-ld64.S
@@ -50,6 +50,8 @@
 //    int8_t output_max;          d13[7]
 //  } xnn_qs8_minmax_params.neonv8;
 
+// iOS does not support 32 bit ARM with Neon DotProduct.
+#ifndef __APPLE__
 BEGIN_FUNCTION xnn_qc8_igemm_minmax_fp32_ukernel_4x8c4__aarch32_neondot_ld64
         ADD     r2, r2, 3               // kc = (kc + 3) & ~3
         BIC     r2, r2, 3
@@ -292,6 +294,7 @@ BEGIN_FUNCTION xnn_qc8_igemm_minmax_fp32_ukernel_4x8c4__aarch32_neondot_ld64
         POP     {r4, r5, r6, r7, r8, r9, r10, r11, pc}
 
 END_FUNCTION xnn_qc8_igemm_minmax_fp32_ukernel_4x8c4__aarch32_neondot_ld64
+#endif  // __APPLE__
 
 #ifdef __ELF__
 .section ".note.GNU-stack","",%progbits
diff --git a/src/qs8-gemm/4x8c4-aarch32-neondot-cortex-a55.S.in b/src/qs8-gemm/4x8c4-aarch32-neondot-cortex-a55.S.in
index b3f3b6e77..429772167 100644
--- a/src/qs8-gemm/4x8c4-aarch32-neondot-cortex-a55.S.in
+++ b/src/qs8-gemm/4x8c4-aarch32-neondot-cortex-a55.S.in
@@ -60,6 +60,8 @@ $else:
   //    int8_t output_max;          d13[7]
   //  } xnn_qs8_minmax_params.neonv8;
 
+// iOS does not support 32 bit ARM with Neon DotProduct.
+#ifndef __APPLE__
 BEGIN_FUNCTION xnn_${DATATYPE.lower()}_gemm_minmax_${REQUANTIZATION.lower()}_ukernel_4x8c4__aarch32_neondot_cortex_a55
         # Push 80 bytes
         PUSH    {r4, r5, r6, r7, r8, r9, r10, r11}  // 32
@@ -363,6 +365,7 @@ BEGIN_FUNCTION xnn_${DATATYPE.lower()}_gemm_minmax_${REQUANTIZATION.lower()}_uke
         BX      lr
 
 END_FUNCTION xnn_${DATATYPE.lower()}_gemm_minmax_${REQUANTIZATION.lower()}_ukernel_4x8c4__aarch32_neondot_cortex_a55
+#endif  // __APPLE__
 
 #ifdef __ELF__
 .section ".note.GNU-stack","",%progbits
diff --git a/src/qs8-gemm/4x8c4-aarch32-neondot-ld64.S.in b/src/qs8-gemm/4x8c4-aarch32-neondot-ld64.S.in
index b88cddfca..98342e67d 100644
--- a/src/qs8-gemm/4x8c4-aarch32-neondot-ld64.S.in
+++ b/src/qs8-gemm/4x8c4-aarch32-neondot-ld64.S.in
@@ -60,6 +60,8 @@ $else:
   //    int8_t output_max;          d13[7]
   //  } xnn_qs8_minmax_params.neonv8;
 
+// iOS does not support 32 bit ARM with Neon DotProduct.
+#ifndef __APPLE__
 BEGIN_FUNCTION xnn_${DATATYPE.lower()}_gemm_minmax_${REQUANTIZATION.lower()}_ukernel_4x8c4__aarch32_neondot_ld64
         # Push 80 bytes
         PUSH    {r4, r5, r6, r7, r8, r9, r10, r11}  // 32
@@ -316,6 +318,7 @@ BEGIN_FUNCTION xnn_${DATATYPE.lower()}_gemm_minmax_${REQUANTIZATION.lower()}_uke
         BX      lr
 
 END_FUNCTION xnn_${DATATYPE.lower()}_gemm_minmax_${REQUANTIZATION.lower()}_ukernel_4x8c4__aarch32_neondot_ld64
+#endif  // __APPLE__
 
