michael@0: /*
michael@0:  * Copyright © 2004, 2005 Red Hat, Inc.
michael@0:  * Copyright © 2004 Nicholas Miell
michael@0:  * Copyright © 2005 Trolltech AS
michael@0:  *
michael@0:  * Permission to use, copy, modify, distribute, and sell this software and its
michael@0:  * documentation for any purpose is hereby granted without fee, provided that
michael@0:  * the above copyright notice appear in all copies and that both that
michael@0:  * copyright notice and this permission notice appear in supporting
michael@0:  * documentation, and that the name of Red Hat not be used in advertising or
michael@0:  * publicity pertaining to distribution of the software without specific,
michael@0:  * written prior permission.  Red Hat makes no representations about the
michael@0:  * suitability of this software for any purpose.  It is provided "as is"
michael@0:  * without express or implied warranty.
michael@0:  *
michael@0:  * THE COPYRIGHT HOLDERS DISCLAIM ALL WARRANTIES WITH REGARD TO THIS
michael@0:  * SOFTWARE, INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND
michael@0:  * FITNESS, IN NO EVENT SHALL THE COPYRIGHT HOLDERS BE LIABLE FOR ANY
michael@0:  * SPECIAL, INDIRECT OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
michael@0:  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN
michael@0:  * AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING
michael@0:  * OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS
michael@0:  * SOFTWARE.
michael@0:  *
michael@0:  * Author:  Søren Sandmann (sandmann@redhat.com)
michael@0:  * Minor Improvements: Nicholas Miell (nmiell@gmail.com)
michael@0:  * MMX code paths for fbcompose.c by Lars Knoll (lars@trolltech.com)
michael@0:  *
michael@0:  * Based on work by Owen Taylor
michael@0:  */
michael@0: 
michael@0: #ifdef HAVE_CONFIG_H
michael@0: #include <config.h>
michael@0: #endif
michael@0: 
michael@0: #if defined USE_X86_MMX || defined USE_ARM_IWMMXT || defined USE_LOONGSON_MMI
michael@0: 
michael@0: #ifdef USE_LOONGSON_MMI
michael@0: #include <loongson-mmintrin.h>
michael@0: #else
michael@0: #include <mmintrin.h>
michael@0: #endif
michael@0: #include "pixman-private.h"
michael@0: #include "pixman-combine32.h"
michael@0: #include "pixman-inlines.h"
michael@0: 
michael@0: #define no_vERBOSE
michael@0: 
michael@0: #ifdef VERBOSE
michael@0: #define CHECKPOINT() error_f ("at %s %d\n", __FUNCTION__, __LINE__)
michael@0: #else
michael@0: #define CHECKPOINT()
michael@0: #endif
michael@0: 
michael@0: #if defined USE_ARM_IWMMXT && __GNUC__ == 4 && __GNUC_MINOR__ < 8
michael@0: /* Empty the multimedia state. For some reason, ARM's mmintrin.h doesn't provide this.  */
michael@0: extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
michael@0: _mm_empty (void)
michael@0: {
michael@0: 
michael@0: }
michael@0: #endif
michael@0: 
michael@0: #ifdef USE_X86_MMX
michael@0: # if (defined(__SUNPRO_C) || defined(_MSC_VER) || defined(_WIN64))
michael@0: #  include <xmmintrin.h>
michael@0: # else
michael@0: /* We have to compile with -msse to use xmmintrin.h, but that causes SSE
michael@0:  * instructions to be generated that we don't want. Just duplicate the
michael@0:  * functions we want to use.  */
michael@0: extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
michael@0: _mm_movemask_pi8 (__m64 __A)
michael@0: {
michael@0:     int ret;
michael@0: 
michael@0:     asm ("pmovmskb %1, %0\n\t"
michael@0: 	: "=r" (ret)
michael@0: 	: "y" (__A)
michael@0:     );
michael@0: 
michael@0:     return ret;
michael@0: }
michael@0: 
michael@0: extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
michael@0: _mm_mulhi_pu16 (__m64 __A, __m64 __B)
michael@0: {
michael@0:     asm ("pmulhuw %1, %0\n\t"
michael@0: 	: "+y" (__A)
michael@0: 	: "y" (__B)
michael@0:     );
michael@0:     return __A;
michael@0: }
michael@0: 
michael@0: #  ifdef __OPTIMIZE__
michael@0: extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
michael@0: _mm_shuffle_pi16 (__m64 __A, int8_t const __N)
michael@0: {
michael@0:     __m64 ret;
michael@0: 
michael@0:     asm ("pshufw %2, %1, %0\n\t"
michael@0: 	: "=y" (ret)
michael@0: 	: "y" (__A), "K" (__N)
michael@0:     );
michael@0: 
michael@0:     return ret;
michael@0: }
michael@0: #  else
michael@0: #   define _mm_shuffle_pi16(A, N)					\
michael@0:     ({									\
michael@0: 	__m64 ret;							\
michael@0: 									\
michael@0: 	asm ("pshufw %2, %1, %0\n\t"					\
michael@0: 	     : "=y" (ret)						\
michael@0: 	     : "y" (A), "K" ((const int8_t)N)				\
michael@0: 	);								\
michael@0: 									\
michael@0: 	ret;								\
michael@0:     })
michael@0: #  endif
michael@0: # endif
michael@0: #endif
michael@0: 
michael@0: #ifndef _MSC_VER
michael@0: #define _MM_SHUFFLE(fp3,fp2,fp1,fp0) \
michael@0:  (((fp3) << 6) | ((fp2) << 4) | ((fp1) << 2) | (fp0))
michael@0: #endif
michael@0: 
michael@0: /* Notes about writing mmx code
michael@0:  *
michael@0:  * give memory operands as the second operand. If you give it as the
michael@0:  * first, gcc will first load it into a register, then use that
michael@0:  * register
michael@0:  *
michael@0:  *   ie. use
michael@0:  *
michael@0:  *         _mm_mullo_pi16 (x, mmx_constant);
michael@0:  *
michael@0:  *   not
michael@0:  *
michael@0:  *         _mm_mullo_pi16 (mmx_constant, x);
michael@0:  *
michael@0:  * Also try to minimize dependencies. i.e. when you need a value, try
michael@0:  * to calculate it from a value that was calculated as early as
michael@0:  * possible.
michael@0:  */
michael@0: 
michael@0: /* --------------- MMX primitives ------------------------------------- */
michael@0: 
michael@0: /* If __m64 is defined as a struct or union, then define M64_MEMBER to be
michael@0:  * the name of the member used to access the data.
michael@0:  * If __m64 requires using mm_cvt* intrinsics functions to convert between
michael@0:  * uint64_t and __m64 values, then define USE_CVT_INTRINSICS.
michael@0:  * If __m64 and uint64_t values can just be cast to each other directly,
michael@0:  * then define USE_M64_CASTS.
michael@0:  * If __m64 is a double datatype, then define USE_M64_DOUBLE.
michael@0:  */
michael@0: #ifdef _MSC_VER
michael@0: # define M64_MEMBER m64_u64
michael@0: #elif defined(__ICC)
michael@0: # define USE_CVT_INTRINSICS
michael@0: #elif defined(USE_LOONGSON_MMI)
michael@0: # define USE_M64_DOUBLE
michael@0: #elif defined(__GNUC__)
michael@0: # define USE_M64_CASTS
michael@0: #elif defined(__SUNPRO_C)
michael@0: # if (__SUNPRO_C >= 0x5120) && !defined(__NOVECTORSIZE__)
michael@0: /* Solaris Studio 12.3 (Sun C 5.12) introduces __attribute__(__vector_size__)
michael@0:  * support, and defaults to using it to define __m64, unless __NOVECTORSIZE__
michael@0:  * is defined.   If it is used, then the mm_cvt* intrinsics must be used.
michael@0:  */
michael@0: #  define USE_CVT_INTRINSICS
michael@0: # else
michael@0: /* For Studio 12.2 or older, or when __attribute__(__vector_size__) is
michael@0:  * disabled, __m64 is defined as a struct containing "unsigned long long l_".
michael@0:  */
michael@0: #  define M64_MEMBER l_
michael@0: # endif
michael@0: #endif
michael@0: 
michael@0: #if defined(USE_M64_CASTS) || defined(USE_CVT_INTRINSICS) || defined(USE_M64_DOUBLE)
michael@0: typedef uint64_t mmxdatafield;
michael@0: #else
michael@0: typedef __m64 mmxdatafield;
michael@0: #endif
michael@0: 
michael@0: typedef struct
michael@0: {
michael@0:     mmxdatafield mmx_4x00ff;
michael@0:     mmxdatafield mmx_4x0080;
michael@0:     mmxdatafield mmx_565_rgb;
michael@0:     mmxdatafield mmx_565_unpack_multiplier;
michael@0:     mmxdatafield mmx_565_pack_multiplier;
michael@0:     mmxdatafield mmx_565_r;
michael@0:     mmxdatafield mmx_565_g;
michael@0:     mmxdatafield mmx_565_b;
michael@0:     mmxdatafield mmx_packed_565_rb;
michael@0:     mmxdatafield mmx_packed_565_g;
michael@0:     mmxdatafield mmx_expand_565_g;
michael@0:     mmxdatafield mmx_expand_565_b;
michael@0:     mmxdatafield mmx_expand_565_r;
michael@0: #ifndef USE_LOONGSON_MMI
michael@0:     mmxdatafield mmx_mask_0;
michael@0:     mmxdatafield mmx_mask_1;
michael@0:     mmxdatafield mmx_mask_2;
michael@0:     mmxdatafield mmx_mask_3;
michael@0: #endif
michael@0:     mmxdatafield mmx_full_alpha;
michael@0:     mmxdatafield mmx_4x0101;
michael@0:     mmxdatafield mmx_ff000000;
michael@0: } mmx_data_t;
michael@0: 
michael@0: #if defined(_MSC_VER)
michael@0: # define MMXDATA_INIT(field, val) { val ## UI64 }
michael@0: #elif defined(M64_MEMBER)       /* __m64 is a struct, not an integral type */
michael@0: # define MMXDATA_INIT(field, val) field =   { val ## ULL }
michael@0: #else                           /* mmxdatafield is an integral type */
michael@0: # define MMXDATA_INIT(field, val) field =   val ## ULL
michael@0: #endif
michael@0: 
michael@0: static const mmx_data_t c =
michael@0: {
michael@0:     MMXDATA_INIT (.mmx_4x00ff,                   0x00ff00ff00ff00ff),
michael@0:     MMXDATA_INIT (.mmx_4x0080,                   0x0080008000800080),
michael@0:     MMXDATA_INIT (.mmx_565_rgb,                  0x000001f0003f001f),
michael@0:     MMXDATA_INIT (.mmx_565_unpack_multiplier,    0x0000008404100840),
michael@0:     MMXDATA_INIT (.mmx_565_pack_multiplier,      0x2000000420000004),
michael@0:     MMXDATA_INIT (.mmx_565_r,                    0x000000f800000000),
michael@0:     MMXDATA_INIT (.mmx_565_g,                    0x0000000000fc0000),
michael@0:     MMXDATA_INIT (.mmx_565_b,                    0x00000000000000f8),
michael@0:     MMXDATA_INIT (.mmx_packed_565_rb,            0x00f800f800f800f8),
michael@0:     MMXDATA_INIT (.mmx_packed_565_g,             0x0000fc000000fc00),
michael@0:     MMXDATA_INIT (.mmx_expand_565_g,             0x07e007e007e007e0),
michael@0:     MMXDATA_INIT (.mmx_expand_565_b,             0x001f001f001f001f),
michael@0:     MMXDATA_INIT (.mmx_expand_565_r,             0xf800f800f800f800),
michael@0: #ifndef USE_LOONGSON_MMI
michael@0:     MMXDATA_INIT (.mmx_mask_0,                   0xffffffffffff0000),
michael@0:     MMXDATA_INIT (.mmx_mask_1,                   0xffffffff0000ffff),
michael@0:     MMXDATA_INIT (.mmx_mask_2,                   0xffff0000ffffffff),
michael@0:     MMXDATA_INIT (.mmx_mask_3,                   0x0000ffffffffffff),
michael@0: #endif
michael@0:     MMXDATA_INIT (.mmx_full_alpha,               0x00ff000000000000),
michael@0:     MMXDATA_INIT (.mmx_4x0101,                   0x0101010101010101),
michael@0:     MMXDATA_INIT (.mmx_ff000000,                 0xff000000ff000000),
michael@0: };
michael@0: 
michael@0: #ifdef USE_CVT_INTRINSICS
michael@0: #    define MC(x) to_m64 (c.mmx_ ## x)
michael@0: #elif defined(USE_M64_CASTS)
michael@0: #    define MC(x) ((__m64)c.mmx_ ## x)
michael@0: #elif defined(USE_M64_DOUBLE)
michael@0: #    define MC(x) (*(__m64 *)&c.mmx_ ## x)
michael@0: #else
michael@0: #    define MC(x) c.mmx_ ## x
michael@0: #endif
michael@0: 
michael@0: static force_inline __m64
michael@0: to_m64 (uint64_t x)
michael@0: {
michael@0: #ifdef USE_CVT_INTRINSICS
michael@0:     return _mm_cvtsi64_m64 (x);
michael@0: #elif defined M64_MEMBER        /* __m64 is a struct, not an integral type */
michael@0:     __m64 res;
michael@0: 
michael@0:     res.M64_MEMBER = x;
michael@0:     return res;
michael@0: #elif defined USE_M64_DOUBLE
michael@0:     return *(__m64 *)&x;
michael@0: #else /* USE_M64_CASTS */
michael@0:     return (__m64)x;
michael@0: #endif
michael@0: }
michael@0: 
michael@0: static force_inline uint64_t
michael@0: to_uint64 (__m64 x)
michael@0: {
michael@0: #ifdef USE_CVT_INTRINSICS
michael@0:     return _mm_cvtm64_si64 (x);
michael@0: #elif defined M64_MEMBER        /* __m64 is a struct, not an integral type */
michael@0:     uint64_t res = x.M64_MEMBER;
michael@0:     return res;
michael@0: #elif defined USE_M64_DOUBLE
michael@0:     return *(uint64_t *)&x;
michael@0: #else /* USE_M64_CASTS */
michael@0:     return (uint64_t)x;
michael@0: #endif
michael@0: }
michael@0: 
michael@0: static force_inline __m64
michael@0: shift (__m64 v,
michael@0:        int   s)
michael@0: {
michael@0:     if (s > 0)
michael@0: 	return _mm_slli_si64 (v, s);
michael@0:     else if (s < 0)
michael@0: 	return _mm_srli_si64 (v, -s);
michael@0:     else
michael@0: 	return v;
michael@0: }
michael@0: 
michael@0: static force_inline __m64
michael@0: negate (__m64 mask)
michael@0: {
michael@0:     return _mm_xor_si64 (mask, MC (4x00ff));
michael@0: }
michael@0: 
michael@0: static force_inline __m64
michael@0: pix_multiply (__m64 a, __m64 b)
michael@0: {
michael@0:     __m64 res;
michael@0: 
michael@0:     res = _mm_mullo_pi16 (a, b);
michael@0:     res = _mm_adds_pu16 (res, MC (4x0080));
michael@0:     res = _mm_mulhi_pu16 (res, MC (4x0101));
michael@0: 
michael@0:     return res;
michael@0: }
michael@0: 
michael@0: static force_inline __m64
michael@0: pix_add (__m64 a, __m64 b)
michael@0: {
michael@0:     return _mm_adds_pu8 (a, b);
michael@0: }
michael@0: 
michael@0: static force_inline __m64
michael@0: expand_alpha (__m64 pixel)
michael@0: {
michael@0:     return _mm_shuffle_pi16 (pixel, _MM_SHUFFLE (3, 3, 3, 3));
michael@0: }
michael@0: 
michael@0: static force_inline __m64
michael@0: expand_alpha_rev (__m64 pixel)
michael@0: {
michael@0:     return _mm_shuffle_pi16 (pixel, _MM_SHUFFLE (0, 0, 0, 0));
michael@0: }
michael@0: 
michael@0: static force_inline __m64
michael@0: invert_colors (__m64 pixel)
michael@0: {
michael@0:     return _mm_shuffle_pi16 (pixel, _MM_SHUFFLE (3, 0, 1, 2));
michael@0: }
michael@0: 
michael@0: static force_inline __m64
michael@0: over (__m64 src,
michael@0:       __m64 srca,
michael@0:       __m64 dest)
michael@0: {
michael@0:     return _mm_adds_pu8 (src, pix_multiply (dest, negate (srca)));
michael@0: }
michael@0: 
michael@0: static force_inline __m64
michael@0: over_rev_non_pre (__m64 src, __m64 dest)
michael@0: {
michael@0:     __m64 srca = expand_alpha (src);
michael@0:     __m64 srcfaaa = _mm_or_si64 (srca, MC (full_alpha));
michael@0: 
michael@0:     return over (pix_multiply (invert_colors (src), srcfaaa), srca, dest);
michael@0: }
michael@0: 
michael@0: static force_inline __m64
michael@0: in (__m64 src, __m64 mask)
michael@0: {
michael@0:     return pix_multiply (src, mask);
michael@0: }
michael@0: 
michael@0: #ifndef _MSC_VER
michael@0: static force_inline __m64
michael@0: in_over (__m64 src, __m64 srca, __m64 mask, __m64 dest)
michael@0: {
michael@0:     return over (in (src, mask), pix_multiply (srca, mask), dest);
michael@0: }
michael@0: 
michael@0: #else
michael@0: 
michael@0: #define in_over(src, srca, mask, dest)					\
michael@0:     over (in (src, mask), pix_multiply (srca, mask), dest)
michael@0: 
michael@0: #endif
michael@0: 
michael@0: /* Elemental unaligned loads */
michael@0: 
michael@0: static force_inline __m64 ldq_u(__m64 *p)
michael@0: {
michael@0: #ifdef USE_X86_MMX
michael@0:     /* x86's alignment restrictions are very relaxed. */
michael@0:     return *(__m64 *)p;
michael@0: #elif defined USE_ARM_IWMMXT
michael@0:     int align = (uintptr_t)p & 7;
michael@0:     __m64 *aligned_p;
michael@0:     if (align == 0)
michael@0: 	return *p;
michael@0:     aligned_p = (__m64 *)((uintptr_t)p & ~7);
michael@0:     return (__m64) _mm_align_si64 (aligned_p[0], aligned_p[1], align);
michael@0: #else
michael@0:     struct __una_u64 { __m64 x __attribute__((packed)); };
michael@0:     const struct __una_u64 *ptr = (const struct __una_u64 *) p;
michael@0:     return (__m64) ptr->x;
michael@0: #endif
michael@0: }
michael@0: 
michael@0: static force_inline uint32_t ldl_u(const uint32_t *p)
michael@0: {
michael@0: #ifdef USE_X86_MMX
michael@0:     /* x86's alignment restrictions are very relaxed. */
michael@0:     return *p;
michael@0: #else
michael@0:     struct __una_u32 { uint32_t x __attribute__((packed)); };
michael@0:     const struct __una_u32 *ptr = (const struct __una_u32 *) p;
michael@0:     return ptr->x;
michael@0: #endif
michael@0: }
michael@0: 
michael@0: static force_inline __m64
michael@0: load (const uint32_t *v)
michael@0: {
michael@0: #ifdef USE_LOONGSON_MMI
michael@0:     __m64 ret;
michael@0:     asm ("lwc1 %0, %1\n\t"
michael@0: 	: "=f" (ret)
michael@0: 	: "m" (*v)
michael@0:     );
michael@0:     return ret;
michael@0: #else
michael@0:     return _mm_cvtsi32_si64 (*v);
michael@0: #endif
michael@0: }
michael@0: 
michael@0: static force_inline __m64
michael@0: load8888 (const uint32_t *v)
michael@0: {
michael@0: #ifdef USE_LOONGSON_MMI
michael@0:     return _mm_unpacklo_pi8_f (*(__m32 *)v, _mm_setzero_si64 ());
michael@0: #else
michael@0:     return _mm_unpacklo_pi8 (load (v), _mm_setzero_si64 ());
michael@0: #endif
michael@0: }
michael@0: 
michael@0: static force_inline __m64
michael@0: load8888u (const uint32_t *v)
michael@0: {
michael@0:     uint32_t l = ldl_u (v);
michael@0:     return load8888 (&l);
michael@0: }
michael@0: 
michael@0: static force_inline __m64
michael@0: pack8888 (__m64 lo, __m64 hi)
michael@0: {
michael@0:     return _mm_packs_pu16 (lo, hi);
michael@0: }
michael@0: 
michael@0: static force_inline void
michael@0: store (uint32_t *dest, __m64 v)
michael@0: {
