Change JIT generator to return uint8_t instead of xnn_status to remove dependency on xnnpack.h

PiperOrigin-RevId: 464912844

30 files changed, 120 insertions, 123 deletions

diff --git a/src/f32-gemm/1x8-aarch64-neonfma-cortex-a75.cc b/src/f32-gemm/1x8-aarch64-neonfma-cortex-a75.cc
index 8f7238959..27329459f 100644
--- a/src/f32-gemm/1x8-aarch64-neonfma-cortex-a75.cc
+++ b/src/f32-gemm/1x8-aarch64-neonfma-cortex-a75.cc
@@ -260,7 +260,7 @@ void Generator::generate(bool prefetch, size_t max_mr, size_t nc_mod_nr, size_t
 } // namespace aarch64
 } // namespace xnnpack
 
-xnn_status xnn_generate_f32_gemm_ukernel_1x8__aarch64_neonfma_cortex_a75(
+xnn_status_t xnn_generate_f32_gemm_ukernel_1x8__aarch64_neonfma_cortex_a75(
     xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params)
 {
   using namespace xnnpack::aarch64;
@@ -270,12 +270,12 @@ xnn_status xnn_generate_f32_gemm_ukernel_1x8__aarch64_neonfma_cortex_a75(
   g.generate(false, max_mr, nc_mod_nr, kc, gemm_params->f32_minmax.min, gemm_params->f32_minmax.max);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
 
-xnn_status xnn_generate_f32_gemm_ukernel_1x8__aarch64_neonfma_prfm_cortex_a75(
+xnn_status_t xnn_generate_f32_gemm_ukernel_1x8__aarch64_neonfma_prfm_cortex_a75(
     xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params)
 {
   using namespace xnnpack::aarch64;
@@ -285,7 +285,7 @@ xnn_status xnn_generate_f32_gemm_ukernel_1x8__aarch64_neonfma_prfm_cortex_a75(
   g.generate(true, max_mr, nc_mod_nr, kc, gemm_params->f32_minmax.min, gemm_params->f32_minmax.max);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
diff --git a/src/f32-gemm/4x8-aarch32-neon-cortex-a53.cc b/src/f32-gemm/4x8-aarch32-neon-cortex-a53.cc
index 28b3d2a93..c55312f21 100644
--- a/src/f32-gemm/4x8-aarch32-neon-cortex-a53.cc
+++ b/src/f32-gemm/4x8-aarch32-neon-cortex-a53.cc
@@ -544,7 +544,7 @@ void Generator::generate(size_t max_mr, size_t nc_mod_nr, size_t kc, float min,
 }  // aarch32
 }  // xnnpack
 
-xnn_status xnn_generate_f32_gemm_ukernel_4x8__aarch32_neon_cortex_a53(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
+xnn_status_t xnn_generate_f32_gemm_ukernel_4x8__aarch32_neon_cortex_a53(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
   using namespace xnnpack::aarch32;
   Generator g(code);
   assert(params != nullptr);
@@ -552,7 +552,7 @@ xnn_status xnn_generate_f32_gemm_ukernel_4x8__aarch32_neon_cortex_a53(xnn_code_b
   g.generate(max_mr, nc_mod_nr, kc, p->f32_minmax.min, p->f32_minmax.max);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
diff --git a/src/f32-gemm/4x8-aarch32-neon-cortex-a55.cc b/src/f32-gemm/4x8-aarch32-neon-cortex-a55.cc
index fd6c6a892..dc65448b2 100644
--- a/src/f32-gemm/4x8-aarch32-neon-cortex-a55.cc
+++ b/src/f32-gemm/4x8-aarch32-neon-cortex-a55.cc
@@ -436,14 +436,14 @@ void Generator::generate(size_t max_mr, size_t nc_mod_nr, size_t kc, const void*
 }  // aarch32
 }  // xnnpack
 
-xnn_status xnn_generate_f32_gemm_ukernel_4x8__aarch32_neon_cortex_a55(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
+xnn_status_t xnn_generate_f32_gemm_ukernel_4x8__aarch32_neon_cortex_a55(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
   using namespace xnnpack::aarch32;
   Generator g(code);
   assert(params != nullptr);
   g.generate(max_mr, nc_mod_nr, kc, nullptr);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
diff --git a/src/f32-gemm/4x8-aarch32-neon-cortex-a7.cc b/src/f32-gemm/4x8-aarch32-neon-cortex-a7.cc
index e7648b93b..9fb8bf161 100644
--- a/src/f32-gemm/4x8-aarch32-neon-cortex-a7.cc
+++ b/src/f32-gemm/4x8-aarch32-neon-cortex-a7.cc
@@ -253,14 +253,14 @@ void Generator::generate(size_t max_mr, size_t nc_mod_nr, size_t kc, const void*
 }  // aarch32
 }  // xnnpack
 
