pixman: Branch 'master' - 7 commits
Matt Turner
mattst88 at kemper.freedesktop.org
Sun Jul 1 13:47:35 PDT 2012
pixman/loongson-mmintrin.h | 116 ++++++++++++++++++
pixman/pixman-mmx.c | 280 ++++++++++++++++++++++++++++++++++++++++++++-
pixman/pixman-sse2.c | 3
3 files changed, 393 insertions(+), 6 deletions(-)
New commits:
commit 4cdf8e9f3aca1925aeca25debb9268877ba3cd3d
Author: Matt Turner <mattst88 at gmail.com>
Date: Sun Jul 1 16:35:46 2012 -0400
sse2: add missing ABGR entires for bilinear src_8888_8888
diff --git a/pixman/pixman-sse2.c b/pixman/pixman-sse2.c
index f665b37..665eead 100644
--- a/pixman/pixman-sse2.c
+++ b/pixman/pixman-sse2.c
@@ -5861,6 +5861,9 @@ static const pixman_fast_path_t sse2_fast_paths[] =
SIMPLE_BILINEAR_FAST_PATH (SRC, a8r8g8b8, a8r8g8b8, sse2_8888_8888),
SIMPLE_BILINEAR_FAST_PATH (SRC, a8r8g8b8, x8r8g8b8, sse2_8888_8888),
SIMPLE_BILINEAR_FAST_PATH (SRC, x8r8g8b8, x8r8g8b8, sse2_8888_8888),
+ SIMPLE_BILINEAR_FAST_PATH (SRC, a8b8g8r8, a8b8g8r8, sse2_8888_8888),
+ SIMPLE_BILINEAR_FAST_PATH (SRC, a8b8g8r8, x8b8g8r8, sse2_8888_8888),
+ SIMPLE_BILINEAR_FAST_PATH (SRC, x8b8g8r8, x8b8g8r8, sse2_8888_8888),
SIMPLE_BILINEAR_FAST_PATH (OVER, a8r8g8b8, x8r8g8b8, sse2_8888_8888),
SIMPLE_BILINEAR_FAST_PATH (OVER, a8b8g8r8, x8b8g8r8, sse2_8888_8888),
commit ef99f9e97260cc55678385a6d691c195f57bd6b1
Author: Matt Turner <mattst88 at gmail.com>
Date: Mon May 21 05:56:58 2012 -0400
loongson: optimize _mm_set_pi* functions with shuffle instructions
diff --git a/pixman/loongson-mmintrin.h b/pixman/loongson-mmintrin.h
index f0931ac..086c6e0 100644
--- a/pixman/loongson-mmintrin.h
+++ b/pixman/loongson-mmintrin.h
@@ -182,9 +182,34 @@ _mm_packs_pi32 (__m64 __m1, __m64 __m2)
return ret;
}
+#define _MM_SHUFFLE(fp3,fp2,fp1,fp0) \
+ (((fp3) << 6) | ((fp2) << 4) | ((fp1) << 2) | (fp0))
extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_set_pi16 (uint16_t __w3, uint16_t __w2, uint16_t __w1, uint16_t __w0)
{
+ if (__builtin_constant_p (__w3) &&
+ __builtin_constant_p (__w2) &&
+ __builtin_constant_p (__w1) &&
+ __builtin_constant_p (__w0))
+ {
+ uint64_t val = ((uint64_t)__w3 << 48)
+ | ((uint64_t)__w2 << 32)
+ | ((uint64_t)__w1 << 16)
+ | ((uint64_t)__w0 << 0);
+ return *(__m64 *)&val;
+ }
+ else if (__w3 == __w2 && __w2 == __w1 && __w1 == __w0)
+ {
+ /* TODO: handle other cases */
+ uint64_t val = __w3;
+ uint64_t imm = _MM_SHUFFLE (0, 0, 0, 0);
+ __m64 ret;
+ asm("pshufh %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (*(__m64 *)&val), "f" (*(__m64 *)&imm)
+ );
+ return ret;
+ }
uint64_t val = ((uint64_t)__w3 << 48)
| ((uint64_t)__w2 << 32)
| ((uint64_t)__w1 << 16)
@@ -195,10 +220,28 @@ _mm_set_pi16 (uint16_t __w3, uint16_t __w2, uint16_t __w1, uint16_t __w0)
extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_set_pi32 (unsigned __i1, unsigned __i0)
{
+ if (__builtin_constant_p (__i1) &&
+ __builtin_constant_p (__i0))
+ {
+ uint64_t val = ((uint64_t)__i1 << 32)
+ | ((uint64_t)__i0 << 0);
+ return *(__m64 *)&val;
+ }
+ else if (__i1 == __i0)
+ {
+ uint64_t imm = _MM_SHUFFLE (1, 0, 1, 0);
+ __m64 ret;
+ asm("pshufh %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (*(__m32 *)&__i1), "f" (*(__m64 *)&imm)
+ );
+ return ret;
+ }
uint64_t val = ((uint64_t)__i1 << 32)
| ((uint64_t)__i0 << 0);
return *(__m64 *)&val;
}
+#undef _MM_SHUFFLE
extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_shuffle_pi16 (__m64 __m, int64_t __n)
commit 9aa8e3a26071739d160496ef9f6126f296c500eb
Author: Matt Turner <mattst88 at gmail.com>
Date: Wed Jun 27 13:00:36 2012 -0400
mmx: optimize bilinear function when using 7-bit precision
Loongson:
image firefox-fishtank 1037.738 1040.218 0.19% 3/3
image firefox-fishtank 1056.611 1057.581 0.20% 3/3
ARM/iwMMXt:
image firefox-fishtank 1487.282 1492.640 0.17% 3/3
image firefox-fishtank 1363.913 1364.366 0.11% 3/3
diff --git a/pixman/pixman-mmx.c b/pixman/pixman-mmx.c
index 0c79f3a..5441d6b 100644
--- a/pixman/pixman-mmx.c
+++ b/pixman/pixman-mmx.c
@@ -3522,11 +3522,14 @@ mmx_composite_over_reverse_n_8888 (pixman_implementation_t *imp,
}
#define BSHIFT ((1 << BILINEAR_INTERPOLATION_BITS))
+#define BMSK (BSHIFT - 1)
#define BILINEAR_DECLARE_VARIABLES \
const __m64 mm_wt = _mm_set_pi16 (wt, wt, wt, wt); \
const __m64 mm_wb = _mm_set_pi16 (wb, wb, wb, wb); \
const __m64 mm_BSHIFT = _mm_set_pi16 (BSHIFT, BSHIFT, BSHIFT, BSHIFT); \
+ const __m64 mm_addc7 = _mm_set_pi16 (0, 1, 0, 1); \
+ const __m64 mm_xorc7 = _mm_set_pi16 (0, BMSK, 0, BMSK); \
const __m64 mm_ux = _mm_set_pi16 (unit_x, unit_x, unit_x, unit_x); \
const __m64 mm_zero = _mm_setzero_si64 (); \
__m64 mm_x = _mm_set_pi16 (vx, vx, vx, vx)
@@ -3544,21 +3547,37 @@ do { \
__m64 b_lo = _mm_mullo_pi16 (_mm_unpacklo_pi8 (b, mm_zero), mm_wb); \
__m64 hi = _mm_add_pi16 (t_hi, b_hi); \
__m64 lo = _mm_add_pi16 (t_lo, b_lo); \
- /* calculate horizontal weights */ \
- __m64 mm_wh_lo = _mm_sub_pi16 (mm_BSHIFT, _mm_srli_pi16 (mm_x, \
+ if (BILINEAR_INTERPOLATION_BITS < 8) \
+ { \
+ /* calculate horizontal weights */ \
+ __m64 mm_wh = _mm_add_pi16 (mm_addc7, _mm_xor_si64 (mm_xorc7, \
+ _mm_srli_pi16 (mm_x, \
+ 16 - BILINEAR_INTERPOLATION_BITS))); \
+ mm_x = _mm_add_pi16 (mm_x, mm_ux); \
+ /* horizontal interpolation */ \
+ __m64 p = _mm_unpacklo_pi16 (lo, hi); \
+ __m64 q = _mm_unpackhi_pi16 (lo, hi); \
+ lo = _mm_madd_pi16 (p, mm_wh); \
+ hi = _mm_madd_pi16 (q, mm_wh); \
+ } \
+ else \
+ { \
+ /* calculate horizontal weights */ \
+ __m64 mm_wh_lo = _mm_sub_pi16 (mm_BSHIFT, _mm_srli_pi16 (mm_x, \