michael@0: #ifdef USE_LOONGSON_MMI
michael@0:     asm ("swc1 %1, %0\n\t"
michael@0: 	: "=m" (*dest)
michael@0: 	: "f" (v)
michael@0: 	: "memory"
michael@0:     );
michael@0: #else
michael@0:     *dest = _mm_cvtsi64_si32 (v);
michael@0: #endif
michael@0: }
michael@0: 
michael@0: static force_inline void
michael@0: store8888 (uint32_t *dest, __m64 v)
michael@0: {
michael@0:     v = pack8888 (v, _mm_setzero_si64 ());
michael@0:     store (dest, v);
michael@0: }
michael@0: 
michael@0: static force_inline pixman_bool_t
michael@0: is_equal (__m64 a, __m64 b)
michael@0: {
michael@0: #ifdef USE_LOONGSON_MMI
michael@0:     /* __m64 is double, we can compare directly. */
michael@0:     return a == b;
michael@0: #else
michael@0:     return _mm_movemask_pi8 (_mm_cmpeq_pi8 (a, b)) == 0xff;
michael@0: #endif
michael@0: }
michael@0: 
michael@0: static force_inline pixman_bool_t
michael@0: is_opaque (__m64 v)
michael@0: {
michael@0: #ifdef USE_LOONGSON_MMI
michael@0:     return is_equal (_mm_and_si64 (v, MC (full_alpha)), MC (full_alpha));
michael@0: #else
michael@0:     __m64 ffs = _mm_cmpeq_pi8 (v, v);
michael@0:     return (_mm_movemask_pi8 (_mm_cmpeq_pi8 (v, ffs)) & 0x40);
michael@0: #endif
michael@0: }
michael@0: 
michael@0: static force_inline pixman_bool_t
michael@0: is_zero (__m64 v)
michael@0: {
michael@0:     return is_equal (v, _mm_setzero_si64 ());
michael@0: }
michael@0: 
michael@0: /* Expand 16 bits positioned at @pos (0-3) of a mmx register into
michael@0:  *
michael@0:  *    00RR00GG00BB
michael@0:  *
michael@0:  * --- Expanding 565 in the low word ---
michael@0:  *
michael@0:  * m = (m << (32 - 3)) | (m << (16 - 5)) | m;
michael@0:  * m = m & (01f0003f001f);
michael@0:  * m = m * (008404100840);
michael@0:  * m = m >> 8;
michael@0:  *
michael@0:  * Note the trick here - the top word is shifted by another nibble to
michael@0:  * avoid it bumping into the middle word
michael@0:  */
michael@0: static force_inline __m64
michael@0: expand565 (__m64 pixel, int pos)
michael@0: {
michael@0:     __m64 p = pixel;
michael@0:     __m64 t1, t2;
michael@0: 
michael@0:     /* move pixel to low 16 bit and zero the rest */
michael@0: #ifdef USE_LOONGSON_MMI
michael@0:     p = loongson_extract_pi16 (p, pos);
michael@0: #else
michael@0:     p = shift (shift (p, (3 - pos) * 16), -48);
michael@0: #endif
michael@0: 
michael@0:     t1 = shift (p, 36 - 11);
michael@0:     t2 = shift (p, 16 - 5);
michael@0: 
michael@0:     p = _mm_or_si64 (t1, p);
michael@0:     p = _mm_or_si64 (t2, p);
michael@0:     p = _mm_and_si64 (p, MC (565_rgb));
michael@0: 
michael@0:     pixel = _mm_mullo_pi16 (p, MC (565_unpack_multiplier));
michael@0:     return _mm_srli_pi16 (pixel, 8);
michael@0: }
michael@0: 
michael@0: /* Expand 4 16 bit pixels in an mmx register into two mmx registers of
michael@0:  *
michael@0:  *    AARRGGBBRRGGBB
michael@0:  */
michael@0: static force_inline void
michael@0: expand_4xpacked565 (__m64 vin, __m64 *vout0, __m64 *vout1, int full_alpha)
michael@0: {
michael@0:     __m64 t0, t1, alpha = _mm_setzero_si64 ();
michael@0:     __m64 r = _mm_and_si64 (vin, MC (expand_565_r));
michael@0:     __m64 g = _mm_and_si64 (vin, MC (expand_565_g));
michael@0:     __m64 b = _mm_and_si64 (vin, MC (expand_565_b));
michael@0:     if (full_alpha)
michael@0: 	alpha = _mm_cmpeq_pi32 (alpha, alpha);
michael@0: 
michael@0:     /* Replicate high bits into empty low bits. */
michael@0:     r = _mm_or_si64 (_mm_srli_pi16 (r, 8), _mm_srli_pi16 (r, 13));
michael@0:     g = _mm_or_si64 (_mm_srli_pi16 (g, 3), _mm_srli_pi16 (g, 9));
michael@0:     b = _mm_or_si64 (_mm_slli_pi16 (b, 3), _mm_srli_pi16 (b, 2));
michael@0: 
michael@0:     r = _mm_packs_pu16 (r, _mm_setzero_si64 ());	/* 00 00 00 00 R3 R2 R1 R0 */
michael@0:     g = _mm_packs_pu16 (g, _mm_setzero_si64 ());	/* 00 00 00 00 G3 G2 G1 G0 */
michael@0:     b = _mm_packs_pu16 (b, _mm_setzero_si64 ());	/* 00 00 00 00 B3 B2 B1 B0 */
michael@0: 
michael@0:     t1 = _mm_unpacklo_pi8 (r, alpha);			/* A3 R3 A2 R2 A1 R1 A0 R0 */
michael@0:     t0 = _mm_unpacklo_pi8 (b, g);			/* G3 B3 G2 B2 G1 B1 G0 B0 */
michael@0: 
michael@0:     *vout0 = _mm_unpacklo_pi16 (t0, t1);		/* A1 R1 G1 B1 A0 R0 G0 B0 */
michael@0:     *vout1 = _mm_unpackhi_pi16 (t0, t1);		/* A3 R3 G3 B3 A2 R2 G2 B2 */
michael@0: }
michael@0: 
michael@0: static force_inline __m64
michael@0: expand8888 (__m64 in, int pos)
michael@0: {
michael@0:     if (pos == 0)
michael@0: 	return _mm_unpacklo_pi8 (in, _mm_setzero_si64 ());
michael@0:     else
michael@0: 	return _mm_unpackhi_pi8 (in, _mm_setzero_si64 ());
michael@0: }
michael@0: 
michael@0: static force_inline __m64
michael@0: expandx888 (__m64 in, int pos)
michael@0: {
michael@0:     return _mm_or_si64 (expand8888 (in, pos), MC (full_alpha));
michael@0: }
michael@0: 
michael@0: static force_inline void
michael@0: expand_4x565 (__m64 vin, __m64 *vout0, __m64 *vout1, __m64 *vout2, __m64 *vout3, int full_alpha)
michael@0: {
michael@0:     __m64 v0, v1;
michael@0:     expand_4xpacked565 (vin, &v0, &v1, full_alpha);
michael@0:     *vout0 = expand8888 (v0, 0);
michael@0:     *vout1 = expand8888 (v0, 1);
michael@0:     *vout2 = expand8888 (v1, 0);
michael@0:     *vout3 = expand8888 (v1, 1);
michael@0: }
michael@0: 
michael@0: static force_inline __m64
michael@0: pack_565 (__m64 pixel, __m64 target, int pos)
michael@0: {
michael@0:     __m64 p = pixel;
michael@0:     __m64 t = target;
michael@0:     __m64 r, g, b;
michael@0: 
michael@0:     r = _mm_and_si64 (p, MC (565_r));
michael@0:     g = _mm_and_si64 (p, MC (565_g));
michael@0:     b = _mm_and_si64 (p, MC (565_b));
michael@0: 
michael@0: #ifdef USE_LOONGSON_MMI
michael@0:     r = shift (r, -(32 - 8));
michael@0:     g = shift (g, -(16 - 3));
michael@0:     b = shift (b, -(0  + 3));
michael@0: 
michael@0:     p = _mm_or_si64 (r, g);
michael@0:     p = _mm_or_si64 (p, b);
michael@0:     return loongson_insert_pi16 (t, p, pos);
michael@0: #else
michael@0:     r = shift (r, -(32 - 8) + pos * 16);
michael@0:     g = shift (g, -(16 - 3) + pos * 16);
michael@0:     b = shift (b, -(0  + 3) + pos * 16);
michael@0: 
michael@0:     if (pos == 0)
michael@0: 	t = _mm_and_si64 (t, MC (mask_0));
michael@0:     else if (pos == 1)
michael@0: 	t = _mm_and_si64 (t, MC (mask_1));
michael@0:     else if (pos == 2)
michael@0: 	t = _mm_and_si64 (t, MC (mask_2));
michael@0:     else if (pos == 3)
michael@0: 	t = _mm_and_si64 (t, MC (mask_3));
michael@0: 
michael@0:     p = _mm_or_si64 (r, t);
michael@0:     p = _mm_or_si64 (g, p);
michael@0: 
michael@0:     return _mm_or_si64 (b, p);
michael@0: #endif
michael@0: }
michael@0: 
michael@0: static force_inline __m64
michael@0: pack_4xpacked565 (__m64 a, __m64 b)
michael@0: {
michael@0:     __m64 rb0 = _mm_and_si64 (a, MC (packed_565_rb));
michael@0:     __m64 rb1 = _mm_and_si64 (b, MC (packed_565_rb));
michael@0: 
michael@0:     __m64 t0 = _mm_madd_pi16 (rb0, MC (565_pack_multiplier));
michael@0:     __m64 t1 = _mm_madd_pi16 (rb1, MC (565_pack_multiplier));
michael@0: 
michael@0:     __m64 g0 = _mm_and_si64 (a, MC (packed_565_g));
michael@0:     __m64 g1 = _mm_and_si64 (b, MC (packed_565_g));
michael@0: 
michael@0:     t0 = _mm_or_si64 (t0, g0);
michael@0:     t1 = _mm_or_si64 (t1, g1);
michael@0: 
michael@0:     t0 = shift(t0, -5);
michael@0: #ifdef USE_ARM_IWMMXT
michael@0:     t1 = shift(t1, -5);
michael@0:     return _mm_packs_pu32 (t0, t1);
michael@0: #else
michael@0:     t1 = shift(t1, -5 + 16);
michael@0:     return _mm_shuffle_pi16 (_mm_or_si64 (t0, t1), _MM_SHUFFLE (3, 1, 2, 0));
michael@0: #endif
michael@0: }
michael@0: 
michael@0: #ifndef _MSC_VER
michael@0: 
michael@0: static force_inline __m64
michael@0: pack_4x565 (__m64 v0, __m64 v1, __m64 v2, __m64 v3)
michael@0: {
michael@0:     return pack_4xpacked565 (pack8888 (v0, v1), pack8888 (v2, v3));
michael@0: }
michael@0: 
michael@0: static force_inline __m64
michael@0: pix_add_mul (__m64 x, __m64 a, __m64 y, __m64 b)
michael@0: {
michael@0:     x = pix_multiply (x, a);
michael@0:     y = pix_multiply (y, b);
michael@0: 
michael@0:     return pix_add (x, y);
michael@0: }
michael@0: 
michael@0: #else
michael@0: 
michael@0: /* MSVC only handles a "pass by register" of up to three SSE intrinsics */
michael@0: 
michael@0: #define pack_4x565(v0, v1, v2, v3) \
michael@0:     pack_4xpacked565 (pack8888 (v0, v1), pack8888 (v2, v3))
michael@0: 
michael@0: #define pix_add_mul(x, a, y, b)	 \
michael@0:     ( x = pix_multiply (x, a),	 \
michael@0:       y = pix_multiply (y, b),	 \
michael@0:       pix_add (x, y) )
michael@0: 
michael@0: #endif
michael@0: 
michael@0: /* --------------- MMX code patch for fbcompose.c --------------------- */
michael@0: 
michael@0: static force_inline __m64
michael@0: combine (const uint32_t *src, const uint32_t *mask)
michael@0: {
michael@0:     __m64 vsrc = load8888 (src);
michael@0: 
michael@0:     if (mask)
michael@0:     {
michael@0: 	__m64 m = load8888 (mask);
michael@0: 
michael@0: 	m = expand_alpha (m);
michael@0: 	vsrc = pix_multiply (vsrc, m);
michael@0:     }
michael@0: 
michael@0:     return vsrc;
michael@0: }
michael@0: 
michael@0: static force_inline __m64
michael@0: core_combine_over_u_pixel_mmx (__m64 vsrc, __m64 vdst)
michael@0: {
michael@0:     vsrc = _mm_unpacklo_pi8 (vsrc, _mm_setzero_si64 ());
michael@0: 
michael@0:     if (is_opaque (vsrc))
michael@0:     {
michael@0: 	return vsrc;
michael@0:     }
michael@0:     else if (!is_zero (vsrc))
michael@0:     {
michael@0: 	return over (vsrc, expand_alpha (vsrc),
michael@0: 		     _mm_unpacklo_pi8 (vdst, _mm_setzero_si64 ()));
michael@0:     }
michael@0: 
michael@0:     return _mm_unpacklo_pi8 (vdst, _mm_setzero_si64 ());
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_combine_over_u (pixman_implementation_t *imp,
michael@0:                     pixman_op_t              op,
michael@0:                     uint32_t *               dest,
michael@0:                     const uint32_t *         src,
michael@0:                     const uint32_t *         mask,
michael@0:                     int                      width)
michael@0: {
michael@0:     const uint32_t *end = dest + width;
michael@0: 
michael@0:     while (dest < end)
michael@0:     {
michael@0: 	__m64 vsrc = combine (src, mask);
michael@0: 
michael@0: 	if (is_opaque (vsrc))
michael@0: 	{
michael@0: 	    store8888 (dest, vsrc);
michael@0: 	}
michael@0: 	else if (!is_zero (vsrc))
michael@0: 	{
michael@0: 	    __m64 sa = expand_alpha (vsrc);
michael@0: 	    store8888 (dest, over (vsrc, sa, load8888 (dest)));
michael@0: 	}
michael@0: 
michael@0: 	++dest;
michael@0: 	++src;
michael@0: 	if (mask)
michael@0: 	    ++mask;
michael@0:     }
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_combine_over_reverse_u (pixman_implementation_t *imp,
michael@0:                             pixman_op_t              op,
michael@0:                             uint32_t *               dest,
michael@0:                             const uint32_t *         src,
michael@0:                             const uint32_t *         mask,
michael@0:                             int                      width)
michael@0: {
michael@0:     const uint32_t *end = dest + width;
michael@0: 
michael@0:     while (dest < end)
michael@0:     {
michael@0: 	__m64 d, da;
michael@0: 	__m64 s = combine (src, mask);
michael@0: 
michael@0: 	d = load8888 (dest);
michael@0: 	da = expand_alpha (d);
michael@0: 	store8888 (dest, over (d, da, s));
michael@0: 
michael@0: 	++dest;
michael@0: 	++src;
michael@0: 	if (mask)
michael@0: 	    mask++;
michael@0:     }
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_combine_in_u (pixman_implementation_t *imp,
michael@0:                   pixman_op_t              op,
michael@0:                   uint32_t *               dest,
michael@0:                   const uint32_t *         src,
michael@0:                   const uint32_t *         mask,
michael@0:                   int                      width)
michael@0: {
michael@0:     const uint32_t *end = dest + width;
michael@0: 
michael@0:     while (dest < end)
michael@0:     {
michael@0: 	__m64 a;
michael@0: 	__m64 x = combine (src, mask);
michael@0: 
michael@0: 	a = load8888 (dest);
michael@0: 	a = expand_alpha (a);
michael@0: 	x = pix_multiply (x, a);
michael@0: 
michael@0: 	store8888 (dest, x);
michael@0: 
michael@0: 	++dest;
michael@0: 	++src;
michael@0: 	if (mask)
michael@0: 	    mask++;
michael@0:     }
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_combine_in_reverse_u (pixman_implementation_t *imp,
michael@0:                           pixman_op_t              op,
michael@0:                           uint32_t *               dest,
michael@0:                           const uint32_t *         src,
michael@0:                           const uint32_t *         mask,
michael@0:                           int                      width)
michael@0: {
michael@0:     const uint32_t *end = dest + width;
michael@0: 
michael@0:     while (dest < end)
michael@0:     {
michael@0: 	__m64 a = combine (src, mask);
michael@0: 	__m64 x;
michael@0: 
michael@0: 	x = load8888 (dest);
michael@0: 	a = expand_alpha (a);
michael@0: 	x = pix_multiply (x, a);
michael@0: 	store8888 (dest, x);
michael@0: 
michael@0: 	++dest;
michael@0: 	++src;
michael@0: 	if (mask)
michael@0: 	    mask++;
michael@0:     }
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_combine_out_u (pixman_implementation_t *imp,
michael@0:                    pixman_op_t              op,
michael@0:                    uint32_t *               dest,
michael@0:                    const uint32_t *         src,
michael@0:                    const uint32_t *         mask,
michael@0:                    int                      width)
michael@0: {
michael@0:     const uint32_t *end = dest + width;
michael@0: 
michael@0:     while (dest < end)
michael@0:     {
michael@0: 	__m64 a;
michael@0: 	__m64 x = combine (src, mask);
michael@0: 
michael@0: 	a = load8888 (dest);
michael@0: 	a = expand_alpha (a);
michael@0: 	a = negate (a);
michael@0: 	x = pix_multiply (x, a);
michael@0: 	store8888 (dest, x);
michael@0: 
michael@0: 	++dest;
michael@0: 	++src;
michael@0: 	if (mask)
michael@0: 	    mask++;
michael@0:     }
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_combine_out_reverse_u (pixman_implementation_t *imp,
michael@0:                            pixman_op_t              op,
michael@0:                            uint32_t *               dest,
michael@0:                            const uint32_t *         src,
michael@0:                            const uint32_t *         mask,
michael@0:                            int                      width)
michael@0: {
michael@0:     const uint32_t *end = dest + width;
michael@0: 
michael@0:     while (dest < end)
michael@0:     {
michael@0: 	__m64 a = combine (src, mask);
michael@0: 	__m64 x;
michael@0: 
michael@0: 	x = load8888 (dest);
michael@0: 	a = expand_alpha (a);
michael@0: 	a = negate (a);
michael@0: 	x = pix_multiply (x, a);
michael@0: 
michael@0: 	store8888 (dest, x);
michael@0: 
michael@0: 	++dest;
michael@0: 	++src;
michael@0: 	if (mask)
michael@0: 	    mask++;
michael@0:     }
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_combine_atop_u (pixman_implementation_t *imp,
michael@0:                     pixman_op_t              op,
michael@0:                     uint32_t *               dest,
michael@0:                     const uint32_t *         src,
michael@0:                     const uint32_t *         mask,
michael@0:                     int                      width)
michael@0: {
michael@0:     const uint32_t *end = dest + width;
michael@0: 
michael@0:     while (dest < end)
michael@0:     {
michael@0: 	__m64 da, d, sia;
michael@0: 	__m64 s = combine (src, mask);
michael@0: 
michael@0: 	d = load8888 (dest);
michael@0: 	sia = expand_alpha (s);
michael@0: 	sia = negate (sia);
michael@0: 	da = expand_alpha (d);
michael@0: 	s = pix_add_mul (s, da, d, sia);
michael@0: 	store8888 (dest, s);
michael@0: 
michael@0: 	++dest;
michael@0: 	++src;
michael@0: 	if (mask)
michael@0: 	    mask++;
michael@0:     }
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_combine_atop_reverse_u (pixman_implementation_t *imp,
michael@0:                             pixman_op_t              op,
michael@0:                             uint32_t *               dest,
michael@0:                             const uint32_t *         src,
michael@0:                             const uint32_t *         mask,
michael@0:                             int                      width)
michael@0: {
michael@0:     const uint32_t *end;
michael@0: 
michael@0:     end = dest + width;
michael@0: 
michael@0:     while (dest < end)
michael@0:     {
michael@0: 	__m64 dia, d, sa;
michael@0: 	__m64 s = combine (src, mask);
michael@0: 
michael@0: 	d = load8888 (dest);
michael@0: 	sa = expand_alpha (s);
michael@0: 	dia = expand_alpha (d);
michael@0: 	dia = negate (dia);
michael@0: 	s = pix_add_mul (s, dia, d, sa);
michael@0: 	store8888 (dest, s);
michael@0: 
michael@0: 	++dest;
michael@0: 	++src;
michael@0: 	if (mask)
michael@0: 	    mask++;
michael@0:     }