-xnn_status xnn_generate_f32_gemm_ukernel_4x8__aarch32_neon_cortex_a7(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
+xnn_status_t xnn_generate_f32_gemm_ukernel_4x8__aarch32_neon_cortex_a7(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
   using namespace xnnpack::aarch32;
   Generator g(code);
   assert(params != nullptr);
   g.generate(max_mr, nc_mod_nr, kc, nullptr);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
diff --git a/src/f32-gemm/4x8-aarch32-neon-cortex-a75.cc b/src/f32-gemm/4x8-aarch32-neon-cortex-a75.cc
index 9e4bf964f..aa6d1286b 100644
--- a/src/f32-gemm/4x8-aarch32-neon-cortex-a75.cc
+++ b/src/f32-gemm/4x8-aarch32-neon-cortex-a75.cc
@@ -386,26 +386,26 @@ void Generator::generate(bool prefetch, size_t max_mr, size_t nc_mod_nr, size_t
 }  // aarch32
 }  // xnnpack
 
-xnn_status xnn_generate_f32_gemm_ukernel_4x8__aarch32_neon_cortex_a75(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
+xnn_status_t xnn_generate_f32_gemm_ukernel_4x8__aarch32_neon_cortex_a75(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
   using namespace xnnpack::aarch32;
   Generator g(code);
   assert(params != nullptr);
   g.generate(false, max_mr, nc_mod_nr, kc, nullptr);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
 
-xnn_status xnn_generate_f32_gemm_ukernel_4x8__aarch32_neon_prfm_cortex_a75(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
+xnn_status_t xnn_generate_f32_gemm_ukernel_4x8__aarch32_neon_prfm_cortex_a75(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
   using namespace xnnpack::aarch32;
   Generator g(code);
   assert(params != nullptr);
   g.generate(true, max_mr, nc_mod_nr, kc, nullptr);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
diff --git a/src/f32-gemm/4x8-aarch32-neon-ld64.cc b/src/f32-gemm/4x8-aarch32-neon-ld64.cc
index ffcd5c098..7bf5f1c26 100644
--- a/src/f32-gemm/4x8-aarch32-neon-ld64.cc
+++ b/src/f32-gemm/4x8-aarch32-neon-ld64.cc
@@ -233,14 +233,14 @@ void Generator::generate(size_t max_mr, size_t nc_mod_nr, size_t kc, const void*
 }  // aarch32
 }  // xnnpack
 
-xnn_status xnn_generate_f32_gemm_ukernel_4x8__aarch32_neon_ld64(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
+xnn_status_t xnn_generate_f32_gemm_ukernel_4x8__aarch32_neon_ld64(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
   using namespace xnnpack::aarch32;
   Generator g(code);
   assert(params != nullptr);
   g.generate(max_mr, nc_mod_nr, kc, nullptr);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
diff --git a/src/f32-gemm/4x8-aarch64-neonfma-cortex-a75.cc b/src/f32-gemm/4x8-aarch64-neonfma-cortex-a75.cc
index 7f29ea3e1..e20a0c9e3 100644
--- a/src/f32-gemm/4x8-aarch64-neonfma-cortex-a75.cc
+++ b/src/f32-gemm/4x8-aarch64-neonfma-cortex-a75.cc
@@ -514,7 +514,7 @@ void Generator::generate(bool prefetch, size_t max_mr, size_t nc_mod_nr, size_t
 }  // aarch64
 }  // xnnpack
 
-xnn_status xnn_generate_f32_gemm_ukernel_4x8__aarch64_neonfma_cortex_a75(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
+xnn_status_t xnn_generate_f32_gemm_ukernel_4x8__aarch64_neonfma_cortex_a75(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
   using namespace xnnpack::aarch64;
   Generator g(code);
   assert(params != nullptr);
@@ -522,12 +522,12 @@ xnn_status xnn_generate_f32_gemm_ukernel_4x8__aarch64_neonfma_cortex_a75(xnn_cod
   g.generate(false, max_mr, nc_mod_nr, kc, gemm_params->f32_minmax.min, gemm_params->f32_minmax.max);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
 
-xnn_status xnn_generate_f32_gemm_ukernel_4x8__aarch64_neonfma_prfm_cortex_a75(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
+xnn_status_t xnn_generate_f32_gemm_ukernel_4x8__aarch64_neonfma_prfm_cortex_a75(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
   using namespace xnnpack::aarch64;
   Generator g(code);
   assert(params != nullptr);
@@ -535,7 +535,7 @@ xnn_status xnn_generate_f32_gemm_ukernel_4x8__aarch64_neonfma_prfm_cortex_a75(xn
   g.generate(true, max_mr, nc_mod_nr, kc, gemm_params->f32_minmax.min, gemm_params->f32_minmax.max);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
diff --git a/src/f32-gemm/6x8-aarch64-neonfma-ld128.cc b/src/f32-gemm/6x8-aarch64-neonfma-ld128.cc
index 1109bb20f..858d55c56 100644
--- a/src/f32-gemm/6x8-aarch64-neonfma-ld128.cc
+++ b/src/f32-gemm/6x8-aarch64-neonfma-ld128.cc
@@ -384,7 +384,7 @@ void Generator::generate(size_t max_mr, size_t nc_mod_nr, size_t kc, float min,
 }  // aarch64
 }  // xnnpack
 
