Merge remote-tracking branch 'qatar/master'

* qatar/master:
  dsputilenc: x86: Convert pixel inline asm to yasm
  libgsm: detect libgsm header path

Merged-by: Michael Niedermayer <michaelni@gmx.at>

3 files changed, 177 insertions, 181 deletions

diff --git a/libavcodec/libgsm.c b/libavcodec/libgsm.c
index f3cfb91f17..fd476e02e2 100644
--- a/libavcodec/libgsm.c
+++ b/libavcodec/libgsm.c
@@ -27,7 +27,12 @@
 
 // The idiosyncrasies of GSM-in-WAV are explained at http://kbs.cs.tu-berlin.de/~jutta/toast.html
 
+#include "config.h"
+#if HAVE_GSM_H
+#include <gsm.h>
+#else
 #include <gsm/gsm.h>
+#endif
 
 #include "libavutil/channel_layout.h"
 #include "libavutil/common.h"
diff --git a/libavcodec/x86/dsputilenc.asm b/libavcodec/x86/dsputilenc.asm
index b1dc99e2a2..55eacf5b96 100644
--- a/libavcodec/x86/dsputilenc.asm
+++ b/libavcodec/x86/dsputilenc.asm
@@ -333,3 +333,155 @@ cglobal sse16, 5, 5, 8
     paddd     m7, m1
     movd     eax, m7         ; return value
     RET
+
+INIT_MMX mmx
+; get_pixels_mmx(DCTELEM *block, const uint8_t *pixels, int line_size)
+cglobal get_pixels, 3,4
+    movsxdifnidn r2, r2d
+    add          r0, 128
+    mov          r3, -128
+    pxor         m7, m7
+.loop:
+    mova         m0, [r1]
+    mova         m2, [r1+r2]
+    mova         m1, m0
+    mova         m3, m2
+    punpcklbw    m0, m7
+    punpckhbw    m1, m7
+    punpcklbw    m2, m7
+    punpckhbw    m3, m7
+    mova [r0+r3+ 0], m0
+    mova [r0+r3+ 8], m1
+    mova [r0+r3+16], m2
+    mova [r0+r3+24], m3
+    lea          r1, [r1+r2*2]
+    add          r3, 32
+    js .loop
+    REP_RET
+
+INIT_XMM sse2
+cglobal get_pixels, 3, 4
+    movsxdifnidn r2, r2d
+    lea          r3, [r2*3]
+    pxor         m4, m4
+    movh         m0, [r1]
+    movh         m1, [r1+r2]
+    movh         m2, [r1+r2*2]
+    movh         m3, [r1+r3]
+    lea          r1, [r1+r2*4]
+    punpcklbw    m0, m4
+    punpcklbw    m1, m4
+    punpcklbw    m2, m4
+    punpcklbw    m3, m4
+    mova       [r0], m0
+    mova  [r0+0x10], m1
+    mova  [r0+0x20], m2
+    mova  [r0+0x30], m3
+    movh         m0, [r1]
+    movh         m1, [r1+r2*1]
+    movh         m2, [r1+r2*2]
+    movh         m3, [r1+r3]
+    punpcklbw    m0, m4
+    punpcklbw    m1, m4
+    punpcklbw    m2, m4
+    punpcklbw    m3, m4
+    mova  [r0+0x40], m0
+    mova  [r0+0x50], m1
+    mova  [r0+0x60], m2
+    mova  [r0+0x70], m3
+    RET
+
+INIT_MMX mmx
+; diff_pixels_mmx(DCTELEM *block, const uint8_t *s1, const unint8_t *s2, stride)
+cglobal diff_pixels, 4,5
+    movsxdifnidn r3, r3d
+    pxor         m7, m7
+    add          r0,  128
+    mov          r4, -128
+.loop:
+    mova         m0, [r1]
+    mova         m2, [r2]
+    mova         m1, m0
+    mova         m3, m2
+    punpcklbw    m0, m7
+    punpckhbw    m1, m7
+    punpcklbw    m2, m7
+    punpckhbw    m3, m7
+    psubw        m0, m2
+    psubw        m1, m3
+    mova  [r0+r4+0], m0
+    mova  [r0+r4+8], m1
+    add          r1, r3
+    add          r2, r3
+    add          r4, 16
+    jne .loop
+    REP_RET
+
+INIT_MMX mmx
+; pix_sum16_mmx(uint8_t * pix, int line_size)
+cglobal pix_sum16, 2, 3
+    movsxdifnidn r1, r1d
+    mov          r2, r1
+    neg          r2
+    shl          r2, 4
+    sub          r0, r2
+    pxor         m7, m7
+    pxor         m6, m6
+.loop:
+    mova         m0, [r0+r2+0]
+    mova         m1, [r0+r2+0]
+    mova         m2, [r0+r2+8]
+    mova         m3, [r0+r2+8]
+    punpcklbw    m0, m7
+    punpckhbw    m1, m7
+    punpcklbw    m2, m7
+    punpckhbw    m3, m7
+    paddw        m1, m0
+    paddw        m3, m2
+    paddw        m3, m1
+    paddw        m6, m3
+    add          r2, r1
+    js .loop
+    mova         m5, m6
+    psrlq        m6, 32
+    paddw        m6, m5
+    mova         m5, m6
+    psrlq        m6, 16
+    paddw        m6, m5
+    movd        eax, m6
+    and         eax, 0xffff
+    RET
+
+INIT_MMX mmx
+; pix_norm1_mmx(uint8_t *pix, int line_size)
+cglobal pix_norm1, 2, 4
+    movsxdifnidn r1, r1d
+    mov          r2, 16
+    pxor         m0, m0
+    pxor         m7, m7
+.loop:
+    mova         m2, [r0+0]
+    mova         m3, [r0+8]
+    mova         m1, m2
+    punpckhbw    m1, m0
+    punpcklbw    m2, m0
+    mova         m4, m3
+    punpckhbw    m3, m0
+    punpcklbw    m4, m0
+    pmaddwd      m1, m1
+    pmaddwd      m2, m2
+    pmaddwd      m3, m3
+    pmaddwd      m4, m4
+    paddd        m2, m1
+    paddd        m4, m3
+    paddd        m7, m2
+    add          r0, r1
+    paddd        m7, m4
+    dec r2
+    jne .loop
+    mova         m1, m7
+    psrlq        m7, 32
+    paddd        m1, m7
+    movd        eax, m1
+    RET
+
diff --git a/libavcodec/x86/dsputilenc_mmx.c b/libavcodec/x86/dsputilenc_mmx.c
index 48fcce477d..8a50a26ce2 100644
--- a/libavcodec/x86/dsputilenc_mmx.c
+++ b/libavcodec/x86/dsputilenc_mmx.c
@@ -30,181 +30,14 @@
 #include "libavcodec/mathops.h"
 #include "dsputil_mmx.h"
 
