From d06dfaa5cbdd20acfd2364b16c0f4ae4ddb30a65 Mon Sep 17 00:00:00 2001
From: Diego Biurrun <diego@biurrun.de>
Date: Mon, 18 Jan 2016 10:52:21 +0100
Subject: [PATCH] x86: huffyuv: Use EXTERNAL_SSSE3_FAST convenience macro where
appropriate
---
libavcodec/x86/huffyuvdsp.asm | 4 ++--
libavcodec/x86/huffyuvdsp_init.c | 10 ++++++----
2 files changed, 8 insertions(+), 6 deletions(-)
diff --git a/libavcodec/x86/huffyuvdsp.asm b/libavcodec/x86/huffyuvdsp.asm
index e7536da9985..692162b5b60 100644
--- a/libavcodec/x86/huffyuvdsp.asm
+++ b/libavcodec/x86/huffyuvdsp.asm
@@ -146,8 +146,8 @@ cglobal add_hfyu_left_pred, 3,3,7, dst, src, w, left
psllq m0, 56
ADD_HFYU_LEFT_LOOP 1, 1
-INIT_XMM sse4
-cglobal add_hfyu_left_pred, 3,3,7, dst, src, w, left
+INIT_XMM ssse3
+cglobal add_hfyu_left_pred_unaligned, 3,3,7, dst, src, w, left
mova m5, [pb_f]
mova m6, [pb_zzzzzzzz77777777]
mova m4, [pb_zzzz3333zzzzbbbb]
diff --git a/libavcodec/x86/huffyuvdsp_init.c b/libavcodec/x86/huffyuvdsp_init.c
index 75537d7a4c3..80e6cfbb12d 100644
--- a/libavcodec/x86/huffyuvdsp_init.c
+++ b/libavcodec/x86/huffyuvdsp_init.c
@@ -31,8 +31,8 @@ void ff_add_hfyu_median_pred_mmxext(uint8_t *dst, const uint8_t *top,
int ff_add_hfyu_left_pred_ssse3(uint8_t *dst, const uint8_t *src,
int w, int left);
-int ff_add_hfyu_left_pred_sse4(uint8_t *dst, const uint8_t *src,
- int w, int left);
+int ff_add_hfyu_left_pred_unaligned_ssse3(uint8_t *dst, const uint8_t *src,
+ int w, int left);
#if HAVE_INLINE_ASM
@@ -124,7 +124,9 @@ av_cold void ff_huffyuvdsp_init_x86(HuffYUVDSPContext *c)
if (EXTERNAL_SSSE3(cpu_flags)) {
c->add_hfyu_left_pred = ff_add_hfyu_left_pred_ssse3;
- if (cpu_flags & AV_CPU_FLAG_SSE4) // not really SSE4, just slow on Conroe
- c->add_hfyu_left_pred = ff_add_hfyu_left_pred_sse4;
+ }
+
+ if (EXTERNAL_SSSE3_FAST(cpu_flags)) {
+ c->add_hfyu_left_pred = ff_add_hfyu_left_pred_unaligned_ssse3;
}
}
--
GitLab