-xnn_status xnn_generate_f32_gemm_ukernel_6x8__aarch64_neonfma_ld128(
+xnn_status_t xnn_generate_f32_gemm_ukernel_6x8__aarch64_neonfma_ld128(
     xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
   using namespace xnnpack::aarch64;
   Generator g(code);
@@ -393,7 +393,7 @@ xnn_status xnn_generate_f32_gemm_ukernel_6x8__aarch64_neonfma_ld128(
   g.generate(max_mr, nc_mod_nr, kc, gemm_params->f32_minmax.min, gemm_params->f32_minmax.max);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
diff --git a/src/f32-gemm/upto6x8-aarch64-neonfma-cortex-a75.cc b/src/f32-gemm/upto6x8-aarch64-neonfma-cortex-a75.cc
index 578fbda47..88b2acd38 100644
--- a/src/f32-gemm/upto6x8-aarch64-neonfma-cortex-a75.cc
+++ b/src/f32-gemm/upto6x8-aarch64-neonfma-cortex-a75.cc
@@ -1349,7 +1349,7 @@ void Generator::generate(bool prefetch, size_t max_mr, size_t nc_mod_nr, size_t
 }  // aarch64
 }  // xnnpack
 
-xnn_status xnn_generate_f32_gemm_ukernel_upto6x8__aarch64_neonfma_cortex_a75(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
+xnn_status_t xnn_generate_f32_gemm_ukernel_upto6x8__aarch64_neonfma_cortex_a75(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
   using namespace xnnpack::aarch64;
   Generator g(code);
   assert(params != nullptr);
@@ -1357,13 +1357,12 @@ xnn_status xnn_generate_f32_gemm_ukernel_upto6x8__aarch64_neonfma_cortex_a75(xnn
   g.generate(false, max_mr, nc_mod_nr, kc, gemm_params->f32_minmax.min, gemm_params->f32_minmax.max);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
-    return xnn_status_invalid_state;
+  return 0;
 }
 
-xnn_status xnn_generate_f32_gemm_ukernel_upto6x8__aarch64_neonfma_prfm_cortex_a75(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
+xnn_status_t xnn_generate_f32_gemm_ukernel_upto6x8__aarch64_neonfma_prfm_cortex_a75(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
   using namespace xnnpack::aarch64;
   Generator g(code);
   assert(params != nullptr);
@@ -1371,7 +1370,7 @@ xnn_status xnn_generate_f32_gemm_ukernel_upto6x8__aarch64_neonfma_prfm_cortex_a7
   g.generate(true, max_mr, nc_mod_nr, kc, gemm_params->f32_minmax.min, gemm_params->f32_minmax.max);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
diff --git a/src/f32-igemm/1x8-aarch64-neonfma-cortex-a75.cc b/src/f32-igemm/1x8-aarch64-neonfma-cortex-a75.cc
index c7980bd22..429909f2a 100644
--- a/src/f32-igemm/1x8-aarch64-neonfma-cortex-a75.cc
+++ b/src/f32-igemm/1x8-aarch64-neonfma-cortex-a75.cc
@@ -292,7 +292,7 @@ void Generator::generate(bool prefetch, size_t max_mr, size_t nc_mod_nr, size_t
 } // namespace aarch64
 } // namespace xnnpack
 
-xnn_status xnn_generate_f32_igemm_ukernel_1x8__aarch64_neonfma_cortex_a75(
+xnn_status_t xnn_generate_f32_igemm_ukernel_1x8__aarch64_neonfma_cortex_a75(
     xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params)
 {
   using namespace xnnpack::aarch64;
@@ -302,12 +302,12 @@ xnn_status xnn_generate_f32_igemm_ukernel_1x8__aarch64_neonfma_cortex_a75(
   g.generate(false, max_mr, nc_mod_nr, kc, ks, jit_params->f32_minmax.min, jit_params->f32_minmax.max);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
 
-xnn_status xnn_generate_f32_igemm_ukernel_1x8__aarch64_neonfma_prfm_cortex_a75(
+xnn_status_t xnn_generate_f32_igemm_ukernel_1x8__aarch64_neonfma_prfm_cortex_a75(
     xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params)
 {
   using namespace xnnpack::aarch64;
@@ -317,7 +317,7 @@ xnn_status xnn_generate_f32_igemm_ukernel_1x8__aarch64_neonfma_prfm_cortex_a75(
   g.generate(true, max_mr, nc_mod_nr, kc, ks, jit_params->f32_minmax.min, jit_params->f32_minmax.max);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
diff --git a/src/f32-igemm/4x8-aarch32-neon-cortex-a53.cc b/src/f32-igemm/4x8-aarch32-neon-cortex-a53.cc
index 387ff8f15..ea07da368 100644
--- a/src/f32-igemm/4x8-aarch32-neon-cortex-a53.cc
+++ b/src/f32-igemm/4x8-aarch32-neon-cortex-a53.cc
@@ -529,14 +529,14 @@ void Generator::generate(size_t max_mr, size_t nc_mod_nr, size_t kc, const void*
 }  // aarch32
 }  // xnnpack
 
