Add MINMAX suffix to GEMM/IGEMM/DWCONV/PPMM micro-kernel names

PiperOrigin-RevId: 305632703
diff --git a/src/xnnpack/dwconv.h b/src/xnnpack/dwconv.h
index 668e038..fd99ef6 100644
--- a/src/xnnpack/dwconv.h
+++ b/src/xnnpack/dwconv.h
@@ -18,7 +18,7 @@
 extern "C" {
 #endif
 
-#define DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(fn_name) \
+#define DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(fn_name) \
   XNN_INTERNAL void fn_name(                                 \
     size_t channels,                                         \
     size_t output_width,                                     \
@@ -29,105 +29,105 @@
     size_t output_increment,                                 \
     const union xnn_f32_minmax_params* params);
 
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up4x4__psimd)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up4x4__psimd_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up8x4__psimd)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up8x4__psimd_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up4x4__sse)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up4x4__sse_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up8x4__sse)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up8x4__sse_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up8x4__avx)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up8x4__avx_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up16x4__avx)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up16x4__avx_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up8x4__fma3)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up8x4__fma3_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up16x4__fma3)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up16x4__fma3_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up16x4__avx512f)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up16x4__avx512f_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up32x4__avx512f)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up32x4__avx512f_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up1x4__wasm)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up1x4__wasm_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up2x4__wasm)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up2x4__wasm_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up1x4__scalar)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up1x4__scalar_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up2x4__scalar)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up2x4__scalar_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up4x4__psimd)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up4x4__psimd_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up8x4__psimd)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up8x4__psimd_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up4x4__sse)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up4x4__sse_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up8x4__sse)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up8x4__sse_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up8x4__avx)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up8x4__avx_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up16x4__avx)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up16x4__avx_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up8x4__fma3)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up8x4__fma3_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up16x4__fma3)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up16x4__fma3_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up16x4__avx512f)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up16x4__avx512f_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up32x4__avx512f)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up32x4__avx512f_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up1x4__wasm)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up1x4__wasm_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up2x4__wasm)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up2x4__wasm_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up1x4__scalar)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up1x4__scalar_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up2x4__scalar)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up2x4__scalar_acc2)
 
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up4x9__neon)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up4x9__neon_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up8x9__neon)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up8x9__neon_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up4x9__neonfma)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up4x9__neonfma_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up8x9__neonfma)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up8x9__neonfma_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up4x9__aarch64_neonfma)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up4x9__aarch64_neonfma_cortex_a55)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up4x9__psimd)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up4x9__psimd_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up8x9__psimd)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up8x9__psimd_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up4x9__sse)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up4x9__sse_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up8x9__sse)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up8x9__sse_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up8x9__avx)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up8x9__avx_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up16x9__avx)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up16x9__avx_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up8x9__fma3)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up8x9__fma3_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up16x9__fma3)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up16x9__fma3_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up16x9__avx512f)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up16x9__avx512f_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up32x9__avx512f)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up32x9__avx512f_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up1x9__wasm)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up1x9__wasm_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up2x9__wasm)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up2x9__wasm_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up1x9__scalar)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up1x9__scalar_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up2x9__scalar)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up2x9__scalar_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up4x9__neon)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up4x9__neon_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up8x9__neon)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up8x9__neon_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up4x9__neonfma)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up4x9__neonfma_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up8x9__neonfma)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up8x9__neonfma_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up4x9__aarch64_neonfma)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up4x9__aarch64_neonfma_cortex_a55)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up4x9__psimd)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up4x9__psimd_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up8x9__psimd)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up8x9__psimd_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up4x9__sse)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up4x9__sse_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up8x9__sse)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up8x9__sse_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up8x9__avx)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up8x9__avx_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up16x9__avx)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up16x9__avx_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up8x9__fma3)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up8x9__fma3_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up16x9__fma3)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up16x9__fma3_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up16x9__avx512f)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up16x9__avx512f_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up32x9__avx512f)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up32x9__avx512f_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up1x9__wasm)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up1x9__wasm_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up2x9__wasm)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up2x9__wasm_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up1x9__scalar)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up1x9__scalar_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up2x9__scalar)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up2x9__scalar_acc2)
 
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up4x25__psimd)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up4x25__psimd_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up8x25__psimd)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up8x25__psimd_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up4x25__sse)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up4x25__sse_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up8x25__sse)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up8x25__sse_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up8x25__avx)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up8x25__avx_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up16x25__avx)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up16x25__avx_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up8x25__fma3)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up8x25__fma3_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up16x25__fma3)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up16x25__fma3_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up16x25__avx512f)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up16x25__avx512f_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up32x25__avx512f)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up32x25__avx512f_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up1x25__wasm)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up1x25__wasm_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up2x25__wasm)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up2x25__wasm_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up1x25__scalar)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up1x25__scalar_acc2)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up2x25__scalar)
-DECLARE_F32_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_ukernel_up2x25__scalar_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up4x25__psimd)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up4x25__psimd_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up8x25__psimd)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up8x25__psimd_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up4x25__sse)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up4x25__sse_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up8x25__sse)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up8x25__sse_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up8x25__avx)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up8x25__avx_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up16x25__avx)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up16x25__avx_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up8x25__fma3)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up8x25__fma3_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up16x25__fma3)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up16x25__fma3_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up16x25__avx512f)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up16x25__avx512f_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up32x25__avx512f)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up32x25__avx512f_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up1x25__wasm)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up1x25__wasm_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up2x25__wasm)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up2x25__wasm_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up1x25__scalar)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up1x25__scalar_acc2)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up2x25__scalar)
+DECLARE_F32_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_f32_dwconv_minmax_ukernel_up2x25__scalar_acc2)
 
 
-#define DECLARE_Q8_DWCONV_UNIPASS_UKERNEL_FUNCTION(fn_name) \
+#define DECLARE_Q8_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(fn_name) \
   XNN_INTERNAL void fn_name(                                \
     size_t channels,                                        \
     size_t output_width,                                    \
@@ -138,10 +138,10 @@
     size_t output_increment,                                \
     const union xnn_q8_gemm_params* params);
 
-DECLARE_Q8_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_q8_dwconv_ukernel_up1x9__scalar)
-DECLARE_Q8_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_q8_dwconv_ukernel_up8x9__aarch32_neon)
-DECLARE_Q8_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_q8_dwconv_ukernel_up8x9__neon)
-DECLARE_Q8_DWCONV_UNIPASS_UKERNEL_FUNCTION(xnn_q8_dwconv_ukernel_up8x9__sse2)
+DECLARE_Q8_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_q8_dwconv_minmax_ukernel_up1x9__scalar)
+DECLARE_Q8_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_q8_dwconv_minmax_ukernel_up8x9__aarch32_neon)
+DECLARE_Q8_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_q8_dwconv_minmax_ukernel_up8x9__neon)
+DECLARE_Q8_DWCONV_MINMAX_UNIPASS_UKERNEL_FUNCTION(xnn_q8_dwconv_minmax_ukernel_up8x9__sse2)
 
 
 #define DECLARE_F32_DWCONV_SPCHW_UKERNEL_FUNCTION(fn_name) \