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_combine_xor_u (pixman_implementation_t *imp,
michael@0:                    pixman_op_t              op,
michael@0:                    uint32_t *               dest,
michael@0:                    const uint32_t *         src,
michael@0:                    const uint32_t *         mask,
michael@0:                    int                      width)
michael@0: {
michael@0:     const uint32_t *end = dest + width;
michael@0: 
michael@0:     while (dest < end)
michael@0:     {
michael@0: 	__m64 dia, d, sia;
michael@0: 	__m64 s = combine (src, mask);
michael@0: 
michael@0: 	d = load8888 (dest);
michael@0: 	sia = expand_alpha (s);
michael@0: 	dia = expand_alpha (d);
michael@0: 	sia = negate (sia);
michael@0: 	dia = negate (dia);
michael@0: 	s = pix_add_mul (s, dia, d, sia);
michael@0: 	store8888 (dest, s);
michael@0: 
michael@0: 	++dest;
michael@0: 	++src;
michael@0: 	if (mask)
michael@0: 	    mask++;
michael@0:     }
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_combine_add_u (pixman_implementation_t *imp,
michael@0:                    pixman_op_t              op,
michael@0:                    uint32_t *               dest,
michael@0:                    const uint32_t *         src,
michael@0:                    const uint32_t *         mask,
michael@0:                    int                      width)
michael@0: {
michael@0:     const uint32_t *end = dest + width;
michael@0: 
michael@0:     while (dest < end)
michael@0:     {
michael@0: 	__m64 d;
michael@0: 	__m64 s = combine (src, mask);
michael@0: 
michael@0: 	d = load8888 (dest);
michael@0: 	s = pix_add (s, d);
michael@0: 	store8888 (dest, s);
michael@0: 
michael@0: 	++dest;
michael@0: 	++src;
michael@0: 	if (mask)
michael@0: 	    mask++;
michael@0:     }
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_combine_saturate_u (pixman_implementation_t *imp,
michael@0:                         pixman_op_t              op,
michael@0:                         uint32_t *               dest,
michael@0:                         const uint32_t *         src,
michael@0:                         const uint32_t *         mask,
michael@0:                         int                      width)
michael@0: {
michael@0:     const uint32_t *end = dest + width;
michael@0: 
michael@0:     while (dest < end)
michael@0:     {
michael@0: 	uint32_t s, sa, da;
michael@0: 	uint32_t d = *dest;
michael@0: 	__m64 ms = combine (src, mask);
michael@0: 	__m64 md = load8888 (dest);
michael@0: 
michael@0: 	store8888(&s, ms);
michael@0: 	da = ~d >> 24;
michael@0: 	sa = s >> 24;
michael@0: 
michael@0: 	if (sa > da)
michael@0: 	{
michael@0: 	    uint32_t quot = DIV_UN8 (da, sa) << 24;
michael@0: 	    __m64 msa = load8888 (&quot);
michael@0: 	    msa = expand_alpha (msa);
michael@0: 	    ms = pix_multiply (ms, msa);
michael@0: 	}
michael@0: 
michael@0: 	md = pix_add (md, ms);
michael@0: 	store8888 (dest, md);
michael@0: 
michael@0: 	++src;
michael@0: 	++dest;
michael@0: 	if (mask)
michael@0: 	    mask++;
michael@0:     }
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_combine_src_ca (pixman_implementation_t *imp,
michael@0:                     pixman_op_t              op,
michael@0:                     uint32_t *               dest,
michael@0:                     const uint32_t *         src,
michael@0:                     const uint32_t *         mask,
michael@0:                     int                      width)
michael@0: {
michael@0:     const uint32_t *end = src + width;
michael@0: 
michael@0:     while (src < end)
michael@0:     {
michael@0: 	__m64 a = load8888 (mask);
michael@0: 	__m64 s = load8888 (src);
michael@0: 
michael@0: 	s = pix_multiply (s, a);
michael@0: 	store8888 (dest, s);
michael@0: 
michael@0: 	++src;
michael@0: 	++mask;
michael@0: 	++dest;
michael@0:     }
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_combine_over_ca (pixman_implementation_t *imp,
michael@0:                      pixman_op_t              op,
michael@0:                      uint32_t *               dest,
michael@0:                      const uint32_t *         src,
michael@0:                      const uint32_t *         mask,
michael@0:                      int                      width)
michael@0: {
michael@0:     const uint32_t *end = src + width;
michael@0: 
michael@0:     while (src < end)
michael@0:     {
michael@0: 	__m64 a = load8888 (mask);
michael@0: 	__m64 s = load8888 (src);
michael@0: 	__m64 d = load8888 (dest);
michael@0: 	__m64 sa = expand_alpha (s);
michael@0: 
michael@0: 	store8888 (dest, in_over (s, sa, a, d));
michael@0: 
michael@0: 	++src;
michael@0: 	++dest;
michael@0: 	++mask;
michael@0:     }
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_combine_over_reverse_ca (pixman_implementation_t *imp,
michael@0:                              pixman_op_t              op,
michael@0:                              uint32_t *               dest,
michael@0:                              const uint32_t *         src,
michael@0:                              const uint32_t *         mask,
michael@0:                              int                      width)
michael@0: {
michael@0:     const uint32_t *end = src + width;
michael@0: 
michael@0:     while (src < end)
michael@0:     {
michael@0: 	__m64 a = load8888 (mask);
michael@0: 	__m64 s = load8888 (src);
michael@0: 	__m64 d = load8888 (dest);
michael@0: 	__m64 da = expand_alpha (d);
michael@0: 
michael@0: 	store8888 (dest, over (d, da, in (s, a)));
michael@0: 
michael@0: 	++src;
michael@0: 	++dest;
michael@0: 	++mask;
michael@0:     }
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_combine_in_ca (pixman_implementation_t *imp,
michael@0:                    pixman_op_t              op,
michael@0:                    uint32_t *               dest,
michael@0:                    const uint32_t *         src,
michael@0:                    const uint32_t *         mask,
michael@0:                    int                      width)
michael@0: {
michael@0:     const uint32_t *end = src + width;
michael@0: 
michael@0:     while (src < end)
michael@0:     {
michael@0: 	__m64 a = load8888 (mask);
michael@0: 	__m64 s = load8888 (src);
michael@0: 	__m64 d = load8888 (dest);
michael@0: 	__m64 da = expand_alpha (d);
michael@0: 
michael@0: 	s = pix_multiply (s, a);
michael@0: 	s = pix_multiply (s, da);
michael@0: 	store8888 (dest, s);
michael@0: 
michael@0: 	++src;
michael@0: 	++dest;
michael@0: 	++mask;
michael@0:     }
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_combine_in_reverse_ca (pixman_implementation_t *imp,
michael@0:                            pixman_op_t              op,
michael@0:                            uint32_t *               dest,
michael@0:                            const uint32_t *         src,
michael@0:                            const uint32_t *         mask,
michael@0:                            int                      width)
michael@0: {
michael@0:     const uint32_t *end = src + width;
michael@0: 
michael@0:     while (src < end)
michael@0:     {
michael@0: 	__m64 a = load8888 (mask);
michael@0: 	__m64 s = load8888 (src);
michael@0: 	__m64 d = load8888 (dest);
michael@0: 	__m64 sa = expand_alpha (s);
michael@0: 
michael@0: 	a = pix_multiply (a, sa);
michael@0: 	d = pix_multiply (d, a);
michael@0: 	store8888 (dest, d);
michael@0: 
michael@0: 	++src;
michael@0: 	++dest;
michael@0: 	++mask;
michael@0:     }
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_combine_out_ca (pixman_implementation_t *imp,
michael@0:                     pixman_op_t              op,
michael@0:                     uint32_t *               dest,
michael@0:                     const uint32_t *         src,
michael@0:                     const uint32_t *         mask,
michael@0:                     int                      width)
michael@0: {
michael@0:     const uint32_t *end = src + width;
michael@0: 
michael@0:     while (src < end)
michael@0:     {
michael@0: 	__m64 a = load8888 (mask);
michael@0: 	__m64 s = load8888 (src);
michael@0: 	__m64 d = load8888 (dest);
michael@0: 	__m64 da = expand_alpha (d);
michael@0: 
michael@0: 	da = negate (da);
michael@0: 	s = pix_multiply (s, a);
michael@0: 	s = pix_multiply (s, da);
michael@0: 	store8888 (dest, s);
michael@0: 
michael@0: 	++src;
michael@0: 	++dest;
michael@0: 	++mask;
michael@0:     }
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_combine_out_reverse_ca (pixman_implementation_t *imp,
michael@0:                             pixman_op_t              op,
michael@0:                             uint32_t *               dest,
michael@0:                             const uint32_t *         src,
michael@0:                             const uint32_t *         mask,
michael@0:                             int                      width)
michael@0: {
michael@0:     const uint32_t *end = src + width;
michael@0: 
michael@0:     while (src < end)
michael@0:     {
michael@0: 	__m64 a = load8888 (mask);
michael@0: 	__m64 s = load8888 (src);
michael@0: 	__m64 d = load8888 (dest);
michael@0: 	__m64 sa = expand_alpha (s);
michael@0: 
michael@0: 	a = pix_multiply (a, sa);
michael@0: 	a = negate (a);
michael@0: 	d = pix_multiply (d, a);
michael@0: 	store8888 (dest, d);
michael@0: 
michael@0: 	++src;
michael@0: 	++dest;
michael@0: 	++mask;
michael@0:     }
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_combine_atop_ca (pixman_implementation_t *imp,
michael@0:                      pixman_op_t              op,
michael@0:                      uint32_t *               dest,
michael@0:                      const uint32_t *         src,
michael@0:                      const uint32_t *         mask,
michael@0:                      int                      width)
michael@0: {
michael@0:     const uint32_t *end = src + width;
michael@0: 
michael@0:     while (src < end)
michael@0:     {
michael@0: 	__m64 a = load8888 (mask);
michael@0: 	__m64 s = load8888 (src);
michael@0: 	__m64 d = load8888 (dest);
michael@0: 	__m64 da = expand_alpha (d);
michael@0: 	__m64 sa = expand_alpha (s);
michael@0: 
michael@0: 	s = pix_multiply (s, a);
michael@0: 	a = pix_multiply (a, sa);
michael@0: 	a = negate (a);
michael@0: 	d = pix_add_mul (d, a, s, da);
michael@0: 	store8888 (dest, d);
michael@0: 
michael@0: 	++src;
michael@0: 	++dest;
michael@0: 	++mask;
michael@0:     }
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_combine_atop_reverse_ca (pixman_implementation_t *imp,
michael@0:                              pixman_op_t              op,
michael@0:                              uint32_t *               dest,
michael@0:                              const uint32_t *         src,
michael@0:                              const uint32_t *         mask,
michael@0:                              int                      width)
michael@0: {
michael@0:     const uint32_t *end = src + width;
michael@0: 
michael@0:     while (src < end)
michael@0:     {
michael@0: 	__m64 a = load8888 (mask);
michael@0: 	__m64 s = load8888 (src);
michael@0: 	__m64 d = load8888 (dest);
michael@0: 	__m64 da = expand_alpha (d);
michael@0: 	__m64 sa = expand_alpha (s);
michael@0: 
michael@0: 	s = pix_multiply (s, a);
michael@0: 	a = pix_multiply (a, sa);
michael@0: 	da = negate (da);
michael@0: 	d = pix_add_mul (d, a, s, da);
michael@0: 	store8888 (dest, d);
michael@0: 
michael@0: 	++src;
michael@0: 	++dest;
michael@0: 	++mask;
michael@0:     }
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_combine_xor_ca (pixman_implementation_t *imp,
michael@0:                     pixman_op_t              op,
michael@0:                     uint32_t *               dest,
michael@0:                     const uint32_t *         src,
michael@0:                     const uint32_t *         mask,
michael@0:                     int                      width)
michael@0: {
michael@0:     const uint32_t *end = src + width;
michael@0: 
michael@0:     while (src < end)
michael@0:     {
michael@0: 	__m64 a = load8888 (mask);
michael@0: 	__m64 s = load8888 (src);
michael@0: 	__m64 d = load8888 (dest);
michael@0: 	__m64 da = expand_alpha (d);
michael@0: 	__m64 sa = expand_alpha (s);
michael@0: 
michael@0: 	s = pix_multiply (s, a);
michael@0: 	a = pix_multiply (a, sa);
michael@0: 	da = negate (da);
michael@0: 	a = negate (a);
michael@0: 	d = pix_add_mul (d, a, s, da);
michael@0: 	store8888 (dest, d);
michael@0: 
michael@0: 	++src;
michael@0: 	++dest;
michael@0: 	++mask;
michael@0:     }
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_combine_add_ca (pixman_implementation_t *imp,
michael@0:                     pixman_op_t              op,
michael@0:                     uint32_t *               dest,
michael@0:                     const uint32_t *         src,
michael@0:                     const uint32_t *         mask,
michael@0:                     int                      width)
michael@0: {
michael@0:     const uint32_t *end = src + width;
michael@0: 
michael@0:     while (src < end)
michael@0:     {
michael@0: 	__m64 a = load8888 (mask);
michael@0: 	__m64 s = load8888 (src);
michael@0: 	__m64 d = load8888 (dest);
michael@0: 
michael@0: 	s = pix_multiply (s, a);
michael@0: 	d = pix_add (s, d);
michael@0: 	store8888 (dest, d);
michael@0: 
michael@0: 	++src;
michael@0: 	++dest;
michael@0: 	++mask;
michael@0:     }
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: /* ------------- MMX code paths called from fbpict.c -------------------- */
michael@0: 
michael@0: static void
michael@0: mmx_composite_over_n_8888 (pixman_implementation_t *imp,
michael@0:                            pixman_composite_info_t *info)
michael@0: {
michael@0:     PIXMAN_COMPOSITE_ARGS (info);
michael@0:     uint32_t src;
michael@0:     uint32_t    *dst_line, *dst;
michael@0:     int32_t w;
michael@0:     int dst_stride;
michael@0:     __m64 vsrc, vsrca;
michael@0: 
michael@0:     CHECKPOINT ();
michael@0: 
michael@0:     src = _pixman_image_get_solid (imp, src_image, dest_image->bits.format);
michael@0: 
michael@0:     if (src == 0)
michael@0: 	return;
michael@0: 
michael@0:     PIXMAN_IMAGE_GET_LINE (dest_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
michael@0: 
michael@0:     vsrc = load8888 (&src);
michael@0:     vsrca = expand_alpha (vsrc);
michael@0: 
michael@0:     while (height--)
michael@0:     {
michael@0: 	dst = dst_line;
michael@0: 	dst_line += dst_stride;
michael@0: 	w = width;
michael@0: 
michael@0: 	CHECKPOINT ();
michael@0: 
michael@0: 	while (w && (uintptr_t)dst & 7)
michael@0: 	{
michael@0: 	    store8888 (dst, over (vsrc, vsrca, load8888 (dst)));
michael@0: 
michael@0: 	    w--;
michael@0: 	    dst++;
michael@0: 	}
michael@0: 
michael@0: 	while (w >= 2)
michael@0: 	{
michael@0: 	    __m64 vdest;
michael@0: 	    __m64 dest0, dest1;
michael@0: 
michael@0: 	    vdest = *(__m64 *)dst;
michael@0: 
michael@0: 	    dest0 = over (vsrc, vsrca, expand8888 (vdest, 0));
michael@0: 	    dest1 = over (vsrc, vsrca, expand8888 (vdest, 1));
michael@0: 
michael@0: 	    *(__m64 *)dst = pack8888 (dest0, dest1);
michael@0: 
michael@0: 	    dst += 2;
michael@0: 	    w -= 2;
michael@0: 	}
michael@0: 
michael@0: 	CHECKPOINT ();
michael@0: 
michael@0: 	if (w)
michael@0: 	{
michael@0: 	    store8888 (dst, over (vsrc, vsrca, load8888 (dst)));
michael@0: 	}
michael@0:     }
michael@0: 
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_composite_over_n_0565 (pixman_implementation_t *imp,
michael@0:                            pixman_composite_info_t *info)
michael@0: {
michael@0:     PIXMAN_COMPOSITE_ARGS (info);
michael@0:     uint32_t src;
michael@0:     uint16_t    *dst_line, *dst;
michael@0:     int32_t w;
michael@0:     int dst_stride;
michael@0:     __m64 vsrc, vsrca;
michael@0: 
michael@0:     CHECKPOINT ();
michael@0: 
michael@0:     src = _pixman_image_get_solid (imp, src_image, dest_image->bits.format);
michael@0: 
michael@0:     if (src == 0)
michael@0: 	return;
michael@0: 
michael@0:     PIXMAN_IMAGE_GET_LINE (dest_image, dest_x, dest_y, uint16_t, dst_stride, dst_line, 1);
michael@0: 
michael@0:     vsrc = load8888 (&src);
michael@0:     vsrca = expand_alpha (vsrc);
michael@0: 
michael@0:     while (height--)
michael@0:     {
michael@0: 	dst = dst_line;
michael@0: 	dst_line += dst_stride;
michael@0: 	w = width;
michael@0: 
michael@0: 	CHECKPOINT ();
michael@0: 
michael@0: 	while (w && (uintptr_t)dst & 7)
michael@0: 	{
michael@0: 	    uint64_t d = *dst;
michael@0: 	    __m64 vdest = expand565 (to_m64 (d), 0);
michael@0: 
michael@0: 	    vdest = pack_565 (over (vsrc, vsrca, vdest), vdest, 0);
michael@0: 	    *dst = to_uint64 (vdest);
michael@0: 
michael@0: 	    w--;
michael@0: 	    dst++;
michael@0: 	}
michael@0: 
michael@0: 	while (w >= 4)
michael@0: 	{
michael@0: 	    __m64 vdest = *(__m64 *)dst;
michael@0: 	    __m64 v0, v1, v2, v3;
michael@0: 
michael@0: 	    expand_4x565 (vdest, &v0, &v1, &v2, &v3, 0);
michael@0: 
michael@0: 	    v0 = over (vsrc, vsrca, v0);
michael@0: 	    v1 = over (vsrc, vsrca, v1);
michael@0: 	    v2 = over (vsrc, vsrca, v2);
michael@0: 	    v3 = over (vsrc, vsrca, v3);
michael@0: 
michael@0: 	    *(__m64 *)dst = pack_4x565 (v0, v1, v2, v3);