-xnn_status xnn_generate_f32_igemm_ukernel_4x8__aarch32_neon_cortex_a53(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
+xnn_status_t xnn_generate_f32_igemm_ukernel_4x8__aarch32_neon_cortex_a53(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
   using namespace xnnpack::aarch32;
   Generator g(code);
   assert(params != nullptr);
   g.generate(max_mr, nc_mod_nr, kc, nullptr);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
diff --git a/src/f32-igemm/4x8-aarch32-neon-cortex-a55.cc b/src/f32-igemm/4x8-aarch32-neon-cortex-a55.cc
index e22eb81c2..f5e3be029 100644
--- a/src/f32-igemm/4x8-aarch32-neon-cortex-a55.cc
+++ b/src/f32-igemm/4x8-aarch32-neon-cortex-a55.cc
@@ -462,14 +462,14 @@ void Generator::generate(size_t max_mr, size_t nc_mod_nr, size_t kc, const void*
 }  // aarch32
 }  // xnnpack
 
-xnn_status xnn_generate_f32_igemm_ukernel_4x8__aarch32_neon_cortex_a55(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
+xnn_status_t xnn_generate_f32_igemm_ukernel_4x8__aarch32_neon_cortex_a55(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
   using namespace xnnpack::aarch32;
   Generator g(code);
   assert(params != nullptr);
   g.generate(max_mr, nc_mod_nr, kc, nullptr);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
diff --git a/src/f32-igemm/4x8-aarch32-neon-cortex-a7.cc b/src/f32-igemm/4x8-aarch32-neon-cortex-a7.cc
index e2448b734..60d48548e 100644
--- a/src/f32-igemm/4x8-aarch32-neon-cortex-a7.cc
+++ b/src/f32-igemm/4x8-aarch32-neon-cortex-a7.cc
@@ -281,14 +281,14 @@ void Generator::generate(size_t max_mr, size_t nc_mod_nr, size_t kc, const void*
 }  // aarch32
 }  // xnnpack
 
-xnn_status xnn_generate_f32_igemm_ukernel_4x8__aarch32_neon_cortex_a7(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
+xnn_status_t xnn_generate_f32_igemm_ukernel_4x8__aarch32_neon_cortex_a7(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
   using namespace xnnpack::aarch32;
   Generator g(code);
   assert(params != nullptr);
   g.generate(max_mr, nc_mod_nr, kc, nullptr);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
diff --git a/src/f32-igemm/4x8-aarch32-neon-cortex-a75.cc b/src/f32-igemm/4x8-aarch32-neon-cortex-a75.cc
index ed232d30c..781c38228 100644
--- a/src/f32-igemm/4x8-aarch32-neon-cortex-a75.cc
+++ b/src/f32-igemm/4x8-aarch32-neon-cortex-a75.cc
@@ -413,26 +413,26 @@ void Generator::generate(bool prefetch, size_t max_mr, size_t nc_mod_nr, size_t
 }  // aarch32
 }  // xnnpack
 
-xnn_status xnn_generate_f32_igemm_ukernel_4x8__aarch32_neon_cortex_a75(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
+xnn_status_t xnn_generate_f32_igemm_ukernel_4x8__aarch32_neon_cortex_a75(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
   using namespace xnnpack::aarch32;
   Generator g(code);
   assert(params != nullptr);
   g.generate(false, max_mr, nc_mod_nr, kc, nullptr);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
 
-xnn_status xnn_generate_f32_igemm_ukernel_4x8__aarch32_neon_prfm_cortex_a75(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
+xnn_status_t xnn_generate_f32_igemm_ukernel_4x8__aarch32_neon_prfm_cortex_a75(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
   using namespace xnnpack::aarch32;
   Generator g(code);
   assert(params != nullptr);
   g.generate(true, max_mr, nc_mod_nr, kc, nullptr);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
diff --git a/src/f32-igemm/4x8-aarch32-neon-ld64.cc b/src/f32-igemm/4x8-aarch32-neon-ld64.cc
index 6f7e91ad4..e26dc6333 100644
--- a/src/f32-igemm/4x8-aarch32-neon-ld64.cc
+++ b/src/f32-igemm/4x8-aarch32-neon-ld64.cc
@@ -260,14 +260,14 @@ void Generator::generate(size_t max_mr, size_t nc_mod_nr, size_t kc, const void*
 }  // aarch32
 }  // xnnpack
 