+void ff_get_pixels_mmx(DCTELEM *block, const uint8_t *pixels, int line_size);
+void ff_get_pixels_sse2(DCTELEM *block, const uint8_t *pixels, int line_size);
+void ff_diff_pixels_mmx(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
+int ff_pix_sum16_mmx(uint8_t * pix, int line_size);
+int ff_pix_norm1_mmx(uint8_t *pix, int line_size);
 
 #if HAVE_INLINE_ASM
 
-static void get_pixels_mmx(DCTELEM *block, const uint8_t *pixels, int line_size)
-{
-    __asm__ volatile(
-        "mov $-128, %%"REG_a"           \n\t"
-        "pxor %%mm7, %%mm7              \n\t"
-        ".p2align 4                     \n\t"
-        "1:                             \n\t"
-        "movq (%0), %%mm0               \n\t"
-        "movq (%0, %2), %%mm2           \n\t"
-        "movq %%mm0, %%mm1              \n\t"
-        "movq %%mm2, %%mm3              \n\t"
-        "punpcklbw %%mm7, %%mm0         \n\t"
-        "punpckhbw %%mm7, %%mm1         \n\t"
-        "punpcklbw %%mm7, %%mm2         \n\t"
-        "punpckhbw %%mm7, %%mm3         \n\t"
-        "movq %%mm0, (%1, %%"REG_a")    \n\t"
-        "movq %%mm1, 8(%1, %%"REG_a")   \n\t"
-        "movq %%mm2, 16(%1, %%"REG_a")  \n\t"
-        "movq %%mm3, 24(%1, %%"REG_a")  \n\t"
-        "add %3, %0                     \n\t"
-        "add $32, %%"REG_a"             \n\t"
-        "js 1b                          \n\t"
-        : "+r" (pixels)
-        : "r" (block+64), "r" ((x86_reg)line_size), "r" ((x86_reg)line_size*2)
-        : "%"REG_a
-    );
-}
-
-static void get_pixels_sse2(DCTELEM *block, const uint8_t *pixels, int line_size)
-{
-    __asm__ volatile(
-        "pxor %%xmm4,      %%xmm4         \n\t"
-        "movq (%0),        %%xmm0         \n\t"
-        "movq (%0, %2),    %%xmm1         \n\t"
-        "movq (%0, %2,2),  %%xmm2         \n\t"
-        "movq (%0, %3),    %%xmm3         \n\t"
-        "lea (%0,%2,4), %0                \n\t"
-        "punpcklbw %%xmm4, %%xmm0         \n\t"
-        "punpcklbw %%xmm4, %%xmm1         \n\t"
-        "punpcklbw %%xmm4, %%xmm2         \n\t"
-        "punpcklbw %%xmm4, %%xmm3         \n\t"
-        "movdqa %%xmm0,      (%1)         \n\t"
-        "movdqa %%xmm1,    16(%1)         \n\t"
-        "movdqa %%xmm2,    32(%1)         \n\t"
-        "movdqa %%xmm3,    48(%1)         \n\t"
-        "movq (%0),        %%xmm0         \n\t"
-        "movq (%0, %2),    %%xmm1         \n\t"
-        "movq (%0, %2,2),  %%xmm2         \n\t"
-        "movq (%0, %3),    %%xmm3         \n\t"
-        "punpcklbw %%xmm4, %%xmm0         \n\t"
-        "punpcklbw %%xmm4, %%xmm1         \n\t"
-        "punpcklbw %%xmm4, %%xmm2         \n\t"
-        "punpcklbw %%xmm4, %%xmm3         \n\t"
-        "movdqa %%xmm0,    64(%1)         \n\t"
-        "movdqa %%xmm1,    80(%1)         \n\t"
-        "movdqa %%xmm2,    96(%1)         \n\t"
-        "movdqa %%xmm3,   112(%1)         \n\t"
-        : "+r" (pixels)
-        : "r" (block), "r" ((x86_reg)line_size), "r" ((x86_reg)line_size*3)
-    );
-}
-
-static inline void diff_pixels_mmx(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride)
-{
-    __asm__ volatile(
-        "pxor %%mm7, %%mm7              \n\t"
-        "mov $-128, %%"REG_a"           \n\t"
-        ".p2align 4                     \n\t"
-        "1:                             \n\t"
-        "movq (%0), %%mm0               \n\t"
-        "movq (%1), %%mm2               \n\t"
-        "movq %%mm0, %%mm1              \n\t"
-        "movq %%mm2, %%mm3              \n\t"
-        "punpcklbw %%mm7, %%mm0         \n\t"
-        "punpckhbw %%mm7, %%mm1         \n\t"
-        "punpcklbw %%mm7, %%mm2         \n\t"
-        "punpckhbw %%mm7, %%mm3         \n\t"
-        "psubw %%mm2, %%mm0             \n\t"
-        "psubw %%mm3, %%mm1             \n\t"