michael@0: 
michael@0: 	    dst += 4;
michael@0: 	    w -= 4;
michael@0: 	}
michael@0: 
michael@0: 	CHECKPOINT ();
michael@0: 
michael@0: 	while (w)
michael@0: 	{
michael@0: 	    uint64_t d = *dst;
michael@0: 	    __m64 vdest = expand565 (to_m64 (d), 0);
michael@0: 
michael@0: 	    vdest = pack_565 (over (vsrc, vsrca, vdest), vdest, 0);
michael@0: 	    *dst = to_uint64 (vdest);
michael@0: 
michael@0: 	    w--;
michael@0: 	    dst++;
michael@0: 	}
michael@0:     }
michael@0: 
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_composite_over_n_8888_8888_ca (pixman_implementation_t *imp,
michael@0:                                    pixman_composite_info_t *info)
michael@0: {
michael@0:     PIXMAN_COMPOSITE_ARGS (info);
michael@0:     uint32_t src;
michael@0:     uint32_t    *dst_line;
michael@0:     uint32_t    *mask_line;
michael@0:     int dst_stride, mask_stride;
michael@0:     __m64 vsrc, vsrca;
michael@0: 
michael@0:     CHECKPOINT ();
michael@0: 
michael@0:     src = _pixman_image_get_solid (imp, src_image, dest_image->bits.format);
michael@0: 
michael@0:     if (src == 0)
michael@0: 	return;
michael@0: 
michael@0:     PIXMAN_IMAGE_GET_LINE (dest_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
michael@0:     PIXMAN_IMAGE_GET_LINE (mask_image, mask_x, mask_y, uint32_t, mask_stride, mask_line, 1);
michael@0: 
michael@0:     vsrc = load8888 (&src);
michael@0:     vsrca = expand_alpha (vsrc);
michael@0: 
michael@0:     while (height--)
michael@0:     {
michael@0: 	int twidth = width;
michael@0: 	uint32_t *p = (uint32_t *)mask_line;
michael@0: 	uint32_t *q = (uint32_t *)dst_line;
michael@0: 
michael@0: 	while (twidth && (uintptr_t)q & 7)
michael@0: 	{
michael@0: 	    uint32_t m = *(uint32_t *)p;
michael@0: 
michael@0: 	    if (m)
michael@0: 	    {
michael@0: 		__m64 vdest = load8888 (q);
michael@0: 		vdest = in_over (vsrc, vsrca, load8888 (&m), vdest);
michael@0: 		store8888 (q, vdest);
michael@0: 	    }
michael@0: 
michael@0: 	    twidth--;
michael@0: 	    p++;
michael@0: 	    q++;
michael@0: 	}
michael@0: 
michael@0: 	while (twidth >= 2)
michael@0: 	{
michael@0: 	    uint32_t m0, m1;
michael@0: 	    m0 = *p;
michael@0: 	    m1 = *(p + 1);
michael@0: 
michael@0: 	    if (m0 | m1)
michael@0: 	    {
michael@0: 		__m64 dest0, dest1;
michael@0: 		__m64 vdest = *(__m64 *)q;
michael@0: 
michael@0: 		dest0 = in_over (vsrc, vsrca, load8888 (&m0),
michael@0: 		                 expand8888 (vdest, 0));
michael@0: 		dest1 = in_over (vsrc, vsrca, load8888 (&m1),
michael@0: 		                 expand8888 (vdest, 1));
michael@0: 
michael@0: 		*(__m64 *)q = pack8888 (dest0, dest1);
michael@0: 	    }
michael@0: 
michael@0: 	    p += 2;
michael@0: 	    q += 2;
michael@0: 	    twidth -= 2;
michael@0: 	}
michael@0: 
michael@0: 	if (twidth)
michael@0: 	{
michael@0: 	    uint32_t m = *(uint32_t *)p;
michael@0: 
michael@0: 	    if (m)
michael@0: 	    {
michael@0: 		__m64 vdest = load8888 (q);
michael@0: 		vdest = in_over (vsrc, vsrca, load8888 (&m), vdest);
michael@0: 		store8888 (q, vdest);
michael@0: 	    }
michael@0: 
michael@0: 	    twidth--;
michael@0: 	    p++;
michael@0: 	    q++;
michael@0: 	}
michael@0: 
michael@0: 	dst_line += dst_stride;
michael@0: 	mask_line += mask_stride;
michael@0:     }
michael@0: 
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_composite_over_8888_n_8888 (pixman_implementation_t *imp,
michael@0:                                 pixman_composite_info_t *info)
michael@0: {
michael@0:     PIXMAN_COMPOSITE_ARGS (info);
michael@0:     uint32_t    *dst_line, *dst;
michael@0:     uint32_t    *src_line, *src;
michael@0:     uint32_t mask;
michael@0:     __m64 vmask;
michael@0:     int dst_stride, src_stride;
michael@0:     int32_t w;
michael@0: 
michael@0:     CHECKPOINT ();
michael@0: 
michael@0:     PIXMAN_IMAGE_GET_LINE (dest_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
michael@0:     PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
michael@0: 
michael@0:     mask = _pixman_image_get_solid (imp, mask_image, dest_image->bits.format);
michael@0:     vmask = expand_alpha (load8888 (&mask));
michael@0: 
michael@0:     while (height--)
michael@0:     {
michael@0: 	dst = dst_line;
michael@0: 	dst_line += dst_stride;
michael@0: 	src = src_line;
michael@0: 	src_line += src_stride;
michael@0: 	w = width;
michael@0: 
michael@0: 	while (w && (uintptr_t)dst & 7)
michael@0: 	{
michael@0: 	    __m64 s = load8888 (src);
michael@0: 	    __m64 d = load8888 (dst);
michael@0: 
michael@0: 	    store8888 (dst, in_over (s, expand_alpha (s), vmask, d));
michael@0: 
michael@0: 	    w--;
michael@0: 	    dst++;
michael@0: 	    src++;
michael@0: 	}
michael@0: 
michael@0: 	while (w >= 2)
michael@0: 	{
michael@0: 	    __m64 vs = ldq_u ((__m64 *)src);
michael@0: 	    __m64 vd = *(__m64 *)dst;
michael@0: 	    __m64 vsrc0 = expand8888 (vs, 0);
michael@0: 	    __m64 vsrc1 = expand8888 (vs, 1);
michael@0: 
michael@0: 	    *(__m64 *)dst = pack8888 (
michael@0: 	        in_over (vsrc0, expand_alpha (vsrc0), vmask, expand8888 (vd, 0)),
michael@0: 	        in_over (vsrc1, expand_alpha (vsrc1), vmask, expand8888 (vd, 1)));
michael@0: 
michael@0: 	    w -= 2;
michael@0: 	    dst += 2;
michael@0: 	    src += 2;
michael@0: 	}
michael@0: 
michael@0: 	if (w)
michael@0: 	{
michael@0: 	    __m64 s = load8888 (src);
michael@0: 	    __m64 d = load8888 (dst);
michael@0: 
michael@0: 	    store8888 (dst, in_over (s, expand_alpha (s), vmask, d));
michael@0: 	}
michael@0:     }
michael@0: 
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_composite_over_x888_n_8888 (pixman_implementation_t *imp,
michael@0:                                 pixman_composite_info_t *info)
michael@0: {
michael@0:     PIXMAN_COMPOSITE_ARGS (info);
michael@0:     uint32_t *dst_line, *dst;
michael@0:     uint32_t *src_line, *src;
michael@0:     uint32_t mask;
michael@0:     __m64 vmask;
michael@0:     int dst_stride, src_stride;
michael@0:     int32_t w;
michael@0:     __m64 srca;
michael@0: 
michael@0:     CHECKPOINT ();
michael@0: 
michael@0:     PIXMAN_IMAGE_GET_LINE (dest_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
michael@0:     PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
michael@0:     mask = _pixman_image_get_solid (imp, mask_image, dest_image->bits.format);
michael@0: 
michael@0:     vmask = expand_alpha (load8888 (&mask));
michael@0:     srca = MC (4x00ff);
michael@0: 
michael@0:     while (height--)
michael@0:     {
michael@0: 	dst = dst_line;
michael@0: 	dst_line += dst_stride;
michael@0: 	src = src_line;
michael@0: 	src_line += src_stride;
michael@0: 	w = width;
michael@0: 
michael@0: 	while (w && (uintptr_t)dst & 7)
michael@0: 	{
michael@0: 	    uint32_t ssrc = *src | 0xff000000;
michael@0: 	    __m64 s = load8888 (&ssrc);
michael@0: 	    __m64 d = load8888 (dst);
michael@0: 
michael@0: 	    store8888 (dst, in_over (s, srca, vmask, d));
michael@0: 
michael@0: 	    w--;
michael@0: 	    dst++;
michael@0: 	    src++;
michael@0: 	}
michael@0: 
michael@0: 	while (w >= 16)
michael@0: 	{
michael@0: 	    __m64 vd0 = *(__m64 *)(dst + 0);
michael@0: 	    __m64 vd1 = *(__m64 *)(dst + 2);
michael@0: 	    __m64 vd2 = *(__m64 *)(dst + 4);
michael@0: 	    __m64 vd3 = *(__m64 *)(dst + 6);
michael@0: 	    __m64 vd4 = *(__m64 *)(dst + 8);
michael@0: 	    __m64 vd5 = *(__m64 *)(dst + 10);
michael@0: 	    __m64 vd6 = *(__m64 *)(dst + 12);
michael@0: 	    __m64 vd7 = *(__m64 *)(dst + 14);
michael@0: 
michael@0: 	    __m64 vs0 = ldq_u ((__m64 *)(src + 0));
michael@0: 	    __m64 vs1 = ldq_u ((__m64 *)(src + 2));
michael@0: 	    __m64 vs2 = ldq_u ((__m64 *)(src + 4));
michael@0: 	    __m64 vs3 = ldq_u ((__m64 *)(src + 6));
michael@0: 	    __m64 vs4 = ldq_u ((__m64 *)(src + 8));
michael@0: 	    __m64 vs5 = ldq_u ((__m64 *)(src + 10));
michael@0: 	    __m64 vs6 = ldq_u ((__m64 *)(src + 12));
michael@0: 	    __m64 vs7 = ldq_u ((__m64 *)(src + 14));
michael@0: 
michael@0: 	    vd0 = pack8888 (
michael@0: 	        in_over (expandx888 (vs0, 0), srca, vmask, expand8888 (vd0, 0)),
michael@0: 	        in_over (expandx888 (vs0, 1), srca, vmask, expand8888 (vd0, 1)));
michael@0: 
michael@0: 	    vd1 = pack8888 (
michael@0: 	        in_over (expandx888 (vs1, 0), srca, vmask, expand8888 (vd1, 0)),
michael@0: 	        in_over (expandx888 (vs1, 1), srca, vmask, expand8888 (vd1, 1)));
michael@0: 
michael@0: 	    vd2 = pack8888 (
michael@0: 	        in_over (expandx888 (vs2, 0), srca, vmask, expand8888 (vd2, 0)),
michael@0: 	        in_over (expandx888 (vs2, 1), srca, vmask, expand8888 (vd2, 1)));
michael@0: 
michael@0: 	    vd3 = pack8888 (
michael@0: 	        in_over (expandx888 (vs3, 0), srca, vmask, expand8888 (vd3, 0)),
michael@0: 	        in_over (expandx888 (vs3, 1), srca, vmask, expand8888 (vd3, 1)));
michael@0: 
michael@0: 	    vd4 = pack8888 (
michael@0: 	        in_over (expandx888 (vs4, 0), srca, vmask, expand8888 (vd4, 0)),
michael@0: 	        in_over (expandx888 (vs4, 1), srca, vmask, expand8888 (vd4, 1)));
michael@0: 
michael@0: 	    vd5 = pack8888 (
michael@0: 	        in_over (expandx888 (vs5, 0), srca, vmask, expand8888 (vd5, 0)),
michael@0: 	        in_over (expandx888 (vs5, 1), srca, vmask, expand8888 (vd5, 1)));
michael@0: 
michael@0: 	    vd6 = pack8888 (
michael@0: 	        in_over (expandx888 (vs6, 0), srca, vmask, expand8888 (vd6, 0)),
michael@0: 	        in_over (expandx888 (vs6, 1), srca, vmask, expand8888 (vd6, 1)));
michael@0: 
michael@0: 	    vd7 = pack8888 (
michael@0: 	        in_over (expandx888 (vs7, 0), srca, vmask, expand8888 (vd7, 0)),
michael@0: 	        in_over (expandx888 (vs7, 1), srca, vmask, expand8888 (vd7, 1)));
michael@0: 
michael@0: 	    *(__m64 *)(dst + 0) = vd0;
michael@0: 	    *(__m64 *)(dst + 2) = vd1;
michael@0: 	    *(__m64 *)(dst + 4) = vd2;
michael@0: 	    *(__m64 *)(dst + 6) = vd3;
michael@0: 	    *(__m64 *)(dst + 8) = vd4;
michael@0: 	    *(__m64 *)(dst + 10) = vd5;
michael@0: 	    *(__m64 *)(dst + 12) = vd6;
michael@0: 	    *(__m64 *)(dst + 14) = vd7;
michael@0: 
michael@0: 	    w -= 16;
michael@0: 	    dst += 16;
michael@0: 	    src += 16;
michael@0: 	}
michael@0: 
michael@0: 	while (w)
michael@0: 	{
michael@0: 	    uint32_t ssrc = *src | 0xff000000;
michael@0: 	    __m64 s = load8888 (&ssrc);
michael@0: 	    __m64 d = load8888 (dst);
michael@0: 
michael@0: 	    store8888 (dst, in_over (s, srca, vmask, d));
michael@0: 
michael@0: 	    w--;
michael@0: 	    dst++;
michael@0: 	    src++;
michael@0: 	}
michael@0:     }
michael@0: 
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_composite_over_8888_8888 (pixman_implementation_t *imp,
michael@0:                               pixman_composite_info_t *info)
michael@0: {
michael@0:     PIXMAN_COMPOSITE_ARGS (info);
michael@0:     uint32_t *dst_line, *dst;
michael@0:     uint32_t *src_line, *src;
michael@0:     uint32_t s;
michael@0:     int dst_stride, src_stride;
michael@0:     uint8_t a;
michael@0:     int32_t w;
michael@0: 
michael@0:     CHECKPOINT ();
michael@0: 
michael@0:     PIXMAN_IMAGE_GET_LINE (dest_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
michael@0:     PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
michael@0: 
michael@0:     while (height--)
michael@0:     {
michael@0: 	dst = dst_line;
michael@0: 	dst_line += dst_stride;
michael@0: 	src = src_line;
michael@0: 	src_line += src_stride;
michael@0: 	w = width;
michael@0: 
michael@0: 	while (w--)
michael@0: 	{
michael@0: 	    s = *src++;
michael@0: 	    a = s >> 24;
michael@0: 
michael@0: 	    if (a == 0xff)
michael@0: 	    {
michael@0: 		*dst = s;
michael@0: 	    }
michael@0: 	    else if (s)
michael@0: 	    {
michael@0: 		__m64 ms, sa;
michael@0: 		ms = load8888 (&s);
michael@0: 		sa = expand_alpha (ms);
michael@0: 		store8888 (dst, over (ms, sa, load8888 (dst)));
michael@0: 	    }
michael@0: 
michael@0: 	    dst++;
michael@0: 	}
michael@0:     }
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_composite_over_8888_0565 (pixman_implementation_t *imp,
michael@0:                               pixman_composite_info_t *info)
michael@0: {
michael@0:     PIXMAN_COMPOSITE_ARGS (info);
michael@0:     uint16_t    *dst_line, *dst;
michael@0:     uint32_t    *src_line, *src;
michael@0:     int dst_stride, src_stride;
michael@0:     int32_t w;
michael@0: 
michael@0:     CHECKPOINT ();
michael@0: 
michael@0:     PIXMAN_IMAGE_GET_LINE (dest_image, dest_x, dest_y, uint16_t, dst_stride, dst_line, 1);
michael@0:     PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
michael@0: 
michael@0: #if 0
michael@0:     /* FIXME */
michael@0:     assert (src_image->drawable == mask_image->drawable);
michael@0: #endif
michael@0: 
michael@0:     while (height--)
michael@0:     {
michael@0: 	dst = dst_line;
michael@0: 	dst_line += dst_stride;
michael@0: 	src = src_line;
michael@0: 	src_line += src_stride;
michael@0: 	w = width;
michael@0: 
michael@0: 	CHECKPOINT ();
michael@0: 
michael@0: 	while (w && (uintptr_t)dst & 7)
michael@0: 	{
michael@0: 	    __m64 vsrc = load8888 (src);
michael@0: 	    uint64_t d = *dst;
michael@0: 	    __m64 vdest = expand565 (to_m64 (d), 0);
michael@0: 
michael@0: 	    vdest = pack_565 (
michael@0: 		over (vsrc, expand_alpha (vsrc), vdest), vdest, 0);
michael@0: 
michael@0: 	    *dst = to_uint64 (vdest);
michael@0: 
michael@0: 	    w--;
michael@0: 	    dst++;
michael@0: 	    src++;
michael@0: 	}
michael@0: 
michael@0: 	CHECKPOINT ();
michael@0: 
michael@0: 	while (w >= 4)
michael@0: 	{
michael@0: 	    __m64 vdest = *(__m64 *)dst;
michael@0: 	    __m64 v0, v1, v2, v3;
michael@0: 	    __m64 vsrc0, vsrc1, vsrc2, vsrc3;
michael@0: 
michael@0: 	    expand_4x565 (vdest, &v0, &v1, &v2, &v3, 0);
michael@0: 
michael@0: 	    vsrc0 = load8888 ((src + 0));
michael@0: 	    vsrc1 = load8888 ((src + 1));
michael@0: 	    vsrc2 = load8888 ((src + 2));
michael@0: 	    vsrc3 = load8888 ((src + 3));
michael@0: 
michael@0: 	    v0 = over (vsrc0, expand_alpha (vsrc0), v0);
michael@0: 	    v1 = over (vsrc1, expand_alpha (vsrc1), v1);
michael@0: 	    v2 = over (vsrc2, expand_alpha (vsrc2), v2);
michael@0: 	    v3 = over (vsrc3, expand_alpha (vsrc3), v3);
michael@0: 
michael@0: 	    *(__m64 *)dst = pack_4x565 (v0, v1, v2, v3);
michael@0: 
michael@0: 	    w -= 4;
michael@0: 	    dst += 4;
michael@0: 	    src += 4;
michael@0: 	}
michael@0: 
michael@0: 	CHECKPOINT ();
michael@0: 
michael@0: 	while (w)
michael@0: 	{
michael@0: 	    __m64 vsrc = load8888 (src);
michael@0: 	    uint64_t d = *dst;
michael@0: 	    __m64 vdest = expand565 (to_m64 (d), 0);
michael@0: 
michael@0: 	    vdest = pack_565 (over (vsrc, expand_alpha (vsrc), vdest), vdest, 0);
michael@0: 
michael@0: 	    *dst = to_uint64 (vdest);
michael@0: 
michael@0: 	    w--;
michael@0: 	    dst++;
michael@0: 	    src++;
michael@0: 	}
michael@0:     }
michael@0: 
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_composite_over_n_8_8888 (pixman_implementation_t *imp,
michael@0:                              pixman_composite_info_t *info)
michael@0: {
michael@0:     PIXMAN_COMPOSITE_ARGS (info);
michael@0:     uint32_t src, srca;
michael@0:     uint32_t *dst_line, *dst;
michael@0:     uint8_t *mask_line, *mask;
michael@0:     int dst_stride, mask_stride;
michael@0:     int32_t w;
michael@0:     __m64 vsrc, vsrca;
michael@0:     uint64_t srcsrc;
michael@0: 
michael@0:     CHECKPOINT ();
michael@0: 
michael@0:     src = _pixman_image_get_solid (imp, src_image, dest_image->bits.format);
michael@0: 
michael@0:     srca = src >> 24;
michael@0:     if (src == 0)
michael@0: 	return;
michael@0: 
michael@0:     srcsrc = (uint64_t)src << 32 | src;
michael@0: 
michael@0:     PIXMAN_IMAGE_GET_LINE (dest_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
michael@0:     PIXMAN_IMAGE_GET_LINE (mask_image, mask_x, mask_y, uint8_t, mask_stride, mask_line, 1);
michael@0: 
michael@0:     vsrc = load8888 (&src);
michael@0:     vsrca = expand_alpha (vsrc);
michael@0: 
michael@0:     while (height--)
michael@0:     {
michael@0: 	dst = dst_line;
michael@0: 	dst_line += dst_stride;
michael@0: 	mask = mask_line;
michael@0: 	mask_line += mask_stride;
michael@0: 	w = width;
michael@0: 
michael@0: 	CHECKPOINT ();
michael@0: 
michael@0: 	while (w && (uintptr_t)dst & 7)
michael@0: 	{
michael@0: 	    uint64_t m = *mask;
michael@0: 
michael@0: 	    if (m)
michael@0: 	    {
michael@0: 		__m64 vdest = in_over (vsrc, vsrca,
michael@0: 				       expand_alpha_rev (to_m64 (m)),
michael@0: 				       load8888 (dst));
michael@0: 
michael@0: 		store8888 (dst, vdest);
michael@0: 	    }
michael@0: 
michael@0: 	    w--;
michael@0: 	    mask++;
michael@0: 	    dst++;
michael@0: 	}
michael@0: 
michael@0: 	CHECKPOINT ();
michael@0: 
michael@0: 	while (w >= 2)
michael@0: 	{
michael@0: 	    uint64_t m0, m1;
michael@0: 
michael@0: 	    m0 = *mask;
michael@0: 	    m1 = *(mask + 1);
michael@0: 
michael@0: 	    if (srca == 0xff && (m0 & m1) == 0xff)
michael@0: 	    {
michael@0: 		*(uint64_t *)dst = srcsrc;
michael@0: 	    }
michael@0: 	    else if (m0 | m1)
michael@0: 	    {
michael@0: 		__m64 vdest;
michael@0: 		__m64 dest0, dest1;
michael@0: 
michael@0: 		vdest = *(__m64 *)dst;
michael@0: 