-xnn_status xnn_generate_f32_igemm_ukernel_4x8__aarch32_neon_ld64(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
+xnn_status_t xnn_generate_f32_igemm_ukernel_4x8__aarch32_neon_ld64(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
   using namespace xnnpack::aarch32;
   Generator g(code);
   assert(params != nullptr);
   g.generate(max_mr, nc_mod_nr, kc, nullptr);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
diff --git a/src/f32-igemm/4x8-aarch64-neonfma-cortex-a75.cc b/src/f32-igemm/4x8-aarch64-neonfma-cortex-a75.cc
index 970f2c2a1..09dfead01 100644
--- a/src/f32-igemm/4x8-aarch64-neonfma-cortex-a75.cc
+++ b/src/f32-igemm/4x8-aarch64-neonfma-cortex-a75.cc
@@ -543,7 +543,7 @@ void Generator::generate(bool prefetch, size_t max_mr, size_t nc_mod_nr, size_t
 }  // aarch64
 }  // xnnpack
 
-xnn_status xnn_generate_f32_igemm_ukernel_4x8__aarch64_neonfma_cortex_a75(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
+xnn_status_t xnn_generate_f32_igemm_ukernel_4x8__aarch64_neonfma_cortex_a75(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
   using namespace xnnpack::aarch64;
   Generator g(code);
   assert(params != nullptr);
@@ -551,12 +551,12 @@ xnn_status xnn_generate_f32_igemm_ukernel_4x8__aarch64_neonfma_cortex_a75(xnn_co
   g.generate(false, max_mr, nc_mod_nr, kc, ks, gemm_params->f32_minmax.min, gemm_params->f32_minmax.max);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
 
-xnn_status xnn_generate_f32_igemm_ukernel_4x8__aarch64_neonfma_prfm_cortex_a75(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
+xnn_status_t xnn_generate_f32_igemm_ukernel_4x8__aarch64_neonfma_prfm_cortex_a75(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
   using namespace xnnpack::aarch64;
   Generator g(code);
   assert(params != nullptr);
@@ -564,7 +564,7 @@ xnn_status xnn_generate_f32_igemm_ukernel_4x8__aarch64_neonfma_prfm_cortex_a75(x
   g.generate(true, max_mr, nc_mod_nr, kc, ks, gemm_params->f32_minmax.min, gemm_params->f32_minmax.max);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
diff --git a/src/f32-igemm/6x8-aarch64-neonfma-ld128.cc b/src/f32-igemm/6x8-aarch64-neonfma-ld128.cc
index 6cbcb69af..93884704a 100644
--- a/src/f32-igemm/6x8-aarch64-neonfma-ld128.cc
+++ b/src/f32-igemm/6x8-aarch64-neonfma-ld128.cc
@@ -414,7 +414,7 @@ void Generator::generate(size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks,
 }  // aarch64
 }  // xnnpack
 
-xnn_status xnn_generate_f32_igemm_ukernel_6x8__aarch64_neonfma_ld128(
+xnn_status_t xnn_generate_f32_igemm_ukernel_6x8__aarch64_neonfma_ld128(
     xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
   using namespace xnnpack::aarch64;
   Generator g(code);
@@ -423,7 +423,7 @@ xnn_status xnn_generate_f32_igemm_ukernel_6x8__aarch64_neonfma_ld128(
   g.generate(max_mr, nc_mod_nr, kc, ks, gemm_params->f32_minmax.min, gemm_params->f32_minmax.max);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
diff --git a/src/f32-igemm/upto6x8-aarch64-neonfma-cortex-a75.cc b/src/f32-igemm/upto6x8-aarch64-neonfma-cortex-a75.cc
index f5f6a14c3..b62ef10a0 100644
--- a/src/f32-igemm/upto6x8-aarch64-neonfma-cortex-a75.cc
+++ b/src/f32-igemm/upto6x8-aarch64-neonfma-cortex-a75.cc
@@ -1384,7 +1384,7 @@ void Generator::generate(bool prefetch, size_t max_mr, size_t nc_mod_nr, size_t
 } // namespace aarch64
 } // namespace xnnpack
 
-xnn_status xnn_generate_f32_igemm_ukernel_upto6x8__aarch64_neonfma_cortex_a75(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
+xnn_status_t xnn_generate_f32_igemm_ukernel_upto6x8__aarch64_neonfma_cortex_a75(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
   using namespace xnnpack::aarch64;
   Generator g(code);
   assert(params != nullptr);
@@ -1392,13 +1392,13 @@ xnn_status xnn_generate_f32_igemm_ukernel_upto6x8__aarch64_neonfma_cortex_a75(xn
   g.generate(false, max_mr, nc_mod_nr, kc, ks, gemm_params->f32_minmax.min, gemm_params->f32_minmax.max);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 
 }
 