16 - BILINEAR_INTERPOLATION_BITS)); \
- __m64 mm_wh_hi = _mm_srli_pi16 (mm_x, \
+ __m64 mm_wh_hi = _mm_srli_pi16 (mm_x, \
16 - BILINEAR_INTERPOLATION_BITS); \
- mm_x = _mm_add_pi16 (mm_x, mm_ux); \
- /* horizontal interpolation */ \
- __m64 mm_lo_lo = _mm_mullo_pi16 (lo, mm_wh_lo); \
- __m64 mm_lo_hi = _mm_mullo_pi16 (hi, mm_wh_hi); \
- __m64 mm_hi_lo = _mm_mulhi_pu16 (lo, mm_wh_lo); \
- __m64 mm_hi_hi = _mm_mulhi_pu16 (hi, mm_wh_hi); \
- lo = _mm_add_pi32 (_mm_unpacklo_pi16 (mm_lo_lo, mm_hi_lo), \
- _mm_unpacklo_pi16 (mm_lo_hi, mm_hi_hi)); \
- hi = _mm_add_pi32 (_mm_unpackhi_pi16 (mm_lo_lo, mm_hi_lo), \
- _mm_unpackhi_pi16 (mm_lo_hi, mm_hi_hi)); \
+ mm_x = _mm_add_pi16 (mm_x, mm_ux); \
+ /* horizontal interpolation */ \
+ __m64 mm_lo_lo = _mm_mullo_pi16 (lo, mm_wh_lo); \
+ __m64 mm_lo_hi = _mm_mullo_pi16 (hi, mm_wh_hi); \
+ __m64 mm_hi_lo = _mm_mulhi_pu16 (lo, mm_wh_lo); \
+ __m64 mm_hi_hi = _mm_mulhi_pu16 (hi, mm_wh_hi); \
+ lo = _mm_add_pi32 (_mm_unpacklo_pi16 (mm_lo_lo, mm_hi_lo), \
+ _mm_unpacklo_pi16 (mm_lo_hi, mm_hi_hi)); \
+ hi = _mm_add_pi32 (_mm_unpackhi_pi16 (mm_lo_lo, mm_hi_lo), \
+ _mm_unpackhi_pi16 (mm_lo_hi, mm_hi_hi)); \
+ } \
/* shift and pack the result */ \
hi = _mm_srli_pi32 (hi, BILINEAR_INTERPOLATION_BITS * 2); \
lo = _mm_srli_pi32 (lo, BILINEAR_INTERPOLATION_BITS * 2); \
commit 1ad6ae6ee8a350f6fe4f30ba928aacf44d04f86e
Author: Matt Turner <mattst88 at gmail.com>
Date: Sun May 20 20:51:08 2012 -0400
mmx: add scaled bilinear over_8888_8_8888
Loongson:
image firefox-fishtank 1665.163 1670.370 0.17% 3/3
image firefox-fishtank 1037.738 1040.218 0.19% 3/3
ARM/iwMMXt:
image firefox-fishtank 2042.723 2045.308 0.10% 3/3
image firefox-fishtank 1487.282 1492.640 0.17% 3/3
diff --git a/pixman/pixman-mmx.c b/pixman/pixman-mmx.c
index bf66a63..0c79f3a 100644
--- a/pixman/pixman-mmx.c
+++ b/pixman/pixman-mmx.c
@@ -3567,6 +3567,12 @@ do { \
pix = lo; \
} while (0)
+#define BILINEAR_SKIP_ONE_PIXEL() \
+do { \
+ vx += unit_x; \
+ mm_x = _mm_add_pi16 (mm_x, mm_ux); \
+} while(0)
+
static force_inline void
scaled_bilinear_scanline_mmx_8888_8888_SRC (uint32_t * dst,
const uint32_t * mask,
@@ -3659,6 +3665,79 @@ FAST_BILINEAR_MAINLOOP_COMMON (mmx_8888_8888_normal_OVER,
scaled_bilinear_scanline_mmx_8888_8888_OVER,
uint32_t, uint32_t, uint32_t,
NORMAL, FLAG_NONE)
+
+static force_inline void
+scaled_bilinear_scanline_mmx_8888_8_8888_OVER (uint32_t * dst,
+ const uint8_t * mask,
+ const uint32_t * src_top,
+ const uint32_t * src_bottom,
+ int32_t w,
+ int wt,
+ int wb,
+ pixman_fixed_t vx,
+ pixman_fixed_t unit_x,
+ pixman_fixed_t max_vx,
+ pixman_bool_t zero_src)
+{
+ BILINEAR_DECLARE_VARIABLES;
+ __m64 pix1, pix2;
+ uint32_t m;
+
+ while (w)
+ {