michael@0: 		dest0 = in_over (vsrc, vsrca, expand_alpha_rev (to_m64 (m0)),
michael@0: 				 expand8888 (vdest, 0));
michael@0: 		dest1 = in_over (vsrc, vsrca, expand_alpha_rev (to_m64 (m1)),
michael@0: 				 expand8888 (vdest, 1));
michael@0: 
michael@0: 		*(__m64 *)dst = pack8888 (dest0, dest1);
michael@0: 	    }
michael@0: 
michael@0: 	    mask += 2;
michael@0: 	    dst += 2;
michael@0: 	    w -= 2;
michael@0: 	}
michael@0: 
michael@0: 	CHECKPOINT ();
michael@0: 
michael@0: 	if (w)
michael@0: 	{
michael@0: 	    uint64_t m = *mask;
michael@0: 
michael@0: 	    if (m)
michael@0: 	    {
michael@0: 		__m64 vdest = load8888 (dst);
michael@0: 
michael@0: 		vdest = in_over (
michael@0: 		    vsrc, vsrca, expand_alpha_rev (to_m64 (m)), vdest);
michael@0: 		store8888 (dst, vdest);
michael@0: 	    }
michael@0: 	}
michael@0:     }
michael@0: 
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static pixman_bool_t
michael@0: mmx_fill (pixman_implementation_t *imp,
michael@0:           uint32_t *               bits,
michael@0:           int                      stride,
michael@0:           int                      bpp,
michael@0:           int                      x,
michael@0:           int                      y,
michael@0:           int                      width,
michael@0:           int                      height,
michael@0:           uint32_t		   filler)
michael@0: {
michael@0:     uint64_t fill;
michael@0:     __m64 vfill;
michael@0:     uint32_t byte_width;
michael@0:     uint8_t     *byte_line;
michael@0: 
michael@0: #if defined __GNUC__ && defined USE_X86_MMX
michael@0:     __m64 v1, v2, v3, v4, v5, v6, v7;
michael@0: #endif
michael@0: 
michael@0:     if (bpp != 16 && bpp != 32 && bpp != 8)
michael@0: 	return FALSE;
michael@0: 
michael@0:     if (bpp == 8)
michael@0:     {
michael@0: 	stride = stride * (int) sizeof (uint32_t) / 1;
michael@0: 	byte_line = (uint8_t *)(((uint8_t *)bits) + stride * y + x);
michael@0: 	byte_width = width;
michael@0: 	stride *= 1;
michael@0:         filler = (filler & 0xff) * 0x01010101;
michael@0:     }
michael@0:     else if (bpp == 16)
michael@0:     {
michael@0: 	stride = stride * (int) sizeof (uint32_t) / 2;
michael@0: 	byte_line = (uint8_t *)(((uint16_t *)bits) + stride * y + x);
michael@0: 	byte_width = 2 * width;
michael@0: 	stride *= 2;
michael@0:         filler = (filler & 0xffff) * 0x00010001;
michael@0:     }
michael@0:     else
michael@0:     {
michael@0: 	stride = stride * (int) sizeof (uint32_t) / 4;
michael@0: 	byte_line = (uint8_t *)(((uint32_t *)bits) + stride * y + x);
michael@0: 	byte_width = 4 * width;
michael@0: 	stride *= 4;
michael@0:     }
michael@0: 
michael@0:     fill = ((uint64_t)filler << 32) | filler;
michael@0:     vfill = to_m64 (fill);
michael@0: 
michael@0: #if defined __GNUC__ && defined USE_X86_MMX
michael@0:     __asm__ (
michael@0:         "movq		%7,	%0\n"
michael@0:         "movq		%7,	%1\n"
michael@0:         "movq		%7,	%2\n"
michael@0:         "movq		%7,	%3\n"
michael@0:         "movq		%7,	%4\n"
michael@0:         "movq		%7,	%5\n"
michael@0:         "movq		%7,	%6\n"
michael@0: 	: "=&y" (v1), "=&y" (v2), "=&y" (v3),
michael@0: 	  "=&y" (v4), "=&y" (v5), "=&y" (v6), "=y" (v7)
michael@0: 	: "y" (vfill));
michael@0: #endif
michael@0: 
michael@0:     while (height--)
michael@0:     {
michael@0: 	int w;
michael@0: 	uint8_t *d = byte_line;
michael@0: 
michael@0: 	byte_line += stride;
michael@0: 	w = byte_width;
michael@0: 
michael@0: 	if (w >= 1 && ((uintptr_t)d & 1))
michael@0: 	{
michael@0: 	    *(uint8_t *)d = (filler & 0xff);
michael@0: 	    w--;
michael@0: 	    d++;
michael@0: 	}
michael@0: 
michael@0: 	if (w >= 2 && ((uintptr_t)d & 3))
michael@0: 	{
michael@0: 	    *(uint16_t *)d = filler;
michael@0: 	    w -= 2;
michael@0: 	    d += 2;
michael@0: 	}
michael@0: 
michael@0: 	while (w >= 4 && ((uintptr_t)d & 7))
michael@0: 	{
michael@0: 	    *(uint32_t *)d = filler;
michael@0: 
michael@0: 	    w -= 4;
michael@0: 	    d += 4;
michael@0: 	}
michael@0: 
michael@0: 	while (w >= 64)
michael@0: 	{
michael@0: #if defined __GNUC__ && defined USE_X86_MMX
michael@0: 	    __asm__ (
michael@0: 	        "movq	%1,	  (%0)\n"
michael@0: 	        "movq	%2,	 8(%0)\n"
michael@0: 	        "movq	%3,	16(%0)\n"
michael@0: 	        "movq	%4,	24(%0)\n"
michael@0: 	        "movq	%5,	32(%0)\n"
michael@0: 	        "movq	%6,	40(%0)\n"
michael@0: 	        "movq	%7,	48(%0)\n"
michael@0: 	        "movq	%8,	56(%0)\n"
michael@0: 		:
michael@0: 		: "r" (d),
michael@0: 		  "y" (vfill), "y" (v1), "y" (v2), "y" (v3),
michael@0: 		  "y" (v4), "y" (v5), "y" (v6), "y" (v7)
michael@0: 		: "memory");
michael@0: #else
michael@0: 	    *(__m64*) (d +  0) = vfill;
michael@0: 	    *(__m64*) (d +  8) = vfill;
michael@0: 	    *(__m64*) (d + 16) = vfill;
michael@0: 	    *(__m64*) (d + 24) = vfill;
michael@0: 	    *(__m64*) (d + 32) = vfill;
michael@0: 	    *(__m64*) (d + 40) = vfill;
michael@0: 	    *(__m64*) (d + 48) = vfill;
michael@0: 	    *(__m64*) (d + 56) = vfill;
michael@0: #endif
michael@0: 	    w -= 64;
michael@0: 	    d += 64;
michael@0: 	}
michael@0: 
michael@0: 	while (w >= 4)
michael@0: 	{
michael@0: 	    *(uint32_t *)d = filler;
michael@0: 
michael@0: 	    w -= 4;
michael@0: 	    d += 4;
michael@0: 	}
michael@0: 	if (w >= 2)
michael@0: 	{
michael@0: 	    *(uint16_t *)d = filler;
michael@0: 	    w -= 2;
michael@0: 	    d += 2;
michael@0: 	}
michael@0: 	if (w >= 1)
michael@0: 	{
michael@0: 	    *(uint8_t *)d = (filler & 0xff);
michael@0: 	    w--;
michael@0: 	    d++;
michael@0: 	}
michael@0: 
michael@0:     }
michael@0: 
michael@0:     _mm_empty ();
michael@0:     return TRUE;
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_composite_src_x888_0565 (pixman_implementation_t *imp,
michael@0:                              pixman_composite_info_t *info)
michael@0: {
michael@0:     PIXMAN_COMPOSITE_ARGS (info);
michael@0:     uint16_t    *dst_line, *dst;
michael@0:     uint32_t    *src_line, *src, s;
michael@0:     int dst_stride, src_stride;
michael@0:     int32_t w;
michael@0: 
michael@0:     PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
michael@0:     PIXMAN_IMAGE_GET_LINE (dest_image, dest_x, dest_y, uint16_t, dst_stride, dst_line, 1);
michael@0: 
michael@0:     while (height--)
michael@0:     {
michael@0: 	dst = dst_line;
michael@0: 	dst_line += dst_stride;
michael@0: 	src = src_line;
michael@0: 	src_line += src_stride;
michael@0: 	w = width;
michael@0: 
michael@0: 	while (w && (uintptr_t)dst & 7)
michael@0: 	{
michael@0: 	    s = *src++;
michael@0: 	    *dst = convert_8888_to_0565 (s);
michael@0: 	    dst++;
michael@0: 	    w--;
michael@0: 	}
michael@0: 
michael@0: 	while (w >= 4)
michael@0: 	{
michael@0: 	    __m64 vdest;
michael@0: 	    __m64 vsrc0 = ldq_u ((__m64 *)(src + 0));
michael@0: 	    __m64 vsrc1 = ldq_u ((__m64 *)(src + 2));
michael@0: 
michael@0: 	    vdest = pack_4xpacked565 (vsrc0, vsrc1);
michael@0: 
michael@0: 	    *(__m64 *)dst = vdest;
michael@0: 
michael@0: 	    w -= 4;
michael@0: 	    src += 4;
michael@0: 	    dst += 4;
michael@0: 	}
michael@0: 
michael@0: 	while (w)
michael@0: 	{
michael@0: 	    s = *src++;
michael@0: 	    *dst = convert_8888_to_0565 (s);
michael@0: 	    dst++;
michael@0: 	    w--;
michael@0: 	}
michael@0:     }
michael@0: 
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_composite_src_n_8_8888 (pixman_implementation_t *imp,
michael@0:                             pixman_composite_info_t *info)
michael@0: {
michael@0:     PIXMAN_COMPOSITE_ARGS (info);
michael@0:     uint32_t src, srca;
michael@0:     uint32_t    *dst_line, *dst;
michael@0:     uint8_t     *mask_line, *mask;
michael@0:     int dst_stride, mask_stride;
michael@0:     int32_t w;
michael@0:     __m64 vsrc;
michael@0:     uint64_t srcsrc;
michael@0: 
michael@0:     CHECKPOINT ();
michael@0: 
michael@0:     src = _pixman_image_get_solid (imp, src_image, dest_image->bits.format);
michael@0: 
michael@0:     srca = src >> 24;
michael@0:     if (src == 0)
michael@0:     {
michael@0: 	mmx_fill (imp, dest_image->bits.bits, dest_image->bits.rowstride,
michael@0: 		  PIXMAN_FORMAT_BPP (dest_image->bits.format),
michael@0: 		  dest_x, dest_y, width, height, 0);
michael@0: 	return;
michael@0:     }
michael@0: 
michael@0:     srcsrc = (uint64_t)src << 32 | src;
michael@0: 
michael@0:     PIXMAN_IMAGE_GET_LINE (dest_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
michael@0:     PIXMAN_IMAGE_GET_LINE (mask_image, mask_x, mask_y, uint8_t, mask_stride, mask_line, 1);
michael@0: 
michael@0:     vsrc = load8888 (&src);
michael@0: 
michael@0:     while (height--)
michael@0:     {
michael@0: 	dst = dst_line;
michael@0: 	dst_line += dst_stride;
michael@0: 	mask = mask_line;
michael@0: 	mask_line += mask_stride;
michael@0: 	w = width;
michael@0: 
michael@0: 	CHECKPOINT ();
michael@0: 
michael@0: 	while (w && (uintptr_t)dst & 7)
michael@0: 	{
michael@0: 	    uint64_t m = *mask;
michael@0: 
michael@0: 	    if (m)
michael@0: 	    {
michael@0: 		__m64 vdest = in (vsrc, expand_alpha_rev (to_m64 (m)));
michael@0: 
michael@0: 		store8888 (dst, vdest);
michael@0: 	    }
michael@0: 	    else
michael@0: 	    {
michael@0: 		*dst = 0;
michael@0: 	    }
michael@0: 
michael@0: 	    w--;
michael@0: 	    mask++;
michael@0: 	    dst++;
michael@0: 	}
michael@0: 
michael@0: 	CHECKPOINT ();
michael@0: 
michael@0: 	while (w >= 2)
michael@0: 	{
michael@0: 	    uint64_t m0, m1;
michael@0: 	    m0 = *mask;
michael@0: 	    m1 = *(mask + 1);
michael@0: 
michael@0: 	    if (srca == 0xff && (m0 & m1) == 0xff)
michael@0: 	    {
michael@0: 		*(uint64_t *)dst = srcsrc;
michael@0: 	    }
michael@0: 	    else if (m0 | m1)
michael@0: 	    {
michael@0: 		__m64 dest0, dest1;
michael@0: 
michael@0: 		dest0 = in (vsrc, expand_alpha_rev (to_m64 (m0)));
michael@0: 		dest1 = in (vsrc, expand_alpha_rev (to_m64 (m1)));
michael@0: 
michael@0: 		*(__m64 *)dst = pack8888 (dest0, dest1);
michael@0: 	    }
michael@0: 	    else
michael@0: 	    {
michael@0: 		*(uint64_t *)dst = 0;
michael@0: 	    }
michael@0: 
michael@0: 	    mask += 2;
michael@0: 	    dst += 2;
michael@0: 	    w -= 2;
michael@0: 	}
michael@0: 
michael@0: 	CHECKPOINT ();
michael@0: 
michael@0: 	if (w)
michael@0: 	{
michael@0: 	    uint64_t m = *mask;
michael@0: 
michael@0: 	    if (m)
michael@0: 	    {
michael@0: 		__m64 vdest = load8888 (dst);
michael@0: 
michael@0: 		vdest = in (vsrc, expand_alpha_rev (to_m64 (m)));
michael@0: 		store8888 (dst, vdest);
michael@0: 	    }
michael@0: 	    else
michael@0: 	    {
michael@0: 		*dst = 0;
michael@0: 	    }
michael@0: 	}
michael@0:     }
michael@0: 
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_composite_over_n_8_0565 (pixman_implementation_t *imp,
michael@0:                              pixman_composite_info_t *info)
michael@0: {
michael@0:     PIXMAN_COMPOSITE_ARGS (info);
michael@0:     uint32_t src, srca;
michael@0:     uint16_t *dst_line, *dst;
michael@0:     uint8_t *mask_line, *mask;
michael@0:     int dst_stride, mask_stride;
michael@0:     int32_t w;
michael@0:     __m64 vsrc, vsrca, tmp;
michael@0:     __m64 srcsrcsrcsrc;
michael@0: 
michael@0:     CHECKPOINT ();
michael@0: 
michael@0:     src = _pixman_image_get_solid (imp, src_image, dest_image->bits.format);
michael@0: 
michael@0:     srca = src >> 24;
michael@0:     if (src == 0)
michael@0: 	return;
michael@0: 
michael@0:     PIXMAN_IMAGE_GET_LINE (dest_image, dest_x, dest_y, uint16_t, dst_stride, dst_line, 1);
michael@0:     PIXMAN_IMAGE_GET_LINE (mask_image, mask_x, mask_y, uint8_t, mask_stride, mask_line, 1);
michael@0: 
michael@0:     vsrc = load8888 (&src);
michael@0:     vsrca = expand_alpha (vsrc);
michael@0: 
michael@0:     tmp = pack_565 (vsrc, _mm_setzero_si64 (), 0);
michael@0:     srcsrcsrcsrc = expand_alpha_rev (tmp);
michael@0: 
michael@0:     while (height--)
michael@0:     {
michael@0: 	dst = dst_line;
michael@0: 	dst_line += dst_stride;
michael@0: 	mask = mask_line;
michael@0: 	mask_line += mask_stride;
michael@0: 	w = width;
michael@0: 
michael@0: 	CHECKPOINT ();
michael@0: 
michael@0: 	while (w && (uintptr_t)dst & 7)
michael@0: 	{
michael@0: 	    uint64_t m = *mask;
michael@0: 
michael@0: 	    if (m)
michael@0: 	    {
michael@0: 		uint64_t d = *dst;
michael@0: 		__m64 vd = to_m64 (d);
michael@0: 		__m64 vdest = in_over (
michael@0: 		    vsrc, vsrca, expand_alpha_rev (to_m64 (m)), expand565 (vd, 0));
michael@0: 
michael@0: 		vd = pack_565 (vdest, _mm_setzero_si64 (), 0);
michael@0: 		*dst = to_uint64 (vd);
michael@0: 	    }
michael@0: 
michael@0: 	    w--;
michael@0: 	    mask++;
michael@0: 	    dst++;
michael@0: 	}
michael@0: 
michael@0: 	CHECKPOINT ();
michael@0: 
michael@0: 	while (w >= 4)
michael@0: 	{
michael@0: 	    uint64_t m0, m1, m2, m3;
michael@0: 	    m0 = *mask;
michael@0: 	    m1 = *(mask + 1);
michael@0: 	    m2 = *(mask + 2);
michael@0: 	    m3 = *(mask + 3);
michael@0: 
michael@0: 	    if (srca == 0xff && (m0 & m1 & m2 & m3) == 0xff)
michael@0: 	    {
michael@0: 		*(__m64 *)dst = srcsrcsrcsrc;
michael@0: 	    }
michael@0: 	    else if (m0 | m1 | m2 | m3)
michael@0: 	    {
michael@0: 		__m64 vdest = *(__m64 *)dst;
michael@0: 		__m64 v0, v1, v2, v3;
michael@0: 		__m64 vm0, vm1, vm2, vm3;
michael@0: 
michael@0: 		expand_4x565 (vdest, &v0, &v1, &v2, &v3, 0);
michael@0: 
michael@0: 		vm0 = to_m64 (m0);
michael@0: 		v0 = in_over (vsrc, vsrca, expand_alpha_rev (vm0), v0);
michael@0: 
michael@0: 		vm1 = to_m64 (m1);
michael@0: 		v1 = in_over (vsrc, vsrca, expand_alpha_rev (vm1), v1);
michael@0: 
michael@0: 		vm2 = to_m64 (m2);
michael@0: 		v2 = in_over (vsrc, vsrca, expand_alpha_rev (vm2), v2);
michael@0: 
michael@0: 		vm3 = to_m64 (m3);
michael@0: 		v3 = in_over (vsrc, vsrca, expand_alpha_rev (vm3), v3);
michael@0: 
michael@0: 		*(__m64 *)dst = pack_4x565 (v0, v1, v2, v3);;
michael@0: 	    }
michael@0: 
michael@0: 	    w -= 4;
michael@0: 	    mask += 4;
michael@0: 	    dst += 4;
michael@0: 	}
michael@0: 
michael@0: 	CHECKPOINT ();
michael@0: 
michael@0: 	while (w)
michael@0: 	{
michael@0: 	    uint64_t m = *mask;
michael@0: 
michael@0: 	    if (m)
michael@0: 	    {
michael@0: 		uint64_t d = *dst;
michael@0: 		__m64 vd = to_m64 (d);
michael@0: 		__m64 vdest = in_over (vsrc, vsrca, expand_alpha_rev (to_m64 (m)),
michael@0: 				       expand565 (vd, 0));
michael@0: 		vd = pack_565 (vdest, _mm_setzero_si64 (), 0);
michael@0: 		*dst = to_uint64 (vd);
michael@0: 	    }
michael@0: 
michael@0: 	    w--;
michael@0: 	    mask++;
michael@0: 	    dst++;
michael@0: 	}
michael@0:     }
michael@0: 
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_composite_over_pixbuf_0565 (pixman_implementation_t *imp,
michael@0:                                 pixman_composite_info_t *info)
michael@0: {
michael@0:     PIXMAN_COMPOSITE_ARGS (info);
michael@0:     uint16_t    *dst_line, *dst;
michael@0:     uint32_t    *src_line, *src;
michael@0:     int dst_stride, src_stride;
michael@0:     int32_t w;
michael@0: 
michael@0:     CHECKPOINT ();
michael@0: 
michael@0:     PIXMAN_IMAGE_GET_LINE (dest_image, dest_x, dest_y, uint16_t, dst_stride, dst_line, 1);
michael@0:     PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
michael@0: 
michael@0: #if 0
michael@0:     /* FIXME */
michael@0:     assert (src_image->drawable == mask_image->drawable);
michael@0: #endif
michael@0: 
michael@0:     while (height--)
michael@0:     {
michael@0: 	dst = dst_line;
michael@0: 	dst_line += dst_stride;
michael@0: 	src = src_line;
michael@0: 	src_line += src_stride;
michael@0: 	w = width;
michael@0: 
michael@0: 	CHECKPOINT ();
michael@0: 
michael@0: 	while (w && (uintptr_t)dst & 7)
michael@0: 	{
michael@0: 	    __m64 vsrc = load8888 (src);
michael@0: 	    uint64_t d = *dst;
michael@0: 	    __m64 vdest = expand565 (to_m64 (d), 0);
michael@0: 
michael@0: 	    vdest = pack_565 (over_rev_non_pre (vsrc, vdest), vdest, 0);
michael@0: 
michael@0: 	    *dst = to_uint64 (vdest);
michael@0: 
michael@0: 	    w--;
michael@0: 	    dst++;
michael@0: 	    src++;
michael@0: 	}
michael@0: 
michael@0: 	CHECKPOINT ();
michael@0: 
michael@0: 	while (w >= 4)
michael@0: 	{
michael@0: 	    uint32_t s0, s1, s2, s3;
michael@0: 	    unsigned char a0, a1, a2, a3;
michael@0: 
michael@0: 	    s0 = *src;
michael@0: 	    s1 = *(src + 1);
michael@0: 	    s2 = *(src + 2);
michael@0: 	    s3 = *(src + 3);
michael@0: 
michael@0: 	    a0 = (s0 >> 24);
michael@0: 	    a1 = (s1 >> 24);
michael@0: 	    a2 = (s2 >> 24);
michael@0: 	    a3 = (s3 >> 24);
michael@0: 
michael@0: 	    if ((a0 & a1 & a2 & a3) == 0xFF)
michael@0: 	    {
michael@0: 		__m64 v0 = invert_colors (load8888 (&s0));
michael@0: 		__m64 v1 = invert_colors (load8888 (&s1));
michael@0: 		__m64 v2 = invert_colors (load8888 (&s2));
michael@0: 		__m64 v3 = invert_colors (load8888 (&s3));
michael@0: 
michael@0: 		*(__m64 *)dst = pack_4x565 (v0, v1, v2, v3);
michael@0: 	    }
michael@0: 	    else if (s0 | s1 | s2 | s3)
michael@0: 	    {
michael@0: 		__m64 vdest = *(__m64 *)dst;
michael@0: 		__m64 v0, v1, v2, v3;
michael@0: 
michael@0: 		__m64 vsrc0 = load8888 (&s0);
michael@0: 		__m64 vsrc1 = load8888 (&s1);
michael@0: 		__m64 vsrc2 = load8888 (&s2);