-xnn_status xnn_generate_f32_igemm_ukernel_upto6x8__aarch64_neonfma_prfm_cortex_a75(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
+xnn_status_t xnn_generate_f32_igemm_ukernel_upto6x8__aarch64_neonfma_prfm_cortex_a75(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
   using namespace xnnpack::aarch64;
   Generator g(code);
   assert(params != nullptr);
@@ -1406,7 +1406,7 @@ xnn_status xnn_generate_f32_igemm_ukernel_upto6x8__aarch64_neonfma_prfm_cortex_a
   g.generate(true, max_mr, nc_mod_nr, kc, ks, gemm_params->f32_minmax.min, gemm_params->f32_minmax.max);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
diff --git a/src/qc8-gemm/4x8-fp32-aarch32-neonv8-mlal-lane-ld64.cc b/src/qc8-gemm/4x8-fp32-aarch32-neonv8-mlal-lane-ld64.cc
index 612651702..ef44e1388 100644
--- a/src/qc8-gemm/4x8-fp32-aarch32-neonv8-mlal-lane-ld64.cc
+++ b/src/qc8-gemm/4x8-fp32-aarch32-neonv8-mlal-lane-ld64.cc
@@ -472,24 +472,24 @@ void Generator::generate(bool prefetch, size_t max_mr, size_t nc_mod_nr, size_t
 }  // aarch32
 }  // xnnpack
 
-xnn_status xnn_generate_qc8_gemm_fp32_ukernel_4x8__aarch32_neonv8_mlal_lane_ld64(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
+xnn_status_t xnn_generate_qc8_gemm_fp32_ukernel_4x8__aarch32_neonv8_mlal_lane_ld64(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
   using namespace xnnpack::aarch32;
   Generator g(code);
   g.generate(false, max_mr, nc_mod_nr, kc, nullptr);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
 
-xnn_status xnn_generate_qc8_gemm_fp32_ukernel_4x8__aarch32_neonv8_mlal_lane_prfm_ld64(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
+xnn_status_t xnn_generate_qc8_gemm_fp32_ukernel_4x8__aarch32_neonv8_mlal_lane_prfm_ld64(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
   using namespace xnnpack::aarch32;
   Generator g(code);
   g.generate(true, max_mr, nc_mod_nr, kc, nullptr);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
diff --git a/src/qc8-gemm/4x8c4-fp32-aarch32-neondot-ld64.cc b/src/qc8-gemm/4x8c4-fp32-aarch32-neondot-ld64.cc
index 56acddbae..f619972a4 100644
--- a/src/qc8-gemm/4x8c4-fp32-aarch32-neondot-ld64.cc
+++ b/src/qc8-gemm/4x8c4-fp32-aarch32-neondot-ld64.cc
@@ -287,13 +287,13 @@ void Generator::generate(size_t max_mr, size_t nc_mod_nr, size_t kc, const void*
 }  // aarch32
 }  // xnnpack
 
-xnn_status xnn_generate_qc8_gemm_fp32_ukernel_4x8c4__aarch32_neondot_ld64(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
+xnn_status_t xnn_generate_qc8_gemm_fp32_ukernel_4x8c4__aarch32_neondot_ld64(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
   using namespace xnnpack::aarch32;
   Generator g(code);
   g.generate(max_mr, nc_mod_nr, kc, nullptr);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
diff --git a/src/qc8-igemm/4x8-fp32-aarch32-neonv8-mlal-lane-ld64.cc b/src/qc8-igemm/4x8-fp32-aarch32-neonv8-mlal-lane-ld64.cc
index 9c3e6b629..1301d22f8 100644
--- a/src/qc8-igemm/4x8-fp32-aarch32-neonv8-mlal-lane-ld64.cc
+++ b/src/qc8-igemm/4x8-fp32-aarch32-neonv8-mlal-lane-ld64.cc
@@ -495,24 +495,24 @@ void Generator::generate(bool prefetch, size_t max_mr, size_t nc_mod_nr, size_t
 }  // aarch32
 }  // xnnpack
 
-xnn_status xnn_generate_qc8_igemm_fp32_ukernel_4x8__aarch32_neonv8_mlal_lane_ld64(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
+xnn_status_t xnn_generate_qc8_igemm_fp32_ukernel_4x8__aarch32_neonv8_mlal_lane_ld64(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
   using namespace xnnpack::aarch32;
   Generator g(code);
   g.generate(false, max_mr, nc_mod_nr, kc, ks, nullptr);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
 
-xnn_status xnn_generate_qc8_igemm_fp32_ukernel_4x8__aarch32_neonv8_mlal_lane_prfm_ld64(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
+xnn_status_t xnn_generate_qc8_igemm_fp32_ukernel_4x8__aarch32_neonv8_mlal_lane_prfm_ld64(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
   using namespace xnnpack::aarch32;
   Generator g(code);
   g.generate(true, max_mr, nc_mod_nr, kc, ks, nullptr);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
diff --git a/src/qc8-igemm/4x8c4-fp32-aarch32-neondot-ld64.cc b/src/qc8-igemm/4x8c4-fp32-aarch32-neondot-ld64.cc
index 3e40acbf0..b51113d81 100644
--- a/src/qc8-igemm/4x8c4-fp32-aarch32-neondot-ld64.cc
+++ b/src/qc8-igemm/4x8c4-fp32-aarch32-neondot-ld64.cc
@@ -313,13 +313,13 @@ void Generator::generate(size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks,
 }  // aarch32
 }  // xnnpack
 