+ m = (uint32_t) *mask++;
+
+ if (m)
+ {
+ BILINEAR_INTERPOLATE_ONE_PIXEL (pix1);
+
+ if (m == 0xff && is_opaque (pix1))
+ {
+ store (dst, pix1);
+ }
+ else
+ {
+ __m64 ms, md, ma, msa;
+
+ pix2 = load (dst);
+ ma = expand_alpha_rev (to_m64 (m));
+ ms = _mm_unpacklo_pi8 (pix1, _mm_setzero_si64 ());
+ md = _mm_unpacklo_pi8 (pix2, _mm_setzero_si64 ());
+
+ msa = expand_alpha (ms);
+
+ store8888 (dst, (in_over (ms, msa, ma, md)));
+ }
+ }
+ else
+ {
+ BILINEAR_SKIP_ONE_PIXEL ();
+ }
+
+ w--;
+ dst++;
+ }
+
+ _mm_empty ();
+}
+
+FAST_BILINEAR_MAINLOOP_COMMON (mmx_8888_8_8888_cover_OVER,
+ scaled_bilinear_scanline_mmx_8888_8_8888_OVER,
+ uint32_t, uint8_t, uint32_t,
+ COVER, FLAG_HAVE_NON_SOLID_MASK)
+FAST_BILINEAR_MAINLOOP_COMMON (mmx_8888_8_8888_pad_OVER,
+ scaled_bilinear_scanline_mmx_8888_8_8888_OVER,
+ uint32_t, uint8_t, uint32_t,
+ PAD, FLAG_HAVE_NON_SOLID_MASK)
+FAST_BILINEAR_MAINLOOP_COMMON (mmx_8888_8_8888_none_OVER,
+ scaled_bilinear_scanline_mmx_8888_8_8888_OVER,
+ uint32_t, uint8_t, uint32_t,
+ NONE, FLAG_HAVE_NON_SOLID_MASK)
+FAST_BILINEAR_MAINLOOP_COMMON (mmx_8888_8_8888_normal_OVER,
+ scaled_bilinear_scanline_mmx_8888_8_8888_OVER,
+ uint32_t, uint8_t, uint32_t,
+ NORMAL, FLAG_HAVE_NON_SOLID_MASK)
+
static uint32_t *
mmx_fetch_x8r8g8b8 (pixman_iter_t *iter, const uint32_t *mask)
{
@@ -3926,6 +4005,11 @@ static const pixman_fast_path_t mmx_fast_paths[] =
SIMPLE_BILINEAR_FAST_PATH (OVER, a8r8g8b8, a8r8g8b8, mmx_8888_8888 ),
SIMPLE_BILINEAR_FAST_PATH (OVER, a8b8g8r8, a8b8g8r8, mmx_8888_8888 ),
+ SIMPLE_BILINEAR_A8_MASK_FAST_PATH (OVER, a8r8g8b8, x8r8g8b8, mmx_8888_8_8888 ),
+ SIMPLE_BILINEAR_A8_MASK_FAST_PATH (OVER, a8b8g8r8, x8b8g8r8, mmx_8888_8_8888 ),
+ SIMPLE_BILINEAR_A8_MASK_FAST_PATH (OVER, a8r8g8b8, a8r8g8b8, mmx_8888_8_8888 ),
+ SIMPLE_BILINEAR_A8_MASK_FAST_PATH (OVER, a8b8g8r8, a8b8g8r8, mmx_8888_8_8888 ),
+
{ PIXMAN_OP_NONE },
};
commit c43de364cbcd195f7d1d6881a6109cbb3d6b73b8
Author: Matt Turner <mattst88 at gmail.com>
Date: Wed Jun 27 12:57:45 2012 -0400
mmx: add scaled bilinear over_8888_8888
Loongson:
image firefox-planet-gnome 157.012 158.087 0.30% 6/6
image firefox-planet-gnome 156.617 157.109 0.15% 5/6
ARM/iwMMXt:
image firefox-planet-gnome 148.086 149.339 0.76% 6/6
image firefox-planet-gnome 144.939 146.123 0.61% 6/6
diff --git a/pixman/pixman-mmx.c b/pixman/pixman-mmx.c
index 018a2ba..bf66a63 100644
--- a/pixman/pixman-mmx.c
+++ b/pixman/pixman-mmx.c
@@ -695,6 +695,24 @@ combine (const uint32_t *src, const uint32_t *mask)
return vsrc;
}
+static force_inline __m64
+core_combine_over_u_pixel_mmx (__m64 vsrc, __m64 vdst)
+{
+ vsrc = _mm_unpacklo_pi8 (vsrc, _mm_setzero_si64 ());
+
+ if (is_opaque (vsrc))
+ {
+ return vsrc;
+ }
+ else if (!is_zero (vsrc))