michael@0: 		__m64 vsrc3 = load8888 (&s3);
michael@0: 
michael@0: 		expand_4x565 (vdest, &v0, &v1, &v2, &v3, 0);
michael@0: 
michael@0: 		v0 = over_rev_non_pre (vsrc0, v0);
michael@0: 		v1 = over_rev_non_pre (vsrc1, v1);
michael@0: 		v2 = over_rev_non_pre (vsrc2, v2);
michael@0: 		v3 = over_rev_non_pre (vsrc3, v3);
michael@0: 
michael@0: 		*(__m64 *)dst = pack_4x565 (v0, v1, v2, v3);
michael@0: 	    }
michael@0: 
michael@0: 	    w -= 4;
michael@0: 	    dst += 4;
michael@0: 	    src += 4;
michael@0: 	}
michael@0: 
michael@0: 	CHECKPOINT ();
michael@0: 
michael@0: 	while (w)
michael@0: 	{
michael@0: 	    __m64 vsrc = load8888 (src);
michael@0: 	    uint64_t d = *dst;
michael@0: 	    __m64 vdest = expand565 (to_m64 (d), 0);
michael@0: 
michael@0: 	    vdest = pack_565 (over_rev_non_pre (vsrc, vdest), vdest, 0);
michael@0: 
michael@0: 	    *dst = to_uint64 (vdest);
michael@0: 
michael@0: 	    w--;
michael@0: 	    dst++;
michael@0: 	    src++;
michael@0: 	}
michael@0:     }
michael@0: 
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_composite_over_pixbuf_8888 (pixman_implementation_t *imp,
michael@0:                                 pixman_composite_info_t *info)
michael@0: {
michael@0:     PIXMAN_COMPOSITE_ARGS (info);
michael@0:     uint32_t    *dst_line, *dst;
michael@0:     uint32_t    *src_line, *src;
michael@0:     int dst_stride, src_stride;
michael@0:     int32_t w;
michael@0: 
michael@0:     CHECKPOINT ();
michael@0: 
michael@0:     PIXMAN_IMAGE_GET_LINE (dest_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
michael@0:     PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
michael@0: 
michael@0: #if 0
michael@0:     /* FIXME */
michael@0:     assert (src_image->drawable == mask_image->drawable);
michael@0: #endif
michael@0: 
michael@0:     while (height--)
michael@0:     {
michael@0: 	dst = dst_line;
michael@0: 	dst_line += dst_stride;
michael@0: 	src = src_line;
michael@0: 	src_line += src_stride;
michael@0: 	w = width;
michael@0: 
michael@0: 	while (w && (uintptr_t)dst & 7)
michael@0: 	{
michael@0: 	    __m64 s = load8888 (src);
michael@0: 	    __m64 d = load8888 (dst);
michael@0: 
michael@0: 	    store8888 (dst, over_rev_non_pre (s, d));
michael@0: 
michael@0: 	    w--;
michael@0: 	    dst++;
michael@0: 	    src++;
michael@0: 	}
michael@0: 
michael@0: 	while (w >= 2)
michael@0: 	{
michael@0: 	    uint32_t s0, s1;
michael@0: 	    unsigned char a0, a1;
michael@0: 	    __m64 d0, d1;
michael@0: 
michael@0: 	    s0 = *src;
michael@0: 	    s1 = *(src + 1);
michael@0: 
michael@0: 	    a0 = (s0 >> 24);
michael@0: 	    a1 = (s1 >> 24);
michael@0: 
michael@0: 	    if ((a0 & a1) == 0xFF)
michael@0: 	    {
michael@0: 		d0 = invert_colors (load8888 (&s0));
michael@0: 		d1 = invert_colors (load8888 (&s1));
michael@0: 
michael@0: 		*(__m64 *)dst = pack8888 (d0, d1);
michael@0: 	    }
michael@0: 	    else if (s0 | s1)
michael@0: 	    {
michael@0: 		__m64 vdest = *(__m64 *)dst;
michael@0: 
michael@0: 		d0 = over_rev_non_pre (load8888 (&s0), expand8888 (vdest, 0));
michael@0: 		d1 = over_rev_non_pre (load8888 (&s1), expand8888 (vdest, 1));
michael@0: 
michael@0: 		*(__m64 *)dst = pack8888 (d0, d1);
michael@0: 	    }
michael@0: 
michael@0: 	    w -= 2;
michael@0: 	    dst += 2;
michael@0: 	    src += 2;
michael@0: 	}
michael@0: 
michael@0: 	if (w)
michael@0: 	{
michael@0: 	    __m64 s = load8888 (src);
michael@0: 	    __m64 d = load8888 (dst);
michael@0: 
michael@0: 	    store8888 (dst, over_rev_non_pre (s, d));
michael@0: 	}
michael@0:     }
michael@0: 
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_composite_over_n_8888_0565_ca (pixman_implementation_t *imp,
michael@0:                                    pixman_composite_info_t *info)
michael@0: {
michael@0:     PIXMAN_COMPOSITE_ARGS (info);
michael@0:     uint32_t src;
michael@0:     uint16_t    *dst_line;
michael@0:     uint32_t    *mask_line;
michael@0:     int dst_stride, mask_stride;
michael@0:     __m64 vsrc, vsrca;
michael@0: 
michael@0:     CHECKPOINT ();
michael@0: 
michael@0:     src = _pixman_image_get_solid (imp, src_image, dest_image->bits.format);
michael@0: 
michael@0:     if (src == 0)
michael@0: 	return;
michael@0: 
michael@0:     PIXMAN_IMAGE_GET_LINE (dest_image, dest_x, dest_y, uint16_t, dst_stride, dst_line, 1);
michael@0:     PIXMAN_IMAGE_GET_LINE (mask_image, mask_x, mask_y, uint32_t, mask_stride, mask_line, 1);
michael@0: 
michael@0:     vsrc = load8888 (&src);
michael@0:     vsrca = expand_alpha (vsrc);
michael@0: 
michael@0:     while (height--)
michael@0:     {
michael@0: 	int twidth = width;
michael@0: 	uint32_t *p = (uint32_t *)mask_line;
michael@0: 	uint16_t *q = (uint16_t *)dst_line;
michael@0: 
michael@0: 	while (twidth && ((uintptr_t)q & 7))
michael@0: 	{
michael@0: 	    uint32_t m = *(uint32_t *)p;
michael@0: 
michael@0: 	    if (m)
michael@0: 	    {
michael@0: 		uint64_t d = *q;
michael@0: 		__m64 vdest = expand565 (to_m64 (d), 0);
michael@0: 		vdest = pack_565 (in_over (vsrc, vsrca, load8888 (&m), vdest), vdest, 0);
michael@0: 		*q = to_uint64 (vdest);
michael@0: 	    }
michael@0: 
michael@0: 	    twidth--;
michael@0: 	    p++;
michael@0: 	    q++;
michael@0: 	}
michael@0: 
michael@0: 	while (twidth >= 4)
michael@0: 	{
michael@0: 	    uint32_t m0, m1, m2, m3;
michael@0: 
michael@0: 	    m0 = *p;
michael@0: 	    m1 = *(p + 1);
michael@0: 	    m2 = *(p + 2);
michael@0: 	    m3 = *(p + 3);
michael@0: 
michael@0: 	    if ((m0 | m1 | m2 | m3))
michael@0: 	    {
michael@0: 		__m64 vdest = *(__m64 *)q;
michael@0: 		__m64 v0, v1, v2, v3;
michael@0: 
michael@0: 		expand_4x565 (vdest, &v0, &v1, &v2, &v3, 0);
michael@0: 
michael@0: 		v0 = in_over (vsrc, vsrca, load8888 (&m0), v0);
michael@0: 		v1 = in_over (vsrc, vsrca, load8888 (&m1), v1);
michael@0: 		v2 = in_over (vsrc, vsrca, load8888 (&m2), v2);
michael@0: 		v3 = in_over (vsrc, vsrca, load8888 (&m3), v3);
michael@0: 
michael@0: 		*(__m64 *)q = pack_4x565 (v0, v1, v2, v3);
michael@0: 	    }
michael@0: 	    twidth -= 4;
michael@0: 	    p += 4;
michael@0: 	    q += 4;
michael@0: 	}
michael@0: 
michael@0: 	while (twidth)
michael@0: 	{
michael@0: 	    uint32_t m;
michael@0: 
michael@0: 	    m = *(uint32_t *)p;
michael@0: 	    if (m)
michael@0: 	    {
michael@0: 		uint64_t d = *q;
michael@0: 		__m64 vdest = expand565 (to_m64 (d), 0);
michael@0: 		vdest = pack_565 (in_over (vsrc, vsrca, load8888 (&m), vdest), vdest, 0);
michael@0: 		*q = to_uint64 (vdest);
michael@0: 	    }
michael@0: 
michael@0: 	    twidth--;
michael@0: 	    p++;
michael@0: 	    q++;
michael@0: 	}
michael@0: 
michael@0: 	mask_line += mask_stride;
michael@0: 	dst_line += dst_stride;
michael@0:     }
michael@0: 
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_composite_in_n_8_8 (pixman_implementation_t *imp,
michael@0:                         pixman_composite_info_t *info)
michael@0: {
michael@0:     PIXMAN_COMPOSITE_ARGS (info);
michael@0:     uint8_t *dst_line, *dst;
michael@0:     uint8_t *mask_line, *mask;
michael@0:     int dst_stride, mask_stride;
michael@0:     int32_t w;
michael@0:     uint32_t src;
michael@0:     uint8_t sa;
michael@0:     __m64 vsrc, vsrca;
michael@0: 
michael@0:     PIXMAN_IMAGE_GET_LINE (dest_image, dest_x, dest_y, uint8_t, dst_stride, dst_line, 1);
michael@0:     PIXMAN_IMAGE_GET_LINE (mask_image, mask_x, mask_y, uint8_t, mask_stride, mask_line, 1);
michael@0: 
michael@0:     src = _pixman_image_get_solid (imp, src_image, dest_image->bits.format);
michael@0: 
michael@0:     sa = src >> 24;
michael@0: 
michael@0:     vsrc = load8888 (&src);
michael@0:     vsrca = expand_alpha (vsrc);
michael@0: 
michael@0:     while (height--)
michael@0:     {
michael@0: 	dst = dst_line;
michael@0: 	dst_line += dst_stride;
michael@0: 	mask = mask_line;
michael@0: 	mask_line += mask_stride;
michael@0: 	w = width;
michael@0: 
michael@0: 	while (w && (uintptr_t)dst & 7)
michael@0: 	{
michael@0: 	    uint16_t tmp;
michael@0: 	    uint8_t a;
michael@0: 	    uint32_t m, d;
michael@0: 
michael@0: 	    a = *mask++;
michael@0: 	    d = *dst;
michael@0: 
michael@0: 	    m = MUL_UN8 (sa, a, tmp);
michael@0: 	    d = MUL_UN8 (m, d, tmp);
michael@0: 
michael@0: 	    *dst++ = d;
michael@0: 	    w--;
michael@0: 	}
michael@0: 
michael@0: 	while (w >= 4)
michael@0: 	{
michael@0: 	    __m64 vmask;
michael@0: 	    __m64 vdest;
michael@0: 
michael@0: 	    vmask = load8888u ((uint32_t *)mask);
michael@0: 	    vdest = load8888 ((uint32_t *)dst);
michael@0: 
michael@0: 	    store8888 ((uint32_t *)dst, in (in (vsrca, vmask), vdest));
michael@0: 
michael@0: 	    dst += 4;
michael@0: 	    mask += 4;
michael@0: 	    w -= 4;
michael@0: 	}
michael@0: 
michael@0: 	while (w--)
michael@0: 	{
michael@0: 	    uint16_t tmp;
michael@0: 	    uint8_t a;
michael@0: 	    uint32_t m, d;
michael@0: 
michael@0: 	    a = *mask++;
michael@0: 	    d = *dst;
michael@0: 
michael@0: 	    m = MUL_UN8 (sa, a, tmp);
michael@0: 	    d = MUL_UN8 (m, d, tmp);
michael@0: 
michael@0: 	    *dst++ = d;
michael@0: 	}
michael@0:     }
michael@0: 
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_composite_in_8_8 (pixman_implementation_t *imp,
michael@0:                       pixman_composite_info_t *info)
michael@0: {
michael@0:     PIXMAN_COMPOSITE_ARGS (info);
michael@0:     uint8_t     *dst_line, *dst;
michael@0:     uint8_t     *src_line, *src;
michael@0:     int src_stride, dst_stride;
michael@0:     int32_t w;
michael@0: 
michael@0:     PIXMAN_IMAGE_GET_LINE (dest_image, dest_x, dest_y, uint8_t, dst_stride, dst_line, 1);
michael@0:     PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint8_t, src_stride, src_line, 1);
michael@0: 
michael@0:     while (height--)
michael@0:     {
michael@0: 	dst = dst_line;
michael@0: 	dst_line += dst_stride;
michael@0: 	src = src_line;
michael@0: 	src_line += src_stride;
michael@0: 	w = width;
michael@0: 
michael@0: 	while (w && (uintptr_t)dst & 3)
michael@0: 	{
michael@0: 	    uint8_t s, d;
michael@0: 	    uint16_t tmp;
michael@0: 
michael@0: 	    s = *src;
michael@0: 	    d = *dst;
michael@0: 
michael@0: 	    *dst = MUL_UN8 (s, d, tmp);
michael@0: 
michael@0: 	    src++;
michael@0: 	    dst++;
michael@0: 	    w--;
michael@0: 	}
michael@0: 
michael@0: 	while (w >= 4)
michael@0: 	{
michael@0: 	    uint32_t *s = (uint32_t *)src;
michael@0: 	    uint32_t *d = (uint32_t *)dst;
michael@0: 
michael@0: 	    store8888 (d, in (load8888u (s), load8888 (d)));
michael@0: 
michael@0: 	    w -= 4;
michael@0: 	    dst += 4;
michael@0: 	    src += 4;
michael@0: 	}
michael@0: 
michael@0: 	while (w--)
michael@0: 	{
michael@0: 	    uint8_t s, d;
michael@0: 	    uint16_t tmp;
michael@0: 
michael@0: 	    s = *src;
michael@0: 	    d = *dst;
michael@0: 
michael@0: 	    *dst = MUL_UN8 (s, d, tmp);
michael@0: 
michael@0: 	    src++;
michael@0: 	    dst++;
michael@0: 	}
michael@0:     }
michael@0: 
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_composite_add_n_8_8 (pixman_implementation_t *imp,
michael@0: 			 pixman_composite_info_t *info)
michael@0: {
michael@0:     PIXMAN_COMPOSITE_ARGS (info);
michael@0:     uint8_t     *dst_line, *dst;
michael@0:     uint8_t     *mask_line, *mask;
michael@0:     int dst_stride, mask_stride;
michael@0:     int32_t w;
michael@0:     uint32_t src;
michael@0:     uint8_t sa;
michael@0:     __m64 vsrc, vsrca;
michael@0: 
michael@0:     PIXMAN_IMAGE_GET_LINE (dest_image, dest_x, dest_y, uint8_t, dst_stride, dst_line, 1);
michael@0:     PIXMAN_IMAGE_GET_LINE (mask_image, mask_x, mask_y, uint8_t, mask_stride, mask_line, 1);
michael@0: 
michael@0:     src = _pixman_image_get_solid (imp, src_image, dest_image->bits.format);
michael@0: 
michael@0:     sa = src >> 24;
michael@0: 
michael@0:     if (src == 0)
michael@0: 	return;
michael@0: 
michael@0:     vsrc = load8888 (&src);
michael@0:     vsrca = expand_alpha (vsrc);
michael@0: 
michael@0:     while (height--)
michael@0:     {
michael@0: 	dst = dst_line;
michael@0: 	dst_line += dst_stride;
michael@0: 	mask = mask_line;
michael@0: 	mask_line += mask_stride;
michael@0: 	w = width;
michael@0: 
michael@0: 	while (w && (uintptr_t)dst & 3)
michael@0: 	{
michael@0: 	    uint16_t tmp;
michael@0: 	    uint16_t a;
michael@0: 	    uint32_t m, d;
michael@0: 	    uint32_t r;
michael@0: 
michael@0: 	    a = *mask++;
michael@0: 	    d = *dst;
michael@0: 
michael@0: 	    m = MUL_UN8 (sa, a, tmp);
michael@0: 	    r = ADD_UN8 (m, d, tmp);
michael@0: 
michael@0: 	    *dst++ = r;
michael@0: 	    w--;
michael@0: 	}
michael@0: 
michael@0: 	while (w >= 4)
michael@0: 	{
michael@0: 	    __m64 vmask;
michael@0: 	    __m64 vdest;
michael@0: 
michael@0: 	    vmask = load8888u ((uint32_t *)mask);
michael@0: 	    vdest = load8888 ((uint32_t *)dst);
michael@0: 
michael@0: 	    store8888 ((uint32_t *)dst, _mm_adds_pu8 (in (vsrca, vmask), vdest));
michael@0: 
michael@0: 	    dst += 4;
michael@0: 	    mask += 4;
michael@0: 	    w -= 4;
michael@0: 	}
michael@0: 
michael@0: 	while (w--)
michael@0: 	{
michael@0: 	    uint16_t tmp;
michael@0: 	    uint16_t a;
michael@0: 	    uint32_t m, d;
michael@0: 	    uint32_t r;
michael@0: 
michael@0: 	    a = *mask++;
michael@0: 	    d = *dst;
michael@0: 
michael@0: 	    m = MUL_UN8 (sa, a, tmp);
michael@0: 	    r = ADD_UN8 (m, d, tmp);
michael@0: 
michael@0: 	    *dst++ = r;
michael@0: 	}
michael@0:     }
michael@0: 
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_composite_add_8_8 (pixman_implementation_t *imp,
michael@0: 		       pixman_composite_info_t *info)
michael@0: {
michael@0:     PIXMAN_COMPOSITE_ARGS (info);
michael@0:     uint8_t *dst_line, *dst;
michael@0:     uint8_t *src_line, *src;
michael@0:     int dst_stride, src_stride;
michael@0:     int32_t w;
michael@0:     uint8_t s, d;
michael@0:     uint16_t t;
michael@0: 
michael@0:     CHECKPOINT ();
michael@0: 
michael@0:     PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint8_t, src_stride, src_line, 1);
michael@0:     PIXMAN_IMAGE_GET_LINE (dest_image, dest_x, dest_y, uint8_t, dst_stride, dst_line, 1);
michael@0: 
michael@0:     while (height--)
michael@0:     {
michael@0: 	dst = dst_line;
michael@0: 	dst_line += dst_stride;
michael@0: 	src = src_line;
michael@0: 	src_line += src_stride;
michael@0: 	w = width;
michael@0: 
michael@0: 	while (w && (uintptr_t)dst & 7)
michael@0: 	{
michael@0: 	    s = *src;
michael@0: 	    d = *dst;
michael@0: 	    t = d + s;
michael@0: 	    s = t | (0 - (t >> 8));
michael@0: 	    *dst = s;
michael@0: 
michael@0: 	    dst++;
michael@0: 	    src++;
michael@0: 	    w--;
michael@0: 	}
michael@0: 
michael@0: 	while (w >= 8)
michael@0: 	{
michael@0: 	    *(__m64*)dst = _mm_adds_pu8 (ldq_u ((__m64 *)src), *(__m64*)dst);
michael@0: 	    dst += 8;
michael@0: 	    src += 8;
michael@0: 	    w -= 8;
michael@0: 	}
michael@0: 
michael@0: 	while (w)
michael@0: 	{
michael@0: 	    s = *src;
michael@0: 	    d = *dst;
michael@0: 	    t = d + s;
michael@0: 	    s = t | (0 - (t >> 8));
michael@0: 	    *dst = s;
michael@0: 
michael@0: 	    dst++;
michael@0: 	    src++;
michael@0: 	    w--;
michael@0: 	}
michael@0:     }
michael@0: 
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_composite_add_0565_0565 (pixman_implementation_t *imp,
michael@0:                              pixman_composite_info_t *info)
michael@0: {
michael@0:     PIXMAN_COMPOSITE_ARGS (info);
michael@0:     uint16_t    *dst_line, *dst;
michael@0:     uint32_t	d;
michael@0:     uint16_t    *src_line, *src;
michael@0:     uint32_t	s;
michael@0:     int dst_stride, src_stride;
michael@0:     int32_t w;
michael@0: 
michael@0:     CHECKPOINT ();
michael@0: 
michael@0:     PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint16_t, src_stride, src_line, 1);
michael@0:     PIXMAN_IMAGE_GET_LINE (dest_image, dest_x, dest_y, uint16_t, dst_stride, dst_line, 1);
michael@0: 
michael@0:     while (height--)
michael@0:     {
michael@0: 	dst = dst_line;
michael@0: 	dst_line += dst_stride;
michael@0: 	src = src_line;
michael@0: 	src_line += src_stride;
michael@0: 	w = width;
michael@0: 
michael@0: 	while (w && (uintptr_t)dst & 7)
michael@0: 	{
michael@0: 	    s = *src++;
michael@0: 	    if (s)
michael@0: 	    {
michael@0: 		d = *dst;
michael@0: 		s = convert_0565_to_8888 (s);
michael@0: 		if (d)
michael@0: 		{
michael@0: 		    d = convert_0565_to_8888 (d);
michael@0: 		    UN8x4_ADD_UN8x4 (s, d);
michael@0: 		}
michael@0: 		*dst = convert_8888_to_0565 (s);
michael@0: 	    }
michael@0: 	    dst++;
michael@0: 	    w--;
michael@0: 	}
michael@0: 
michael@0: 	while (w >= 4)
michael@0: 	{
michael@0: 	    __m64 vdest = *(__m64 *)dst;
michael@0: 	    __m64 vsrc = ldq_u ((__m64 *)src);
michael@0: 	    __m64 vd0, vd1;
michael@0: 	    __m64 vs0, vs1;
michael@0: 
michael@0: 	    expand_4xpacked565 (vdest, &vd0, &vd1, 0);
michael@0: 	    expand_4xpacked565 (vsrc, &vs0, &vs1, 0);
michael@0: 
michael@0: 	    vd0 = _mm_adds_pu8 (vd0, vs0);
michael@0: 	    vd1 = _mm_adds_pu8 (vd1, vs1);
michael@0: 
michael@0: 	    *(__m64 *)dst = pack_4xpacked565 (vd0, vd1);
michael@0: 
michael@0: 	    dst += 4;
michael@0: 	    src += 4;
michael@0: 	    w -= 4;
michael@0: 	}
michael@0: 
michael@0: 	while (w--)
michael@0: 	{
michael@0: 	    s = *src++;
michael@0: 	    if (s)
