From 9209cd746b7a81d0536df6dadd6a0b0b983291cb Mon Sep 17 00:00:00 2001 From: Matt Turner Date: Tue, 19 Jun 2012 00:30:51 -0400 Subject: 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 --- pixman/loongson-mmintrin.h | 73 +++++++++++++++++++++++++++++++++++ pixman/pixman-mmx.c | 96 ++++++++++++++++++++++++++++++++++++++++++++++ 2 files changed, 169 insertions(+) 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 @@ -44,6 +44,28 @@ _mm_setzero_si64 (void) return 0.0; } +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) { @@ -149,6 +171,35 @@ _mm_packs_pu16 (__m64 __m1, __m64 __m2) return ret; } +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) { @@ -192,6 +243,17 @@ _mm_srli_pi16 (__m64 __m, int64_t __count) return ret; } +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) { @@ -203,6 +265,17 @@ _mm_srli_si64 (__m64 __m, int64_t __count) return ret; } +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) { 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 }, }; -- cgit v1.2.3