+ {
+ return over (vsrc, expand_alpha (vsrc),
+ _mm_unpacklo_pi8 (vdst, _mm_setzero_si64 ()));
+ }
+
+ return _mm_unpacklo_pi8 (vdst, _mm_setzero_si64 ());
+}
+
static void
mmx_combine_over_u (pixman_implementation_t *imp,
pixman_op_t op,
@@ -3546,7 +3564,7 @@ do { \
lo = _mm_srli_pi32 (lo, BILINEAR_INTERPOLATION_BITS * 2); \
lo = _mm_packs_pi32 (lo, hi); \
lo = _mm_packs_pu16 (lo, lo); \
- store (&pix, lo); \
+ pix = lo; \
} while (0)
static force_inline void
@@ -3563,12 +3581,13 @@ scaled_bilinear_scanline_mmx_8888_8888_SRC (uint32_t * dst,
pixman_bool_t zero_src)
{
BILINEAR_DECLARE_VARIABLES;
- uint32_t pix;
+ __m64 pix;
while (w--)
{
BILINEAR_INTERPOLATE_ONE_PIXEL (pix);
- *dst++ = pix;
+ store (dst, pix);
+ dst++;
}
_mm_empty ();
@@ -3591,6 +3610,55 @@ FAST_BILINEAR_MAINLOOP_COMMON (mmx_8888_8888_normal_SRC,
uint32_t, uint32_t, uint32_t,
NORMAL, FLAG_NONE)
+static force_inline void
+scaled_bilinear_scanline_mmx_8888_8888_OVER (uint32_t * dst,
+ const uint32_t * mask,
+ const uint32_t * src_top,
+ const uint32_t * src_bottom,
+ int32_t w,
+ int wt,
+ int wb,
+ pixman_fixed_t vx,
+ pixman_fixed_t unit_x,
+ pixman_fixed_t max_vx,
+ pixman_bool_t zero_src)
+{
+ BILINEAR_DECLARE_VARIABLES;
+ __m64 pix1, pix2;
+
+ while (w)
+ {
+ BILINEAR_INTERPOLATE_ONE_PIXEL (pix1);
+
+ if (!is_zero (pix1))
+ {
+ pix2 = load (dst);
+ store8888 (dst, core_combine_over_u_pixel_mmx (pix1, pix2));
+ }
+
+ w--;
+ dst++;
+ }
+
+ _mm_empty ();
+}
+
+FAST_BILINEAR_MAINLOOP_COMMON (mmx_8888_8888_cover_OVER,
+ scaled_bilinear_scanline_mmx_8888_8888_OVER,
+ uint32_t, uint32_t, uint32_t,
+ COVER, FLAG_NONE)
+FAST_BILINEAR_MAINLOOP_COMMON (mmx_8888_8888_pad_OVER,
+ scaled_bilinear_scanline_mmx_8888_8888_OVER,
+ uint32_t, uint32_t, uint32_t,
+ PAD, FLAG_NONE)
+FAST_BILINEAR_MAINLOOP_COMMON (mmx_8888_8888_none_OVER,
+ scaled_bilinear_scanline_mmx_8888_8888_OVER,
+ uint32_t, uint32_t, uint32_t,
+ NONE, FLAG_NONE)
+FAST_BILINEAR_MAINLOOP_COMMON (mmx_8888_8888_normal_OVER,
+ scaled_bilinear_scanline_mmx_8888_8888_OVER,
+ uint32_t, uint32_t, uint32_t,
+ NORMAL, FLAG_NONE)
static uint32_t *
mmx_fetch_x8r8g8b8 (pixman_iter_t *iter, const uint32_t *mask)
{
@@ -3853,6 +3921,11 @@ static const pixman_fast_path_t mmx_fast_paths[] =
SIMPLE_BILINEAR_FAST_PATH (SRC, a8b8g8r8, x8b8g8r8, mmx_8888_8888 ),
SIMPLE_BILINEAR_FAST_PATH (SRC, x8b8g8r8, x8b8g8r8, mmx_8888_8888 ),
+ SIMPLE_BILINEAR_FAST_PATH (OVER, a8r8g8b8, x8r8g8b8, mmx_8888_8888 ),
+ SIMPLE_BILINEAR_FAST_PATH (OVER, a8b8g8r8, x8b8g8r8, mmx_8888_8888 ),
+ SIMPLE_BILINEAR_FAST_PATH (OVER, a8r8g8b8, a8r8g8b8, mmx_8888_8888 ),
+ SIMPLE_BILINEAR_FAST_PATH (OVER, a8b8g8r8, a8b8g8r8, mmx_8888_8888 ),
+
{ PIXMAN_OP_NONE },
};
commit 9209cd746b7a81d0536df6dadd6a0b0b983291cb