-        "movq %%mm0, (%2, %%"REG_a")    \n\t"
-        "movq %%mm1, 8(%2, %%"REG_a")   \n\t"
-        "add %3, %0                     \n\t"
-        "add %3, %1                     \n\t"
-        "add $16, %%"REG_a"             \n\t"
-        "jnz 1b                         \n\t"
-        : "+r" (s1), "+r" (s2)
-        : "r" (block+64), "r" ((x86_reg)stride)
-        : "%"REG_a
-    );
-}
-
-static int pix_sum16_mmx(uint8_t * pix, int line_size){
-    const int h=16;
-    int sum;
-    x86_reg index= -line_size*h;
-
-    __asm__ volatile(
-                "pxor %%mm7, %%mm7              \n\t"
-                "pxor %%mm6, %%mm6              \n\t"
-                "1:                             \n\t"
-                "movq (%2, %1), %%mm0           \n\t"
-                "movq (%2, %1), %%mm1           \n\t"
-                "movq 8(%2, %1), %%mm2          \n\t"
-                "movq 8(%2, %1), %%mm3          \n\t"
-                "punpcklbw %%mm7, %%mm0         \n\t"
-                "punpckhbw %%mm7, %%mm1         \n\t"
-                "punpcklbw %%mm7, %%mm2         \n\t"
-                "punpckhbw %%mm7, %%mm3         \n\t"
-                "paddw %%mm0, %%mm1             \n\t"
-                "paddw %%mm2, %%mm3             \n\t"
-                "paddw %%mm1, %%mm3             \n\t"
-                "paddw %%mm3, %%mm6             \n\t"
-                "add %3, %1                     \n\t"
-                " js 1b                         \n\t"
-                "movq %%mm6, %%mm5              \n\t"
-                "psrlq $32, %%mm6               \n\t"
-                "paddw %%mm5, %%mm6             \n\t"
-                "movq %%mm6, %%mm5              \n\t"
-                "psrlq $16, %%mm6               \n\t"
-                "paddw %%mm5, %%mm6             \n\t"
-                "movd %%mm6, %0                 \n\t"
-                "andl $0xFFFF, %0               \n\t"
-                : "=&r" (sum), "+r" (index)
-                : "r" (pix - index), "r" ((x86_reg)line_size)
-        );
-
-        return sum;
-}
-
-static int pix_norm1_mmx(uint8_t *pix, int line_size) {
-    int tmp;
-  __asm__ volatile (
-      "movl $16,%%ecx\n"
-      "pxor %%mm0,%%mm0\n"
-      "pxor %%mm7,%%mm7\n"
-      "1:\n"
-      "movq (%0),%%mm2\n"       /* mm2 = pix[0-7] */
-      "movq 8(%0),%%mm3\n"      /* mm3 = pix[8-15] */
-
-      "movq %%mm2,%%mm1\n"      /* mm1 = mm2 = pix[0-7] */
-
-      "punpckhbw %%mm0,%%mm1\n" /* mm1 = [pix4-7] */
-      "punpcklbw %%mm0,%%mm2\n" /* mm2 = [pix0-3] */
-
-      "movq %%mm3,%%mm4\n"      /* mm4 = mm3 = pix[8-15] */
-      "punpckhbw %%mm0,%%mm3\n" /* mm3 = [pix12-15] */
-      "punpcklbw %%mm0,%%mm4\n" /* mm4 = [pix8-11] */
-
-      "pmaddwd %%mm1,%%mm1\n"   /* mm1 = (pix0^2+pix1^2,pix2^2+pix3^2) */
-      "pmaddwd %%mm2,%%mm2\n"   /* mm2 = (pix4^2+pix5^2,pix6^2+pix7^2) */
-
-      "pmaddwd %%mm3,%%mm3\n"
-      "pmaddwd %%mm4,%%mm4\n"
-
-      "paddd %%mm1,%%mm2\n"     /* mm2 = (pix0^2+pix1^2+pix4^2+pix5^2,
-                                          pix2^2+pix3^2+pix6^2+pix7^2) */
-      "paddd %%mm3,%%mm4\n"
-      "paddd %%mm2,%%mm7\n"
-
-      "add %2, %0\n"
-      "paddd %%mm4,%%mm7\n"
-      "dec %%ecx\n"
-      "jnz 1b\n"
-
-      "movq %%mm7,%%mm1\n"
-      "psrlq $32, %%mm7\n"      /* shift hi dword to lo */
-      "paddd %%mm7,%%mm1\n"
-      "movd %%mm1,%1\n"
-      : "+r" (pix), "=r"(tmp) : "r" ((x86_reg)line_size) : "%ecx" );
-    return tmp;
-}
-
 static int sse8_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
     int tmp;
   __asm__ volatile (
@@ -1112,10 +945,23 @@ hadamard_func(ssse3)
 void ff_dsputilenc_init_mmx(DSPContext* c, AVCodecContext *avctx)
 {
     int mm_flags = av_get_cpu_flags();
-
-#if HAVE_INLINE_ASM
     int bit_depth = avctx->bits_per_raw_sample;
 