michael@0: 	    {
michael@0: 		d = *dst;
michael@0: 		s = convert_0565_to_8888 (s);
michael@0: 		if (d)
michael@0: 		{
michael@0: 		    d = convert_0565_to_8888 (d);
michael@0: 		    UN8x4_ADD_UN8x4 (s, d);
michael@0: 		}
michael@0: 		*dst = convert_8888_to_0565 (s);
michael@0: 	    }
michael@0: 	    dst++;
michael@0: 	}
michael@0:     }
michael@0: 
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_composite_add_8888_8888 (pixman_implementation_t *imp,
michael@0:                              pixman_composite_info_t *info)
michael@0: {
michael@0:     PIXMAN_COMPOSITE_ARGS (info);
michael@0:     uint32_t    *dst_line, *dst;
michael@0:     uint32_t    *src_line, *src;
michael@0:     int dst_stride, src_stride;
michael@0:     int32_t w;
michael@0: 
michael@0:     CHECKPOINT ();
michael@0: 
michael@0:     PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
michael@0:     PIXMAN_IMAGE_GET_LINE (dest_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
michael@0: 
michael@0:     while (height--)
michael@0:     {
michael@0: 	dst = dst_line;
michael@0: 	dst_line += dst_stride;
michael@0: 	src = src_line;
michael@0: 	src_line += src_stride;
michael@0: 	w = width;
michael@0: 
michael@0: 	while (w && (uintptr_t)dst & 7)
michael@0: 	{
michael@0: 	    store (dst, _mm_adds_pu8 (load ((const uint32_t *)src),
michael@0: 	                              load ((const uint32_t *)dst)));
michael@0: 	    dst++;
michael@0: 	    src++;
michael@0: 	    w--;
michael@0: 	}
michael@0: 
michael@0: 	while (w >= 2)
michael@0: 	{
michael@0: 	    *(__m64 *)dst = _mm_adds_pu8 (ldq_u ((__m64 *)src), *(__m64*)dst);
michael@0: 	    dst += 2;
michael@0: 	    src += 2;
michael@0: 	    w -= 2;
michael@0: 	}
michael@0: 
michael@0: 	if (w)
michael@0: 	{
michael@0: 	    store (dst, _mm_adds_pu8 (load ((const uint32_t *)src),
michael@0: 	                              load ((const uint32_t *)dst)));
michael@0: 
michael@0: 	}
michael@0:     }
michael@0: 
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static pixman_bool_t
michael@0: mmx_blt (pixman_implementation_t *imp,
michael@0:          uint32_t *               src_bits,
michael@0:          uint32_t *               dst_bits,
michael@0:          int                      src_stride,
michael@0:          int                      dst_stride,
michael@0:          int                      src_bpp,
michael@0:          int                      dst_bpp,
michael@0:          int                      src_x,
michael@0:          int                      src_y,
michael@0:          int                      dest_x,
michael@0:          int                      dest_y,
michael@0:          int                      width,
michael@0:          int                      height)
michael@0: {
michael@0:     uint8_t *   src_bytes;
michael@0:     uint8_t *   dst_bytes;
michael@0:     int byte_width;
michael@0: 
michael@0:     if (src_bpp != dst_bpp)
michael@0: 	return FALSE;
michael@0: 
michael@0:     if (src_bpp == 16)
michael@0:     {
michael@0: 	src_stride = src_stride * (int) sizeof (uint32_t) / 2;
michael@0: 	dst_stride = dst_stride * (int) sizeof (uint32_t) / 2;
michael@0: 	src_bytes = (uint8_t *)(((uint16_t *)src_bits) + src_stride * (src_y) + (src_x));
michael@0: 	dst_bytes = (uint8_t *)(((uint16_t *)dst_bits) + dst_stride * (dest_y) + (dest_x));
michael@0: 	byte_width = 2 * width;
michael@0: 	src_stride *= 2;
michael@0: 	dst_stride *= 2;
michael@0:     }
michael@0:     else if (src_bpp == 32)
michael@0:     {
michael@0: 	src_stride = src_stride * (int) sizeof (uint32_t) / 4;
michael@0: 	dst_stride = dst_stride * (int) sizeof (uint32_t) / 4;
michael@0: 	src_bytes = (uint8_t *)(((uint32_t *)src_bits) + src_stride * (src_y) + (src_x));
michael@0: 	dst_bytes = (uint8_t *)(((uint32_t *)dst_bits) + dst_stride * (dest_y) + (dest_x));
michael@0: 	byte_width = 4 * width;
michael@0: 	src_stride *= 4;
michael@0: 	dst_stride *= 4;
michael@0:     }
michael@0:     else
michael@0:     {
michael@0: 	return FALSE;
michael@0:     }
michael@0: 
michael@0:     while (height--)
michael@0:     {
michael@0: 	int w;
michael@0: 	uint8_t *s = src_bytes;
michael@0: 	uint8_t *d = dst_bytes;
michael@0: 	src_bytes += src_stride;
michael@0: 	dst_bytes += dst_stride;
michael@0: 	w = byte_width;
michael@0: 
michael@0: 	if (w >= 1 && ((uintptr_t)d & 1))
michael@0: 	{
michael@0: 	    *(uint8_t *)d = *(uint8_t *)s;
michael@0: 	    w -= 1;
michael@0: 	    s += 1;
michael@0: 	    d += 1;
michael@0: 	}
michael@0: 
michael@0: 	if (w >= 2 && ((uintptr_t)d & 3))
michael@0: 	{
michael@0: 	    *(uint16_t *)d = *(uint16_t *)s;
michael@0: 	    w -= 2;
michael@0: 	    s += 2;
michael@0: 	    d += 2;
michael@0: 	}
michael@0: 
michael@0: 	while (w >= 4 && ((uintptr_t)d & 7))
michael@0: 	{
michael@0: 	    *(uint32_t *)d = ldl_u ((uint32_t *)s);
michael@0: 
michael@0: 	    w -= 4;
michael@0: 	    s += 4;
michael@0: 	    d += 4;
michael@0: 	}
michael@0: 
michael@0: 	while (w >= 64)
michael@0: 	{
michael@0: #if (defined (__GNUC__) || (defined(__SUNPRO_C) && (__SUNPRO_C >= 0x590))) && defined USE_X86_MMX
michael@0: 	    __asm__ (
michael@0: 	        "movq	  (%1),	  %%mm0\n"
michael@0: 	        "movq	 8(%1),	  %%mm1\n"
michael@0: 	        "movq	16(%1),	  %%mm2\n"
michael@0: 	        "movq	24(%1),	  %%mm3\n"
michael@0: 	        "movq	32(%1),	  %%mm4\n"
michael@0: 	        "movq	40(%1),	  %%mm5\n"
michael@0: 	        "movq	48(%1),	  %%mm6\n"
michael@0: 	        "movq	56(%1),	  %%mm7\n"
michael@0: 
michael@0: 	        "movq	%%mm0,	  (%0)\n"
michael@0: 	        "movq	%%mm1,	 8(%0)\n"
michael@0: 	        "movq	%%mm2,	16(%0)\n"
michael@0: 	        "movq	%%mm3,	24(%0)\n"
michael@0: 	        "movq	%%mm4,	32(%0)\n"
michael@0: 	        "movq	%%mm5,	40(%0)\n"
michael@0: 	        "movq	%%mm6,	48(%0)\n"
michael@0: 	        "movq	%%mm7,	56(%0)\n"
michael@0: 		:
michael@0: 		: "r" (d), "r" (s)
michael@0: 		: "memory",
michael@0: 		  "%mm0", "%mm1", "%mm2", "%mm3",
michael@0: 		  "%mm4", "%mm5", "%mm6", "%mm7");
michael@0: #else
michael@0: 	    __m64 v0 = ldq_u ((__m64 *)(s + 0));
michael@0: 	    __m64 v1 = ldq_u ((__m64 *)(s + 8));
michael@0: 	    __m64 v2 = ldq_u ((__m64 *)(s + 16));
michael@0: 	    __m64 v3 = ldq_u ((__m64 *)(s + 24));
michael@0: 	    __m64 v4 = ldq_u ((__m64 *)(s + 32));
michael@0: 	    __m64 v5 = ldq_u ((__m64 *)(s + 40));
michael@0: 	    __m64 v6 = ldq_u ((__m64 *)(s + 48));
michael@0: 	    __m64 v7 = ldq_u ((__m64 *)(s + 56));
michael@0: 	    *(__m64 *)(d + 0)  = v0;
michael@0: 	    *(__m64 *)(d + 8)  = v1;
michael@0: 	    *(__m64 *)(d + 16) = v2;
michael@0: 	    *(__m64 *)(d + 24) = v3;
michael@0: 	    *(__m64 *)(d + 32) = v4;
michael@0: 	    *(__m64 *)(d + 40) = v5;
michael@0: 	    *(__m64 *)(d + 48) = v6;
michael@0: 	    *(__m64 *)(d + 56) = v7;
michael@0: #endif
michael@0: 
michael@0: 	    w -= 64;
michael@0: 	    s += 64;
michael@0: 	    d += 64;
michael@0: 	}
michael@0: 	while (w >= 4)
michael@0: 	{
michael@0: 	    *(uint32_t *)d = ldl_u ((uint32_t *)s);
michael@0: 
michael@0: 	    w -= 4;
michael@0: 	    s += 4;
michael@0: 	    d += 4;
michael@0: 	}
michael@0: 	if (w >= 2)
michael@0: 	{
michael@0: 	    *(uint16_t *)d = *(uint16_t *)s;
michael@0: 	    w -= 2;
michael@0: 	    s += 2;
michael@0: 	    d += 2;
michael@0: 	}
michael@0:     }
michael@0: 
michael@0:     _mm_empty ();
michael@0: 
michael@0:     return TRUE;
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_composite_copy_area (pixman_implementation_t *imp,
michael@0:                          pixman_composite_info_t *info)
michael@0: {
michael@0:     PIXMAN_COMPOSITE_ARGS (info);
michael@0: 
michael@0:     mmx_blt (imp, src_image->bits.bits,
michael@0: 	     dest_image->bits.bits,
michael@0: 	     src_image->bits.rowstride,
michael@0: 	     dest_image->bits.rowstride,
michael@0: 	     PIXMAN_FORMAT_BPP (src_image->bits.format),
michael@0: 	     PIXMAN_FORMAT_BPP (dest_image->bits.format),
michael@0: 	     src_x, src_y, dest_x, dest_y, width, height);
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_composite_over_x888_8_8888 (pixman_implementation_t *imp,
michael@0:                                 pixman_composite_info_t *info)
michael@0: {
michael@0:     PIXMAN_COMPOSITE_ARGS (info);
michael@0:     uint32_t  *src, *src_line;
michael@0:     uint32_t  *dst, *dst_line;
michael@0:     uint8_t  *mask, *mask_line;
michael@0:     int src_stride, mask_stride, dst_stride;
michael@0:     int32_t w;
michael@0: 
michael@0:     PIXMAN_IMAGE_GET_LINE (dest_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
michael@0:     PIXMAN_IMAGE_GET_LINE (mask_image, mask_x, mask_y, uint8_t, mask_stride, mask_line, 1);
michael@0:     PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
michael@0: 
michael@0:     while (height--)
michael@0:     {
michael@0: 	src = src_line;
michael@0: 	src_line += src_stride;
michael@0: 	dst = dst_line;
michael@0: 	dst_line += dst_stride;
michael@0: 	mask = mask_line;
michael@0: 	mask_line += mask_stride;
michael@0: 
michael@0: 	w = width;
michael@0: 
michael@0: 	while (w--)
michael@0: 	{
michael@0: 	    uint64_t m = *mask;
michael@0: 
michael@0: 	    if (m)
michael@0: 	    {
michael@0: 		uint32_t ssrc = *src | 0xff000000;
michael@0: 		__m64 s = load8888 (&ssrc);
michael@0: 
michael@0: 		if (m == 0xff)
michael@0: 		{
michael@0: 		    store8888 (dst, s);
michael@0: 		}
michael@0: 		else
michael@0: 		{
michael@0: 		    __m64 sa = expand_alpha (s);
michael@0: 		    __m64 vm = expand_alpha_rev (to_m64 (m));
michael@0: 		    __m64 vdest = in_over (s, sa, vm, load8888 (dst));
michael@0: 
michael@0: 		    store8888 (dst, vdest);
michael@0: 		}
michael@0: 	    }
michael@0: 
michael@0: 	    mask++;
michael@0: 	    dst++;
michael@0: 	    src++;
michael@0: 	}
michael@0:     }
michael@0: 
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: mmx_composite_over_reverse_n_8888 (pixman_implementation_t *imp,
michael@0:                                    pixman_composite_info_t *info)
michael@0: {
michael@0:     PIXMAN_COMPOSITE_ARGS (info);
michael@0:     uint32_t src;
michael@0:     uint32_t    *dst_line, *dst;
michael@0:     int32_t w;
michael@0:     int dst_stride;
michael@0:     __m64 vsrc;
michael@0: 
michael@0:     CHECKPOINT ();
michael@0: 
michael@0:     src = _pixman_image_get_solid (imp, src_image, dest_image->bits.format);
michael@0: 
michael@0:     if (src == 0)
michael@0: 	return;
michael@0: 
michael@0:     PIXMAN_IMAGE_GET_LINE (dest_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
michael@0: 
michael@0:     vsrc = load8888 (&src);
michael@0: 
michael@0:     while (height--)
michael@0:     {
michael@0: 	dst = dst_line;
michael@0: 	dst_line += dst_stride;
michael@0: 	w = width;
michael@0: 
michael@0: 	CHECKPOINT ();
michael@0: 
michael@0: 	while (w && (uintptr_t)dst & 7)
michael@0: 	{
michael@0: 	    __m64 vdest = load8888 (dst);
michael@0: 
michael@0: 	    store8888 (dst, over (vdest, expand_alpha (vdest), vsrc));
michael@0: 
michael@0: 	    w--;
michael@0: 	    dst++;
michael@0: 	}
michael@0: 
michael@0: 	while (w >= 2)
michael@0: 	{
michael@0: 	    __m64 vdest = *(__m64 *)dst;
michael@0: 	    __m64 dest0 = expand8888 (vdest, 0);
michael@0: 	    __m64 dest1 = expand8888 (vdest, 1);
michael@0: 
michael@0: 
michael@0: 	    dest0 = over (dest0, expand_alpha (dest0), vsrc);
michael@0: 	    dest1 = over (dest1, expand_alpha (dest1), vsrc);
michael@0: 
michael@0: 	    *(__m64 *)dst = pack8888 (dest0, dest1);
michael@0: 
michael@0: 	    dst += 2;
michael@0: 	    w -= 2;
michael@0: 	}
michael@0: 
michael@0: 	CHECKPOINT ();
michael@0: 
michael@0: 	if (w)
michael@0: 	{
michael@0: 	    __m64 vdest = load8888 (dst);
michael@0: 
michael@0: 	    store8888 (dst, over (vdest, expand_alpha (vdest), vsrc));
michael@0: 	}
michael@0:     }
michael@0: 
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: #define BSHIFT ((1 << BILINEAR_INTERPOLATION_BITS))
michael@0: #define BMSK (BSHIFT - 1)
michael@0: 
michael@0: #define BILINEAR_DECLARE_VARIABLES						\
michael@0:     const __m64 mm_wt = _mm_set_pi16 (wt, wt, wt, wt);				\
michael@0:     const __m64 mm_wb = _mm_set_pi16 (wb, wb, wb, wb);				\
michael@0:     const __m64 mm_BSHIFT = _mm_set_pi16 (BSHIFT, BSHIFT, BSHIFT, BSHIFT);	\
michael@0:     const __m64 mm_addc7 = _mm_set_pi16 (0, 1, 0, 1);				\
michael@0:     const __m64 mm_xorc7 = _mm_set_pi16 (0, BMSK, 0, BMSK);			\
michael@0:     const __m64 mm_ux = _mm_set_pi16 (unit_x, unit_x, unit_x, unit_x);		\
michael@0:     const __m64 mm_zero = _mm_setzero_si64 ();					\
michael@0:     __m64 mm_x = _mm_set_pi16 (vx, vx, vx, vx)
michael@0: 
michael@0: #define BILINEAR_INTERPOLATE_ONE_PIXEL(pix)					\
michael@0: do {										\
michael@0:     /* fetch 2x2 pixel block into 2 mmx registers */				\
michael@0:     __m64 t = ldq_u ((__m64 *)&src_top [pixman_fixed_to_int (vx)]);		\
michael@0:     __m64 b = ldq_u ((__m64 *)&src_bottom [pixman_fixed_to_int (vx)]);		\
michael@0:     /* vertical interpolation */						\
michael@0:     __m64 t_hi = _mm_mullo_pi16 (_mm_unpackhi_pi8 (t, mm_zero), mm_wt);		\
michael@0:     __m64 t_lo = _mm_mullo_pi16 (_mm_unpacklo_pi8 (t, mm_zero), mm_wt);		\
michael@0:     __m64 b_hi = _mm_mullo_pi16 (_mm_unpackhi_pi8 (b, mm_zero), mm_wb);		\
michael@0:     __m64 b_lo = _mm_mullo_pi16 (_mm_unpacklo_pi8 (b, mm_zero), mm_wb);		\
michael@0:     __m64 hi = _mm_add_pi16 (t_hi, b_hi);					\
michael@0:     __m64 lo = _mm_add_pi16 (t_lo, b_lo);					\
michael@0:     vx += unit_x;								\
michael@0:     if (BILINEAR_INTERPOLATION_BITS < 8)					\
michael@0:     {										\
michael@0: 	/* calculate horizontal weights */					\
michael@0: 	__m64 mm_wh = _mm_add_pi16 (mm_addc7, _mm_xor_si64 (mm_xorc7,		\
michael@0: 			  _mm_srli_pi16 (mm_x,					\
michael@0: 					 16 - BILINEAR_INTERPOLATION_BITS)));	\
michael@0: 	/* horizontal interpolation */						\
michael@0: 	__m64 p = _mm_unpacklo_pi16 (lo, hi);					\
michael@0: 	__m64 q = _mm_unpackhi_pi16 (lo, hi);					\
michael@0: 	lo = _mm_madd_pi16 (p, mm_wh);						\
michael@0: 	hi = _mm_madd_pi16 (q, mm_wh);						\
michael@0:     }										\
michael@0:     else									\
michael@0:     {										\
michael@0: 	/* calculate horizontal weights */					\
michael@0: 	__m64 mm_wh_lo = _mm_sub_pi16 (mm_BSHIFT, _mm_srli_pi16 (mm_x,		\
michael@0: 					16 - BILINEAR_INTERPOLATION_BITS));	\
michael@0: 	__m64 mm_wh_hi = _mm_srli_pi16 (mm_x,					\
michael@0: 					16 - BILINEAR_INTERPOLATION_BITS);	\
michael@0: 	/* horizontal interpolation */						\
michael@0: 	__m64 mm_lo_lo = _mm_mullo_pi16 (lo, mm_wh_lo);				\
michael@0: 	__m64 mm_lo_hi = _mm_mullo_pi16 (hi, mm_wh_hi);				\
michael@0: 	__m64 mm_hi_lo = _mm_mulhi_pu16 (lo, mm_wh_lo);				\
michael@0: 	__m64 mm_hi_hi = _mm_mulhi_pu16 (hi, mm_wh_hi);				\
michael@0: 	lo = _mm_add_pi32 (_mm_unpacklo_pi16 (mm_lo_lo, mm_hi_lo),		\
michael@0: 			   _mm_unpacklo_pi16 (mm_lo_hi, mm_hi_hi));		\
michael@0: 	hi = _mm_add_pi32 (_mm_unpackhi_pi16 (mm_lo_lo, mm_hi_lo),		\
michael@0: 			   _mm_unpackhi_pi16 (mm_lo_hi, mm_hi_hi));		\
michael@0:     }										\
michael@0:     mm_x = _mm_add_pi16 (mm_x, mm_ux);						\
michael@0:     /* shift and pack the result */						\
michael@0:     hi = _mm_srli_pi32 (hi, BILINEAR_INTERPOLATION_BITS * 2);			\
michael@0:     lo = _mm_srli_pi32 (lo, BILINEAR_INTERPOLATION_BITS * 2);			\
michael@0:     lo = _mm_packs_pi32 (lo, hi);						\
michael@0:     lo = _mm_packs_pu16 (lo, lo);						\
michael@0:     pix = lo;									\
michael@0: } while (0)
michael@0: 
michael@0: #define BILINEAR_SKIP_ONE_PIXEL()						\
michael@0: do {										\
michael@0:     vx += unit_x;								\
michael@0:     mm_x = _mm_add_pi16 (mm_x, mm_ux);						\
michael@0: } while(0)
michael@0: 
michael@0: static force_inline void
michael@0: scaled_bilinear_scanline_mmx_8888_8888_SRC (uint32_t *       dst,
michael@0: 					    const uint32_t * mask,
michael@0: 					    const uint32_t * src_top,
michael@0: 					    const uint32_t * src_bottom,
michael@0: 					    int32_t          w,
michael@0: 					    int              wt,
michael@0: 					    int              wb,
michael@0: 					    pixman_fixed_t   vx,
michael@0: 					    pixman_fixed_t   unit_x,
michael@0: 					    pixman_fixed_t   max_vx,
michael@0: 					    pixman_bool_t    zero_src)
michael@0: {
michael@0:     BILINEAR_DECLARE_VARIABLES;
michael@0:     __m64 pix;
michael@0: 
michael@0:     while (w--)
michael@0:     {
michael@0: 	BILINEAR_INTERPOLATE_ONE_PIXEL (pix);
michael@0: 	store (dst, pix);
michael@0: 	dst++;
michael@0:     }
michael@0: 
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: FAST_BILINEAR_MAINLOOP_COMMON (mmx_8888_8888_cover_SRC,
michael@0: 			       scaled_bilinear_scanline_mmx_8888_8888_SRC,
michael@0: 			       uint32_t, uint32_t, uint32_t,
michael@0: 			       COVER, FLAG_NONE)
michael@0: FAST_BILINEAR_MAINLOOP_COMMON (mmx_8888_8888_pad_SRC,
michael@0: 			       scaled_bilinear_scanline_mmx_8888_8888_SRC,
michael@0: 			       uint32_t, uint32_t, uint32_t,
michael@0: 			       PAD, FLAG_NONE)
michael@0: FAST_BILINEAR_MAINLOOP_COMMON (mmx_8888_8888_none_SRC,
michael@0: 			       scaled_bilinear_scanline_mmx_8888_8888_SRC,
michael@0: 			       uint32_t, uint32_t, uint32_t,