Author: Matt Turner <mattst88 at gmail.com>
Date: Tue Jun 19 00:30:51 2012 -0400
mmx: add scaled bilinear src_8888_8888
Loongson:
image firefox-planet-gnome 170.025 170.229 0.09% 3/4
image firefox-planet-gnome 157.012 158.087 0.30% 6/6
ARM/iwMMXt:
image firefox-planet-gnome 164.192 164.875 0.34% 3/4
image firefox-planet-gnome 148.086 149.339 0.76% 6/6
diff --git a/pixman/loongson-mmintrin.h b/pixman/loongson-mmintrin.h
index 1a114fe..f0931ac 100644
--- a/pixman/loongson-mmintrin.h
+++ b/pixman/loongson-mmintrin.h
@@ -45,6 +45,28 @@ _mm_setzero_si64 (void)
}
extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_add_pi16 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+ asm("paddh %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+ return ret;
+}
+
+extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_add_pi32 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+ asm("paddw %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+ return ret;
+}
+
+extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_adds_pu16 (__m64 __m1, __m64 __m2)
{
__m64 ret;
@@ -150,6 +172,35 @@ _mm_packs_pu16 (__m64 __m1, __m64 __m2)
}
extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_packs_pi32 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+ asm("packsswh %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+ return ret;
+}
+
+extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_set_pi16 (uint16_t __w3, uint16_t __w2, uint16_t __w1, uint16_t __w0)
+{
+ uint64_t val = ((uint64_t)__w3 << 48)
+ | ((uint64_t)__w2 << 32)
+ | ((uint64_t)__w1 << 16)
+ | ((uint64_t)__w0 << 0);
+ return *(__m64 *)&val;
+}
+
+extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_set_pi32 (unsigned __i1, unsigned __i0)
+{
+ uint64_t val = ((uint64_t)__i1 << 32)
+ | ((uint64_t)__i0 << 0);
+ return *(__m64 *)&val;
+}
+
+extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_shuffle_pi16 (__m64 __m, int64_t __n)
{
__m64 ret;
@@ -193,6 +244,17 @@ _mm_srli_pi16 (__m64 __m, int64_t __count)
}
extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_srli_pi32 (__m64 __m, int64_t __count)
+{
+ __m64 ret;
+ asm("psrlw %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m), "f" (*(__m64 *)&__count)
+ );
+ return ret;
+}
+
+extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_srli_si64 (__m64 __m, int64_t __count)
{
__m64 ret;
@@ -204,6 +266,17 @@ _mm_srli_si64 (__m64 __m, int64_t __count)
}
extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_sub_pi16 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+ asm("psubh %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+ return ret;
+}
+
+extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_unpackhi_pi8 (__m64 __m1, __m64 __m2)