+#if HAVE_YASM
+    if (EXTERNAL_MMX(mm_flags)) {
+        if (bit_depth <= 8)
+            c->get_pixels = ff_get_pixels_mmx;
+        c->diff_pixels = ff_diff_pixels_mmx;
+        c->pix_sum = ff_pix_sum16_mmx;
+
+        c->pix_norm1 = ff_pix_norm1_mmx;
+    }
+    if (EXTERNAL_SSE2(mm_flags))
+        if (bit_depth <= 8)
+            c->get_pixels = ff_get_pixels_sse2;
+#endif /* HAVE_YASM */
+
+#if HAVE_INLINE_ASM
     if (mm_flags & AV_CPU_FLAG_MMX) {
         const int dct_algo = avctx->dct_algo;
         if (avctx->bits_per_raw_sample <= 8 &&
@@ -1129,15 +975,10 @@ void ff_dsputilenc_init_mmx(DSPContext* c, AVCodecContext *avctx)
             }
         }
 
-        if (bit_depth <= 8)
-            c->get_pixels = get_pixels_mmx;
-        c->diff_pixels = diff_pixels_mmx;
-        c->pix_sum = pix_sum16_mmx;
 
         c->diff_bytes= diff_bytes_mmx;
         c->sum_abs_dctelem= sum_abs_dctelem_mmx;
 
-        c->pix_norm1 = pix_norm1_mmx;
         c->sse[0] = sse16_mmx;
         c->sse[1] = sse8_mmx;
         c->vsad[4]= vsad_intra16_mmx;
@@ -1167,8 +1008,6 @@ void ff_dsputilenc_init_mmx(DSPContext* c, AVCodecContext *avctx)
         }
 
         if(mm_flags & AV_CPU_FLAG_SSE2){
-            if (bit_depth <= 8)
-                c->get_pixels = get_pixels_sse2;
             c->sum_abs_dctelem= sum_abs_dctelem_sse2;
         }