 #ifdef __ELF__
 .section ".note.GNU-stack","",%progbits
diff --git a/src/qs8-gemm/gen/4x8c4-minmax-rndnu-aarch32-neondot-cortex-a55.S b/src/qs8-gemm/gen/4x8c4-minmax-rndnu-aarch32-neondot-cortex-a55.S
index 8629e37c2..504cbab79 100644
--- a/src/qs8-gemm/gen/4x8c4-minmax-rndnu-aarch32-neondot-cortex-a55.S
+++ b/src/qs8-gemm/gen/4x8c4-minmax-rndnu-aarch32-neondot-cortex-a55.S
@@ -51,6 +51,8 @@
 //    int8_t output_max;          d13[7]
 //  } rndnu_neon;
 
+// iOS does not support 32 bit ARM with Neon DotProduct.
+#ifndef __APPLE__
 BEGIN_FUNCTION xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4__aarch32_neondot_cortex_a55
         # Push 80 bytes
         PUSH    {r4, r5, r6, r7, r8, r9, r10, r11}  // 32
@@ -320,6 +322,7 @@ BEGIN_FUNCTION xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4__aarch32_neondot_cortex_a
         BX      lr
 
 END_FUNCTION xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4__aarch32_neondot_cortex_a55
+#endif  // __APPLE__
 
 #ifdef __ELF__
 .section ".note.GNU-stack","",%progbits
diff --git a/src/qs8-gemm/gen/4x8c4-minmax-rndnu-aarch32-neondot-ld64.S b/src/qs8-gemm/gen/4x8c4-minmax-rndnu-aarch32-neondot-ld64.S
index 94bad0879..9198a9d90 100644
--- a/src/qs8-gemm/gen/4x8c4-minmax-rndnu-aarch32-neondot-ld64.S
+++ b/src/qs8-gemm/gen/4x8c4-minmax-rndnu-aarch32-neondot-ld64.S
@@ -51,6 +51,8 @@
 //    int8_t output_max;          d13[7]
 //  } rndnu_neon;
 
+// iOS does not support 32 bit ARM with Neon DotProduct.
+#ifndef __APPLE__
 BEGIN_FUNCTION xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4__aarch32_neondot_ld64
         # Push 80 bytes
         PUSH    {r4, r5, r6, r7, r8, r9, r10, r11}  // 32
@@ -273,6 +275,7 @@ BEGIN_FUNCTION xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4__aarch32_neondot_ld64
         BX      lr
 
 END_FUNCTION xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4__aarch32_neondot_ld64
+#endif  // __APPLE__
 
 #ifdef __ELF__
 .section ".note.GNU-stack","",%progbits
diff --git a/src/qs8-igemm/4x8c4-aarch32-neondot-cortex-a55.S.in b/src/qs8-igemm/4x8c4-aarch32-neondot-cortex-a55.S.in
index 58f2ab37d..f6470f05d 100644
--- a/src/qs8-igemm/4x8c4-aarch32-neondot-cortex-a55.S.in
+++ b/src/qs8-igemm/4x8c4-aarch32-neondot-cortex-a55.S.in
@@ -62,6 +62,8 @@ $else:
   //    int8_t output_max;          d13[7]
   //  } xnn_qs8_minmax_params.neonv8;
 
+// iOS does not support 32 bit ARM with Neon DotProduct.
+#ifndef __APPLE__
 BEGIN_FUNCTION xnn_${DATATYPE.lower()}_igemm_minmax_${REQUANTIZATION.lower()}_ukernel_4x8c4__aarch32_neondot_cortex_a55
         ADD     r2, r2, 3               // kc = (kc + 3) & ~3
         BIC     r2, r2, 3
@@ -387,6 +389,7 @@ BEGIN_FUNCTION xnn_${DATATYPE.lower()}_igemm_minmax_${REQUANTIZATION.lower()}_uk
         POP     {r4, r5, r6, r7, r8, r9, r10, r11, pc}
 
 END_FUNCTION xnn_${DATATYPE.lower()}_igemm_minmax_${REQUANTIZATION.lower()}_ukernel_4x8c4__aarch32_neondot_cortex_a55
+#endif  // __APPLE__
 