{
__m64 ret;
diff --git a/pixman/pixman-mmx.c b/pixman/pixman-mmx.c
index 0ebe119..018a2ba 100644
--- a/pixman/pixman-mmx.c
+++ b/pixman/pixman-mmx.c
@@ -42,6 +42,7 @@
#endif
#include "pixman-private.h"
#include "pixman-combine32.h"
+#include "pixman-inlines.h"
#define no_vERBOSE
@@ -3502,6 +3503,94 @@ mmx_composite_over_reverse_n_8888 (pixman_implementation_t *imp,
_mm_empty ();
}
+#define BSHIFT ((1 << BILINEAR_INTERPOLATION_BITS))
+
+#define BILINEAR_DECLARE_VARIABLES \
+ const __m64 mm_wt = _mm_set_pi16 (wt, wt, wt, wt); \
+ const __m64 mm_wb = _mm_set_pi16 (wb, wb, wb, wb); \
+ const __m64 mm_BSHIFT = _mm_set_pi16 (BSHIFT, BSHIFT, BSHIFT, BSHIFT); \
+ const __m64 mm_ux = _mm_set_pi16 (unit_x, unit_x, unit_x, unit_x); \
+ const __m64 mm_zero = _mm_setzero_si64 (); \
+ __m64 mm_x = _mm_set_pi16 (vx, vx, vx, vx)
+
+#define BILINEAR_INTERPOLATE_ONE_PIXEL(pix) \
+do { \
+ /* fetch 2x2 pixel block into 2 mmx registers */ \
+ __m64 t = ldq_u ((__m64 *)&src_top [pixman_fixed_to_int (vx)]); \
+ __m64 b = ldq_u ((__m64 *)&src_bottom [pixman_fixed_to_int (vx)]); \
+ vx += unit_x; \
+ /* vertical interpolation */ \
+ __m64 t_hi = _mm_mullo_pi16 (_mm_unpackhi_pi8 (t, mm_zero), mm_wt); \
+ __m64 t_lo = _mm_mullo_pi16 (_mm_unpacklo_pi8 (t, mm_zero), mm_wt); \
+ __m64 b_hi = _mm_mullo_pi16 (_mm_unpackhi_pi8 (b, mm_zero), mm_wb); \
+ __m64 b_lo = _mm_mullo_pi16 (_mm_unpacklo_pi8 (b, mm_zero), mm_wb); \
+ __m64 hi = _mm_add_pi16 (t_hi, b_hi); \
+ __m64 lo = _mm_add_pi16 (t_lo, b_lo); \
+ /* calculate horizontal weights */ \
+ __m64 mm_wh_lo = _mm_sub_pi16 (mm_BSHIFT, _mm_srli_pi16 (mm_x, \
+ 16 - BILINEAR_INTERPOLATION_BITS)); \
+ __m64 mm_wh_hi = _mm_srli_pi16 (mm_x, \
+ 16 - BILINEAR_INTERPOLATION_BITS); \
+ mm_x = _mm_add_pi16 (mm_x, mm_ux); \
+ /* horizontal interpolation */ \
+ __m64 mm_lo_lo = _mm_mullo_pi16 (lo, mm_wh_lo); \
+ __m64 mm_lo_hi = _mm_mullo_pi16 (hi, mm_wh_hi); \
+ __m64 mm_hi_lo = _mm_mulhi_pu16 (lo, mm_wh_lo); \
+ __m64 mm_hi_hi = _mm_mulhi_pu16 (hi, mm_wh_hi); \
+ lo = _mm_add_pi32 (_mm_unpacklo_pi16 (mm_lo_lo, mm_hi_lo), \
+ _mm_unpacklo_pi16 (mm_lo_hi, mm_hi_hi)); \
+ hi = _mm_add_pi32 (_mm_unpackhi_pi16 (mm_lo_lo, mm_hi_lo), \
+ _mm_unpackhi_pi16 (mm_lo_hi, mm_hi_hi)); \
+ /* shift and pack the result */ \
+ hi = _mm_srli_pi32 (hi, BILINEAR_INTERPOLATION_BITS * 2); \
+ lo = _mm_srli_pi32 (lo, BILINEAR_INTERPOLATION_BITS * 2); \
+ lo = _mm_packs_pi32 (lo, hi); \
+ lo = _mm_packs_pu16 (lo, lo); \
+ store (&pix, lo); \
+} while (0)
+
+static force_inline void
+scaled_bilinear_scanline_mmx_8888_8888_SRC (uint32_t * dst,
+ const uint32_t * mask,
+ const uint32_t * src_top,
+ const uint32_t * src_bottom,
+ int32_t w,
+ int wt,
+ int wb,
+ pixman_fixed_t vx,
