[FFmpeg-devel] [PATCH 01/14] swscale/range_convert: call arch-specific init functions from main init function

[Ramiro Polla]

This commit also fixes the issue that the call to ff_sws_init_range_convert()
from sws_init_swscale() was not setting up the arch-specific optimizations.
---
 libswscale/aarch64/swscale.c                  |  5 ++++-
 libswscale/loongarch/swscale_init_loongarch.c |  1 -
 libswscale/riscv/swscale.c                    |  7 ++++---
 libswscale/swscale.c                          | 10 ++++++++++
 libswscale/swscale_internal.h                 |  1 +
 libswscale/utils.c                            | 10 +---------
 libswscale/x86/swscale.c                      |  2 --
 7 files changed, 20 insertions(+), 16 deletions(-)

diff --git a/libswscale/aarch64/swscale.c b/libswscale/aarch64/swscale.c
index eb907284e7..863627d7c3 100644
--- a/libswscale/aarch64/swscale.c
+++ b/libswscale/aarch64/swscale.c
@@ -225,6 +225,9 @@ void ff_chrRangeToJpeg_neon(int16_t *dstU, int16_t *dstV, int width);
 
 av_cold void ff_sws_init_range_convert_aarch64(SwsContext *c)
 {
+    int cpu_flags = av_get_cpu_flags();
+
+    if (have_neon(cpu_flags)) {
     if (c->srcRange != c->dstRange && !isAnyRGB(c->dstFormat)) {
         if (c->dstBpc <= 14) {
             if (c->srcRange) {
@@ -236,6 +239,7 @@ av_cold void ff_sws_init_range_convert_aarch64(SwsContext *c)
             }
         }
     }
+    }
 }
 
 av_cold void ff_sws_init_swscale_aarch64(SwsContext *c)
@@ -296,6 +300,5 @@ av_cold void ff_sws_init_swscale_aarch64(SwsContext *c)
         default:
             break;
         }
-        ff_sws_init_range_convert_aarch64(c);
     }
 }
diff --git a/libswscale/loongarch/swscale_init_loongarch.c b/libswscale/loongarch/swscale_init_loongarch.c
index 2a95ede6d9..88ad21a103 100644
--- a/libswscale/loongarch/swscale_init_loongarch.c
+++ b/libswscale/loongarch/swscale_init_loongarch.c
@@ -95,7 +95,6 @@ av_cold void ff_sws_init_swscale_loongarch(SwsContext *c)
         }
     }
 #endif // #if HAVE_LASX
-    ff_sws_init_range_convert_loongarch(c);
 }
 
 av_cold void rgb2rgb_init_loongarch(void)
diff --git a/libswscale/riscv/swscale.c b/libswscale/riscv/swscale.c
index c452d93e5d..ad579308c5 100644
--- a/libswscale/riscv/swscale.c
+++ b/libswscale/riscv/swscale.c
@@ -26,9 +26,11 @@ void ff_range_chr_to_jpeg_16_rvv(int16_t *, int16_t *, int);
 void ff_range_lum_from_jpeg_16_rvv(int16_t *, int);
 void ff_range_chr_from_jpeg_16_rvv(int16_t *, int16_t *, int);
 
-av_cold static void ff_sws_init_range_convert_riscv(SwsContext *c, int flags)
+av_cold void ff_sws_init_range_convert_riscv(SwsContext *c)
 {
 #if HAVE_RVV
+    int flags = av_get_cpu_flags();
+
     static const struct {
         void (*lum)(int16_t *, int);
         void (*chr)(int16_t *, int16_t *, int);
@@ -67,9 +69,9 @@ RVV_INPUT(rgba32);
 
 av_cold void ff_sws_init_swscale_riscv(SwsContext *c)
 {
+#if HAVE_RVV
     int flags = av_get_cpu_flags();
 
-#if HAVE_RVV
     if ((flags & AV_CPU_FLAG_RVV_I32) && (flags & AV_CPU_FLAG_RVB)) {
         switch (c->srcFormat) {
             case AV_PIX_FMT_ABGR:
@@ -122,5 +124,4 @@ av_cold void ff_sws_init_swscale_riscv(SwsContext *c)
         }
     }
 #endif
-    ff_sws_init_range_convert_riscv(c, flags);
 }
diff --git a/libswscale/swscale.c b/libswscale/swscale.c
index 8b6a3a84b4..7f47dab4b6 100644
--- a/libswscale/swscale.c
+++ b/libswscale/swscale.c
@@ -554,6 +554,16 @@ av_cold void ff_sws_init_range_convert(SwsContext *c)
             }
         }
     }
+
+#if ARCH_AARCH64
+    ff_sws_init_range_convert_aarch64(c);
+#elif ARCH_LOONGARCH64
+    ff_sws_init_range_convert_loongarch(c);
+#elif ARCH_RISCV
+    ff_sws_init_range_convert_riscv(c);
+#elif ARCH_X86
+    ff_sws_init_range_convert_x86(c);
+#endif
 }
 
 static av_cold void sws_init_swscale(SwsContext *c)
diff --git a/libswscale/swscale_internal.h b/libswscale/swscale_internal.h
index 50127d288f..66be22ac05 100644
--- a/libswscale/swscale_internal.h
+++ b/libswscale/swscale_internal.h
@@ -697,6 +697,7 @@ void ff_updateMMXDitherTables(SwsContext *c, int dstY);
 av_cold void ff_sws_init_range_convert(SwsContext *c);
 av_cold void ff_sws_init_range_convert_aarch64(SwsContext *c);
 av_cold void ff_sws_init_range_convert_loongarch(SwsContext *c);
+av_cold void ff_sws_init_range_convert_riscv(SwsContext *c);
 av_cold void ff_sws_init_range_convert_x86(SwsContext *c);
 
 SwsFunc ff_yuv2rgb_init_x86(SwsContext *c);
diff --git a/libswscale/utils.c b/libswscale/utils.c
index c3154d82c1..4dec29ad96 100644
--- a/libswscale/utils.c
+++ b/libswscale/utils.c
@@ -1078,16 +1078,8 @@ int sws_setColorspaceDetails(struct SwsContext *c, const int inv_table[4],
     c->srcRange   = srcRange;
     c->dstRange   = dstRange;
 
-    if (need_reinit) {
+    if (need_reinit)
         ff_sws_init_range_convert(c);
-#if ARCH_AARCH64
-        ff_sws_init_range_convert_aarch64(c);
-#elif ARCH_LOONGARCH64
-        ff_sws_init_range_convert_loongarch(c);
-#elif ARCH_X86
-        ff_sws_init_range_convert_x86(c);
-#endif
-    }
 
     c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);
     c->srcFormatBpp = av_get_bits_per_pixel(desc_src);
diff --git a/libswscale/x86/swscale.c b/libswscale/x86/swscale.c
index 43319fd6b2..c82311d87b 100644
--- a/libswscale/x86/swscale.c
+++ b/libswscale/x86/swscale.c
@@ -851,6 +851,4 @@ switch(c->dstBpc){ \
     }
 
 #endif
-
-    ff_sws_init_range_convert_x86(c);
 }
-- 
2.30.2