 #ifdef __ELF__
 .section ".note.GNU-stack","",%progbits
diff --git a/src/qs8-igemm/4x8c4-aarch32-neondot-ld64.S.in b/src/qs8-igemm/4x8c4-aarch32-neondot-ld64.S.in
index 5361caf6b..1889e1a9f 100644
--- a/src/qs8-igemm/4x8c4-aarch32-neondot-ld64.S.in
+++ b/src/qs8-igemm/4x8c4-aarch32-neondot-ld64.S.in
@@ -62,6 +62,8 @@ $else:
   //    int8_t output_max;          d13[7]
   //  } xnn_qs8_minmax_params.neonv8;
 
+// iOS does not support 32 bit ARM with Neon DotProduct.
+#ifndef __APPLE__
 BEGIN_FUNCTION xnn_${DATATYPE.lower()}_igemm_minmax_${REQUANTIZATION.lower()}_ukernel_4x8c4__aarch32_neondot_ld64
         ADD     r2, r2, 3               // kc = (kc + 3) & ~3
         BIC     r2, r2, 3
@@ -340,6 +342,7 @@ BEGIN_FUNCTION xnn_${DATATYPE.lower()}_igemm_minmax_${REQUANTIZATION.lower()}_uk
         POP     {r4, r5, r6, r7, r8, r9, r10, r11, pc}
 
 END_FUNCTION xnn_${DATATYPE.lower()}_igemm_minmax_${REQUANTIZATION.lower()}_ukernel_4x8c4__aarch32_neondot_ld64
+#endif  // __APPLE__
 
 #ifdef __ELF__
 .section ".note.GNU-stack","",%progbits
diff --git a/src/qs8-igemm/gen/4x8c4-minmax-rndnu-aarch32-neondot-cortex-a55.S b/src/qs8-igemm/gen/4x8c4-minmax-rndnu-aarch32-neondot-cortex-a55.S
index 9de362cf6..a0f12af78 100644
--- a/src/qs8-igemm/gen/4x8c4-minmax-rndnu-aarch32-neondot-cortex-a55.S
+++ b/src/qs8-igemm/gen/4x8c4-minmax-rndnu-aarch32-neondot-cortex-a55.S
@@ -53,6 +53,8 @@
 //    int8_t output_max;          d13[7]
 //  } rndnu_neon;
 
+// iOS does not support 32 bit ARM with Neon DotProduct.
+#ifndef __APPLE__
 BEGIN_FUNCTION xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4__aarch32_neondot_cortex_a55
         ADD     r2, r2, 3               // kc = (kc + 3) & ~3
         BIC     r2, r2, 3
@@ -344,6 +346,7 @@ BEGIN_FUNCTION xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4__aarch32_neondot_cortex_
         POP     {r4, r5, r6, r7, r8, r9, r10, r11, pc}
 
 END_FUNCTION xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4__aarch32_neondot_cortex_a55
+#endif  // __APPLE__
 
 #ifdef __ELF__
 .section ".note.GNU-stack","",%progbits
diff --git a/src/qs8-igemm/gen/4x8c4-minmax-rndnu-aarch32-neondot-ld64.S b/src/qs8-igemm/gen/4x8c4-minmax-rndnu-aarch32-neondot-ld64.S
index a28f00e7d..d6e06bf9e 100644
--- a/src/qs8-igemm/gen/4x8c4-minmax-rndnu-aarch32-neondot-ld64.S
+++ b/src/qs8-igemm/gen/4x8c4-minmax-rndnu-aarch32-neondot-ld64.S
@@ -53,6 +53,8 @@
 //    int8_t output_max;          d13[7]
 //  } rndnu_neon;
 
+// iOS does not support 32 bit ARM with Neon DotProduct.
+#ifndef __APPLE__
 BEGIN_FUNCTION xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4__aarch32_neondot_ld64
         ADD     r2, r2, 3               // kc = (kc + 3) & ~3
         BIC     r2, r2, 3
@@ -297,6 +299,7 @@ BEGIN_FUNCTION xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4__aarch32_neondot_ld64
         POP     {r4, r5, r6, r7, r8, r9, r10, r11, pc}
 
 END_FUNCTION xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4__aarch32_neondot_ld64
+#endif  // __APPLE__
 
 #ifdef __ELF__
 .section ".note.GNU-stack","",%progbits