+ pixman_fixed_t unit_x,
+ pixman_fixed_t max_vx,
+ pixman_bool_t zero_src)
+{
+ BILINEAR_DECLARE_VARIABLES;
+ uint32_t pix;
+
+ while (w--)
+ {
+ BILINEAR_INTERPOLATE_ONE_PIXEL (pix);
+ *dst++ = pix;
+ }
+
+ _mm_empty ();
+}
+
+FAST_BILINEAR_MAINLOOP_COMMON (mmx_8888_8888_cover_SRC,
+ scaled_bilinear_scanline_mmx_8888_8888_SRC,
+ uint32_t, uint32_t, uint32_t,
+ COVER, FLAG_NONE)
+FAST_BILINEAR_MAINLOOP_COMMON (mmx_8888_8888_pad_SRC,
+ scaled_bilinear_scanline_mmx_8888_8888_SRC,
+ uint32_t, uint32_t, uint32_t,
+ PAD, FLAG_NONE)
+FAST_BILINEAR_MAINLOOP_COMMON (mmx_8888_8888_none_SRC,
+ scaled_bilinear_scanline_mmx_8888_8888_SRC,
+ uint32_t, uint32_t, uint32_t,
+ NONE, FLAG_NONE)
+FAST_BILINEAR_MAINLOOP_COMMON (mmx_8888_8888_normal_SRC,
+ scaled_bilinear_scanline_mmx_8888_8888_SRC,
+ uint32_t, uint32_t, uint32_t,
+ NORMAL, FLAG_NONE)
+
static uint32_t *
mmx_fetch_x8r8g8b8 (pixman_iter_t *iter, const uint32_t *mask)
{
@@ -3757,6 +3846,13 @@ static const pixman_fast_path_t mmx_fast_paths[] =
PIXMAN_STD_FAST_PATH (IN, a8, null, a8, mmx_composite_in_8_8 ),
PIXMAN_STD_FAST_PATH (IN, solid, a8, a8, mmx_composite_in_n_8_8 ),
+ SIMPLE_BILINEAR_FAST_PATH (SRC, a8r8g8b8, a8r8g8b8, mmx_8888_8888 ),
+ SIMPLE_BILINEAR_FAST_PATH (SRC, a8r8g8b8, x8r8g8b8, mmx_8888_8888 ),
+ SIMPLE_BILINEAR_FAST_PATH (SRC, x8r8g8b8, x8r8g8b8, mmx_8888_8888 ),
+ SIMPLE_BILINEAR_FAST_PATH (SRC, a8b8g8r8, a8b8g8r8, mmx_8888_8888 ),
+ SIMPLE_BILINEAR_FAST_PATH (SRC, a8b8g8r8, x8b8g8r8, mmx_8888_8888 ),
+ SIMPLE_BILINEAR_FAST_PATH (SRC, x8b8g8r8, x8b8g8r8, mmx_8888_8888 ),
+
{ PIXMAN_OP_NONE },
};
commit 51f27d7364d66e47d882ee531b6655368159231a
Author: Matt Turner <mattst88 at gmail.com>
Date: Thu Jun 28 12:17:16 2012 -0400
mmx: Use expand_alpha instead of mask/shift
diff --git a/pixman/pixman-mmx.c b/pixman/pixman-mmx.c
index d869c04..0ebe119 100644
--- a/pixman/pixman-mmx.c
+++ b/pixman/pixman-mmx.c
@@ -1599,9 +1599,7 @@ mmx_composite_over_8888_n_8888 (pixman_implementation_t *imp,
PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
mask = _pixman_image_get_solid (imp, mask_image, dest_image->bits.format);
- mask &= 0xff000000;
- mask = mask | mask >> 8 | mask >> 16 | mask >> 24;
- vmask = load8888 (&mask);
+ vmask = expand_alpha (load8888 (&mask));
while (height--)
{
@@ -1670,9 +1668,7 @@ mmx_composite_over_x888_n_8888 (pixman_implementation_t *imp,
PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
mask = _pixman_image_get_solid (imp, mask_image, dest_image->bits.format);
- mask &= 0xff000000;
- mask = mask | mask >> 8 | mask >> 16 | mask >> 24;
- vmask = load8888 (&mask);
+ vmask = expand_alpha (load8888 (&mask));
srca = MC (4x00ff);
while (height--)
More information about the xorg-commit
mailing list