-xnn_status xnn_generate_qc8_igemm_fp32_ukernel_4x8c4__aarch32_neondot_ld64(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
+xnn_status_t xnn_generate_qc8_igemm_fp32_ukernel_4x8c4__aarch32_neondot_ld64(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
   using namespace xnnpack::aarch32;
   Generator g(code);
   g.generate(max_mr, nc_mod_nr, kc, ks, nullptr);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
diff --git a/src/qs8-gemm/4x8-rndnu-aarch32-neon-mlal-lane-ld64.cc b/src/qs8-gemm/4x8-rndnu-aarch32-neon-mlal-lane-ld64.cc
index 0f60ba8eb..e6b6e8e59 100644
--- a/src/qs8-gemm/4x8-rndnu-aarch32-neon-mlal-lane-ld64.cc
+++ b/src/qs8-gemm/4x8-rndnu-aarch32-neon-mlal-lane-ld64.cc
@@ -477,24 +477,24 @@ void Generator::generate(bool prefetch, size_t max_mr, size_t nc_mod_nr, size_t
 }  // aarch32
 }  // xnnpack
 
-xnn_status xnn_generate_qs8_gemm_rndnu_ukernel_4x8__aarch32_neon_mlal_lane_ld64(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
+xnn_status_t xnn_generate_qs8_gemm_rndnu_ukernel_4x8__aarch32_neon_mlal_lane_ld64(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
   using namespace xnnpack::aarch32;
   Generator g(code);
   g.generate(false, max_mr, nc_mod_nr, kc, nullptr);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
 
-xnn_status xnn_generate_qs8_gemm_rndnu_ukernel_4x8__aarch32_neon_mlal_lane_prfm_ld64(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
+xnn_status_t xnn_generate_qs8_gemm_rndnu_ukernel_4x8__aarch32_neon_mlal_lane_prfm_ld64(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
   using namespace xnnpack::aarch32;
   Generator g(code);
   g.generate(true, max_mr, nc_mod_nr, kc, nullptr);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
diff --git a/src/qs8-gemm/4x8c4-rndnu-aarch32-neondot-ld64.cc b/src/qs8-gemm/4x8c4-rndnu-aarch32-neondot-ld64.cc
index d04b23355..e12b0e546 100644
--- a/src/qs8-gemm/4x8c4-rndnu-aarch32-neondot-ld64.cc
+++ b/src/qs8-gemm/4x8c4-rndnu-aarch32-neondot-ld64.cc
@@ -292,13 +292,13 @@ void Generator::generate(size_t max_mr, size_t nc_mod_nr, size_t kc, const void*
 }  // aarch32
 }  // xnnpack
 
-xnn_status xnn_generate_qs8_gemm_rndnu_ukernel_4x8c4__aarch32_neondot_ld64(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
+xnn_status_t xnn_generate_qs8_gemm_rndnu_ukernel_4x8c4__aarch32_neondot_ld64(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void* params) {
   using namespace xnnpack::aarch32;
   Generator g(code);
   g.generate(max_mr, nc_mod_nr, kc, nullptr);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
diff --git a/src/qs8-igemm/4x8-rndnu-aarch32-neon-mlal-lane-ld64.cc b/src/qs8-igemm/4x8-rndnu-aarch32-neon-mlal-lane-ld64.cc
index 99acb710f..43e62bab1 100644
--- a/src/qs8-igemm/4x8-rndnu-aarch32-neon-mlal-lane-ld64.cc
+++ b/src/qs8-igemm/4x8-rndnu-aarch32-neon-mlal-lane-ld64.cc
@@ -500,24 +500,24 @@ void Generator::generate(bool prefetch, size_t max_mr, size_t nc_mod_nr, size_t
 }  // aarch32
 }  // xnnpack
 
-xnn_status xnn_generate_qs8_igemm_rndnu_ukernel_4x8__aarch32_neon_mlal_lane_ld64(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
+xnn_status_t xnn_generate_qs8_igemm_rndnu_ukernel_4x8__aarch32_neon_mlal_lane_ld64(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
   using namespace xnnpack::aarch32;
   Generator g(code);
   g.generate(false, max_mr, nc_mod_nr, kc, ks, nullptr);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
 
-xnn_status xnn_generate_qs8_igemm_rndnu_ukernel_4x8__aarch32_neon_mlal_lane_prfm_ld64(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
+xnn_status_t xnn_generate_qs8_igemm_rndnu_ukernel_4x8__aarch32_neon_mlal_lane_prfm_ld64(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
   using namespace xnnpack::aarch32;
   Generator g(code);
   g.generate(true, max_mr, nc_mod_nr, kc, ks, nullptr);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
diff --git a/src/qs8-igemm/4x8c4-rndnu-aarch32-neondot-ld64.cc b/src/qs8-igemm/4x8c4-rndnu-aarch32-neondot-ld64.cc
index 560fbb114..7a407c75a 100644
--- a/src/qs8-igemm/4x8c4-rndnu-aarch32-neondot-ld64.cc
+++ b/src/qs8-igemm/4x8c4-rndnu-aarch32-neondot-ld64.cc
@@ -318,13 +318,13 @@ void Generator::generate(size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks,
 }  // aarch32
 }  // xnnpack
 
-xnn_status xnn_generate_qs8_igemm_rndnu_ukernel_4x8c4__aarch32_neondot_ld64(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
+xnn_status_t xnn_generate_qs8_igemm_rndnu_ukernel_4x8c4__aarch32_neondot_ld64(xnn_code_buffer* code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void* params) {
   using namespace xnnpack::aarch32;
   Generator g(code);
   g.generate(max_mr, nc_mod_nr, kc, ks, nullptr);
   g.finalize();
   if (g.error() != xnnpack::Error::kNoError) {
-    return xnn_status_invalid_state;
+    return 3;
   }
-  return xnn_status_success;
+  return 0;
 }
diff --git a/src/xnnpack/gemm.h b/src/xnnpack/gemm.h
index a8a43a94c..0f960e925 100644
--- a/src/xnnpack/gemm.h
+++ b/src/xnnpack/gemm.h
@@ -11,8 +11,6 @@
 #include <stddef.h>
 #include <stdint.h>
 
-#include <xnnpack.h>  // For xnn_status
-
 #include <xnnpack/common.h>
 #include <xnnpack/microparams.h>
 
@@ -1918,7 +1916,7 @@ DECLARE_QC8_GEMM_MINMAX_UKERNEL_FUNCTION(xnn_qc8_gemm_minmax_fp32_ukernel_4x2__s
 DECLARE_QC8_GEMM_MINMAX_UKERNEL_FUNCTION(xnn_qc8_gemm_minmax_fp32_ukernel_4x4__scalar_lrintf)
 
 #define DECLARE_GENERATE_GEMM_UKERNEL_FUNCTION(fn_name) \
-  XNN_INTERNAL enum xnn_status fn_name(                 \
+  XNN_INTERNAL xnn_status_t fn_name(                    \
       struct xnn_code_buffer* code,                     \
       size_t max_mr,                                    \
       size_t nc_mod_nr,                                 \
diff --git a/src/xnnpack/igemm.h b/src/xnnpack/igemm.h
index 8ef6bb130..55fe92394 100644
--- a/src/xnnpack/igemm.h
+++ b/src/xnnpack/igemm.h
@@ -11,8 +11,6 @@
 #include <stddef.h>
 #include <stdint.h>
 
-#include <xnnpack.h>  // For xnn_status
-
 #include <xnnpack/common.h>
 #include <xnnpack/microparams.h>
 
@@ -1626,7 +1624,7 @@ DECLARE_QC8_IGEMM_MINMAX_UKERNEL_FUNCTION(xnn_qc8_igemm_minmax_fp32_ukernel_4x2_
 DECLARE_QC8_IGEMM_MINMAX_UKERNEL_FUNCTION(xnn_qc8_igemm_minmax_fp32_ukernel_4x4__scalar_lrintf)
 
 #define DECLARE_GENERATE_IGEMM_UKERNEL_FUNCTION(fn_name) \
-  XNN_INTERNAL enum xnn_status fn_name(                  \
+  XNN_INTERNAL xnn_status_t fn_name(                     \
       struct xnn_code_buffer* code,                      \
       size_t max_mr,                                     \
       size_t nc_mod_nr,                                  \
diff --git a/src/xnnpack/microparams.h b/src/xnnpack/microparams.h
index 7b291f414..f33f8b66f 100644
--- a/src/xnnpack/microparams.h
+++ b/src/xnnpack/microparams.h
@@ -2479,3 +2479,5 @@ struct jit_gemm_params {
     float max;
   } f32_minmax;
 };
+
+typedef int xnn_status_t;
diff --git a/src/xnnpack/params.h b/src/xnnpack/params.h
index 37352a1f5..abe7ff49f 100644
--- a/src/xnnpack/params.h
+++ b/src/xnnpack/params.h
@@ -1885,9 +1885,9 @@ typedef void (*xnn_init_qc8_scale_params_fn)(
   const float scale[XNN_MIN_ELEMENTS(1)],
   void* packed_w);
 
-typedef enum xnn_status (*xnn_jit_gemm_code_generator_function)(
+typedef xnn_status_t (*xnn_jit_gemm_code_generator_function)(
     struct xnn_code_buffer *code, size_t max_mr, size_t nc_mod_nr, size_t kc, const void *params);
-typedef enum xnn_status (*xnn_jit_igemm_code_generator_function)(
+typedef xnn_status_t (*xnn_jit_igemm_code_generator_function)(
     struct xnn_code_buffer *code, size_t max_mr, size_t nc_mod_nr, size_t kc, size_t ks, const void *params);
 
 struct xnn_hmp_gemm_ukernel {