michael@0: 			       NONE, FLAG_NONE)
michael@0: FAST_BILINEAR_MAINLOOP_COMMON (mmx_8888_8888_normal_SRC,
michael@0: 			       scaled_bilinear_scanline_mmx_8888_8888_SRC,
michael@0: 			       uint32_t, uint32_t, uint32_t,
michael@0: 			       NORMAL, FLAG_NONE)
michael@0: 
michael@0: static force_inline void
michael@0: scaled_bilinear_scanline_mmx_8888_8888_OVER (uint32_t *       dst,
michael@0: 					     const uint32_t * mask,
michael@0: 					     const uint32_t * src_top,
michael@0: 					     const uint32_t * src_bottom,
michael@0: 					     int32_t          w,
michael@0: 					     int              wt,
michael@0: 					     int              wb,
michael@0: 					     pixman_fixed_t   vx,
michael@0: 					     pixman_fixed_t   unit_x,
michael@0: 					     pixman_fixed_t   max_vx,
michael@0: 					     pixman_bool_t    zero_src)
michael@0: {
michael@0:     BILINEAR_DECLARE_VARIABLES;
michael@0:     __m64 pix1, pix2;
michael@0: 
michael@0:     while (w)
michael@0:     {
michael@0: 	BILINEAR_INTERPOLATE_ONE_PIXEL (pix1);
michael@0: 
michael@0: 	if (!is_zero (pix1))
michael@0: 	{
michael@0: 	    pix2 = load (dst);
michael@0: 	    store8888 (dst, core_combine_over_u_pixel_mmx (pix1, pix2));
michael@0: 	}
michael@0: 
michael@0: 	w--;
michael@0: 	dst++;
michael@0:     }
michael@0: 
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: FAST_BILINEAR_MAINLOOP_COMMON (mmx_8888_8888_cover_OVER,
michael@0: 			       scaled_bilinear_scanline_mmx_8888_8888_OVER,
michael@0: 			       uint32_t, uint32_t, uint32_t,
michael@0: 			       COVER, FLAG_NONE)
michael@0: FAST_BILINEAR_MAINLOOP_COMMON (mmx_8888_8888_pad_OVER,
michael@0: 			       scaled_bilinear_scanline_mmx_8888_8888_OVER,
michael@0: 			       uint32_t, uint32_t, uint32_t,
michael@0: 			       PAD, FLAG_NONE)
michael@0: FAST_BILINEAR_MAINLOOP_COMMON (mmx_8888_8888_none_OVER,
michael@0: 			       scaled_bilinear_scanline_mmx_8888_8888_OVER,
michael@0: 			       uint32_t, uint32_t, uint32_t,
michael@0: 			       NONE, FLAG_NONE)
michael@0: FAST_BILINEAR_MAINLOOP_COMMON (mmx_8888_8888_normal_OVER,
michael@0: 			       scaled_bilinear_scanline_mmx_8888_8888_OVER,
michael@0: 			       uint32_t, uint32_t, uint32_t,
michael@0: 			       NORMAL, FLAG_NONE)
michael@0: 
michael@0: static force_inline void
michael@0: scaled_bilinear_scanline_mmx_8888_8_8888_OVER (uint32_t *       dst,
michael@0: 					       const uint8_t  * mask,
michael@0: 					       const uint32_t * src_top,
michael@0: 					       const uint32_t * src_bottom,
michael@0: 					       int32_t          w,
michael@0: 					       int              wt,
michael@0: 					       int              wb,
michael@0: 					       pixman_fixed_t   vx,
michael@0: 					       pixman_fixed_t   unit_x,
michael@0: 					       pixman_fixed_t   max_vx,
michael@0: 					       pixman_bool_t    zero_src)
michael@0: {
michael@0:     BILINEAR_DECLARE_VARIABLES;
michael@0:     __m64 pix1, pix2;
michael@0:     uint32_t m;
michael@0: 
michael@0:     while (w)
michael@0:     {
michael@0: 	m = (uint32_t) *mask++;
michael@0: 
michael@0: 	if (m)
michael@0: 	{
michael@0: 	    BILINEAR_INTERPOLATE_ONE_PIXEL (pix1);
michael@0: 
michael@0: 	    if (m == 0xff && is_opaque (pix1))
michael@0: 	    {
michael@0: 		store (dst, pix1);
michael@0: 	    }
michael@0: 	    else
michael@0: 	    {
michael@0: 		__m64 ms, md, ma, msa;
michael@0: 
michael@0: 		pix2 = load (dst);
michael@0: 		ma = expand_alpha_rev (to_m64 (m));
michael@0: 		ms = _mm_unpacklo_pi8 (pix1, _mm_setzero_si64 ());
michael@0: 		md = _mm_unpacklo_pi8 (pix2, _mm_setzero_si64 ());
michael@0: 
michael@0: 		msa = expand_alpha (ms);
michael@0: 
michael@0: 		store8888 (dst, (in_over (ms, msa, ma, md)));
michael@0: 	    }
michael@0: 	}
michael@0: 	else
michael@0: 	{
michael@0: 	    BILINEAR_SKIP_ONE_PIXEL ();
michael@0: 	}
michael@0: 
michael@0: 	w--;
michael@0: 	dst++;
michael@0:     }
michael@0: 
michael@0:     _mm_empty ();
michael@0: }
michael@0: 
michael@0: FAST_BILINEAR_MAINLOOP_COMMON (mmx_8888_8_8888_cover_OVER,
michael@0: 			       scaled_bilinear_scanline_mmx_8888_8_8888_OVER,
michael@0: 			       uint32_t, uint8_t, uint32_t,
michael@0: 			       COVER, FLAG_HAVE_NON_SOLID_MASK)
michael@0: FAST_BILINEAR_MAINLOOP_COMMON (mmx_8888_8_8888_pad_OVER,
michael@0: 			       scaled_bilinear_scanline_mmx_8888_8_8888_OVER,
michael@0: 			       uint32_t, uint8_t, uint32_t,
michael@0: 			       PAD, FLAG_HAVE_NON_SOLID_MASK)
michael@0: FAST_BILINEAR_MAINLOOP_COMMON (mmx_8888_8_8888_none_OVER,
michael@0: 			       scaled_bilinear_scanline_mmx_8888_8_8888_OVER,
michael@0: 			       uint32_t, uint8_t, uint32_t,
michael@0: 			       NONE, FLAG_HAVE_NON_SOLID_MASK)
michael@0: FAST_BILINEAR_MAINLOOP_COMMON (mmx_8888_8_8888_normal_OVER,
michael@0: 			       scaled_bilinear_scanline_mmx_8888_8_8888_OVER,
michael@0: 			       uint32_t, uint8_t, uint32_t,
michael@0: 			       NORMAL, FLAG_HAVE_NON_SOLID_MASK)
michael@0: 
michael@0: static uint32_t *
michael@0: mmx_fetch_x8r8g8b8 (pixman_iter_t *iter, const uint32_t *mask)
michael@0: {
michael@0:     int w = iter->width;
michael@0:     uint32_t *dst = iter->buffer;
michael@0:     uint32_t *src = (uint32_t *)iter->bits;
michael@0: 
michael@0:     iter->bits += iter->stride;
michael@0: 
michael@0:     while (w && ((uintptr_t)dst) & 7)
michael@0:     {
michael@0: 	*dst++ = (*src++) | 0xff000000;
michael@0: 	w--;
michael@0:     }
michael@0: 
michael@0:     while (w >= 8)
michael@0:     {
michael@0: 	__m64 vsrc1 = ldq_u ((__m64 *)(src + 0));
michael@0: 	__m64 vsrc2 = ldq_u ((__m64 *)(src + 2));
michael@0: 	__m64 vsrc3 = ldq_u ((__m64 *)(src + 4));
michael@0: 	__m64 vsrc4 = ldq_u ((__m64 *)(src + 6));
michael@0: 
michael@0: 	*(__m64 *)(dst + 0) = _mm_or_si64 (vsrc1, MC (ff000000));
michael@0: 	*(__m64 *)(dst + 2) = _mm_or_si64 (vsrc2, MC (ff000000));
michael@0: 	*(__m64 *)(dst + 4) = _mm_or_si64 (vsrc3, MC (ff000000));
michael@0: 	*(__m64 *)(dst + 6) = _mm_or_si64 (vsrc4, MC (ff000000));
michael@0: 
michael@0: 	dst += 8;
michael@0: 	src += 8;
michael@0: 	w -= 8;
michael@0:     }
michael@0: 
michael@0:     while (w)
michael@0:     {
michael@0: 	*dst++ = (*src++) | 0xff000000;
michael@0: 	w--;
michael@0:     }
michael@0: 
michael@0:     _mm_empty ();
michael@0:     return iter->buffer;
michael@0: }
michael@0: 
michael@0: static uint32_t *
michael@0: mmx_fetch_r5g6b5 (pixman_iter_t *iter, const uint32_t *mask)
michael@0: {
michael@0:     int w = iter->width;
michael@0:     uint32_t *dst = iter->buffer;
michael@0:     uint16_t *src = (uint16_t *)iter->bits;
michael@0: 
michael@0:     iter->bits += iter->stride;
michael@0: 
michael@0:     while (w && ((uintptr_t)dst) & 0x0f)
michael@0:     {
michael@0: 	uint16_t s = *src++;
michael@0: 
michael@0: 	*dst++ = convert_0565_to_8888 (s);
michael@0: 	w--;
michael@0:     }
michael@0: 
michael@0:     while (w >= 4)
michael@0:     {
michael@0: 	__m64 vsrc = ldq_u ((__m64 *)src);
michael@0: 	__m64 mm0, mm1;
michael@0: 
michael@0: 	expand_4xpacked565 (vsrc, &mm0, &mm1, 1);
michael@0: 
michael@0: 	*(__m64 *)(dst + 0) = mm0;
michael@0: 	*(__m64 *)(dst + 2) = mm1;
michael@0: 
michael@0: 	dst += 4;
michael@0: 	src += 4;
michael@0: 	w -= 4;
michael@0:     }
michael@0: 
michael@0:     while (w)
michael@0:     {
michael@0: 	uint16_t s = *src++;
michael@0: 
michael@0: 	*dst++ = convert_0565_to_8888 (s);
michael@0: 	w--;
michael@0:     }
michael@0: 
michael@0:     _mm_empty ();
michael@0:     return iter->buffer;
michael@0: }
michael@0: 
michael@0: static uint32_t *
michael@0: mmx_fetch_a8 (pixman_iter_t *iter, const uint32_t *mask)
michael@0: {
michael@0:     int w = iter->width;
michael@0:     uint32_t *dst = iter->buffer;
michael@0:     uint8_t *src = iter->bits;
michael@0: 
michael@0:     iter->bits += iter->stride;
michael@0: 
michael@0:     while (w && (((uintptr_t)dst) & 15))
michael@0:     {
michael@0:         *dst++ = *(src++) << 24;
michael@0:         w--;
michael@0:     }
michael@0: 
michael@0:     while (w >= 8)
michael@0:     {
michael@0: 	__m64 mm0 = ldq_u ((__m64 *)src);
michael@0: 
michael@0: 	__m64 mm1 = _mm_unpacklo_pi8  (_mm_setzero_si64(), mm0);
michael@0: 	__m64 mm2 = _mm_unpackhi_pi8  (_mm_setzero_si64(), mm0);
michael@0: 	__m64 mm3 = _mm_unpacklo_pi16 (_mm_setzero_si64(), mm1);
michael@0: 	__m64 mm4 = _mm_unpackhi_pi16 (_mm_setzero_si64(), mm1);
michael@0: 	__m64 mm5 = _mm_unpacklo_pi16 (_mm_setzero_si64(), mm2);
michael@0: 	__m64 mm6 = _mm_unpackhi_pi16 (_mm_setzero_si64(), mm2);
michael@0: 
michael@0: 	*(__m64 *)(dst + 0) = mm3;
michael@0: 	*(__m64 *)(dst + 2) = mm4;
michael@0: 	*(__m64 *)(dst + 4) = mm5;
michael@0: 	*(__m64 *)(dst + 6) = mm6;
michael@0: 
michael@0: 	dst += 8;
michael@0: 	src += 8;
michael@0: 	w -= 8;
michael@0:     }
michael@0: 
michael@0:     while (w)
michael@0:     {
michael@0: 	*dst++ = *(src++) << 24;
michael@0: 	w--;
michael@0:     }
michael@0: 
michael@0:     _mm_empty ();
michael@0:     return iter->buffer;
michael@0: }
michael@0: 
michael@0: typedef struct
michael@0: {
michael@0:     pixman_format_code_t	format;
michael@0:     pixman_iter_get_scanline_t	get_scanline;
michael@0: } fetcher_info_t;
michael@0: 
michael@0: static const fetcher_info_t fetchers[] =
michael@0: {
michael@0:     { PIXMAN_x8r8g8b8,		mmx_fetch_x8r8g8b8 },
michael@0:     { PIXMAN_r5g6b5,		mmx_fetch_r5g6b5 },
michael@0:     { PIXMAN_a8,		mmx_fetch_a8 },
michael@0:     { PIXMAN_null }
michael@0: };
michael@0: 
michael@0: static pixman_bool_t
michael@0: mmx_src_iter_init (pixman_implementation_t *imp, pixman_iter_t *iter)
michael@0: {
michael@0:     pixman_image_t *image = iter->image;
michael@0: 
michael@0: #define FLAGS								\
michael@0:     (FAST_PATH_STANDARD_FLAGS | FAST_PATH_ID_TRANSFORM |		\
michael@0:      FAST_PATH_BITS_IMAGE | FAST_PATH_SAMPLES_COVER_CLIP_NEAREST)
michael@0: 
michael@0:     if ((iter->iter_flags & ITER_NARROW)			&&
michael@0: 	(iter->image_flags & FLAGS) == FLAGS)
michael@0:     {
michael@0: 	const fetcher_info_t *f;
michael@0: 
michael@0: 	for (f = &fetchers[0]; f->format != PIXMAN_null; f++)
michael@0: 	{
michael@0: 	    if (image->common.extended_format_code == f->format)
michael@0: 	    {
michael@0: 		uint8_t *b = (uint8_t *)image->bits.bits;
michael@0: 		int s = image->bits.rowstride * 4;
michael@0: 
michael@0: 		iter->bits = b + s * iter->y + iter->x * PIXMAN_FORMAT_BPP (f->format) / 8;
michael@0: 		iter->stride = s;
michael@0: 
michael@0: 		iter->get_scanline = f->get_scanline;
michael@0: 		return TRUE;
michael@0: 	    }
michael@0: 	}
michael@0:     }
michael@0: 
michael@0:     return FALSE;
michael@0: }
michael@0: 
michael@0: static const pixman_fast_path_t mmx_fast_paths[] =
michael@0: {
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, solid,    a8,       r5g6b5,   mmx_composite_over_n_8_0565       ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, solid,    a8,       b5g6r5,   mmx_composite_over_n_8_0565       ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, solid,    a8,       a8r8g8b8, mmx_composite_over_n_8_8888       ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, solid,    a8,       x8r8g8b8, mmx_composite_over_n_8_8888       ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, solid,    a8,       a8b8g8r8, mmx_composite_over_n_8_8888       ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, solid,    a8,       x8b8g8r8, mmx_composite_over_n_8_8888       ),
michael@0:     PIXMAN_STD_FAST_PATH_CA (OVER, solid,    a8r8g8b8, a8r8g8b8, mmx_composite_over_n_8888_8888_ca ),
michael@0:     PIXMAN_STD_FAST_PATH_CA (OVER, solid,    a8r8g8b8, x8r8g8b8, mmx_composite_over_n_8888_8888_ca ),
michael@0:     PIXMAN_STD_FAST_PATH_CA (OVER, solid,    a8r8g8b8, r5g6b5,   mmx_composite_over_n_8888_0565_ca ),
michael@0:     PIXMAN_STD_FAST_PATH_CA (OVER, solid,    a8b8g8r8, a8b8g8r8, mmx_composite_over_n_8888_8888_ca ),
michael@0:     PIXMAN_STD_FAST_PATH_CA (OVER, solid,    a8b8g8r8, x8b8g8r8, mmx_composite_over_n_8888_8888_ca ),
michael@0:     PIXMAN_STD_FAST_PATH_CA (OVER, solid,    a8b8g8r8, b5g6r5,   mmx_composite_over_n_8888_0565_ca ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, pixbuf,   pixbuf,   a8r8g8b8, mmx_composite_over_pixbuf_8888    ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, pixbuf,   pixbuf,   x8r8g8b8, mmx_composite_over_pixbuf_8888    ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, pixbuf,   pixbuf,   r5g6b5,   mmx_composite_over_pixbuf_0565    ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, rpixbuf,  rpixbuf,  a8b8g8r8, mmx_composite_over_pixbuf_8888    ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, rpixbuf,  rpixbuf,  x8b8g8r8, mmx_composite_over_pixbuf_8888    ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, rpixbuf,  rpixbuf,  b5g6r5,   mmx_composite_over_pixbuf_0565    ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, x8r8g8b8, solid,    a8r8g8b8, mmx_composite_over_x888_n_8888    ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, x8r8g8b8, solid,    x8r8g8b8, mmx_composite_over_x888_n_8888    ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, x8b8g8r8, solid,    a8b8g8r8, mmx_composite_over_x888_n_8888    ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, x8b8g8r8, solid,    x8b8g8r8, mmx_composite_over_x888_n_8888    ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, a8r8g8b8, solid,    a8r8g8b8, mmx_composite_over_8888_n_8888    ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, a8r8g8b8, solid,    x8r8g8b8, mmx_composite_over_8888_n_8888    ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, a8b8g8r8, solid,    a8b8g8r8, mmx_composite_over_8888_n_8888    ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, a8b8g8r8, solid,    x8b8g8r8, mmx_composite_over_8888_n_8888    ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, x8r8g8b8, a8,       x8r8g8b8, mmx_composite_over_x888_8_8888    ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, x8r8g8b8, a8,       a8r8g8b8, mmx_composite_over_x888_8_8888    ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, x8b8g8r8, a8,       x8b8g8r8, mmx_composite_over_x888_8_8888    ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, x8b8g8r8, a8,       a8b8g8r8, mmx_composite_over_x888_8_8888    ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, solid,    null,     a8r8g8b8, mmx_composite_over_n_8888         ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, solid,    null,     x8r8g8b8, mmx_composite_over_n_8888         ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, solid,    null,     r5g6b5,   mmx_composite_over_n_0565         ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, solid,    null,     b5g6r5,   mmx_composite_over_n_0565         ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, x8r8g8b8, null,     x8r8g8b8, mmx_composite_copy_area           ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, x8b8g8r8, null,     x8b8g8r8, mmx_composite_copy_area           ),
michael@0: 
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, a8r8g8b8, null,     a8r8g8b8, mmx_composite_over_8888_8888      ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, a8r8g8b8, null,     x8r8g8b8, mmx_composite_over_8888_8888      ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, a8r8g8b8, null,     r5g6b5,   mmx_composite_over_8888_0565      ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, a8b8g8r8, null,     a8b8g8r8, mmx_composite_over_8888_8888      ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, a8b8g8r8, null,     x8b8g8r8, mmx_composite_over_8888_8888      ),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER, a8b8g8r8, null,     b5g6r5,   mmx_composite_over_8888_0565      ),
michael@0: 
michael@0:     PIXMAN_STD_FAST_PATH    (OVER_REVERSE, solid, null, a8r8g8b8, mmx_composite_over_reverse_n_8888),
michael@0:     PIXMAN_STD_FAST_PATH    (OVER_REVERSE, solid, null, a8b8g8r8, mmx_composite_over_reverse_n_8888),
michael@0: 
michael@0:     PIXMAN_STD_FAST_PATH    (ADD,  r5g6b5,   null,     r5g6b5,   mmx_composite_add_0565_0565       ),
michael@0:     PIXMAN_STD_FAST_PATH    (ADD,  b5g6r5,   null,     b5g6r5,   mmx_composite_add_0565_0565       ),
michael@0:     PIXMAN_STD_FAST_PATH    (ADD,  a8r8g8b8, null,     a8r8g8b8, mmx_composite_add_8888_8888       ),
michael@0:     PIXMAN_STD_FAST_PATH    (ADD,  a8b8g8r8, null,     a8b8g8r8, mmx_composite_add_8888_8888       ),
michael@0:     PIXMAN_STD_FAST_PATH    (ADD,  a8,       null,     a8,       mmx_composite_add_8_8		   ),
michael@0:     PIXMAN_STD_FAST_PATH    (ADD,  solid,    a8,       a8,       mmx_composite_add_n_8_8           ),
michael@0: 
michael@0:     PIXMAN_STD_FAST_PATH    (SRC,  a8r8g8b8, null,     r5g6b5,   mmx_composite_src_x888_0565       ),
michael@0:     PIXMAN_STD_FAST_PATH    (SRC,  a8b8g8r8, null,     b5g6r5,   mmx_composite_src_x888_0565       ),
michael@0:     PIXMAN_STD_FAST_PATH    (SRC,  x8r8g8b8, null,     r5g6b5,   mmx_composite_src_x888_0565       ),
michael@0:     PIXMAN_STD_FAST_PATH    (SRC,  x8b8g8r8, null,     b5g6r5,   mmx_composite_src_x888_0565       ),
michael@0:     PIXMAN_STD_FAST_PATH    (SRC,  solid,    a8,       a8r8g8b8, mmx_composite_src_n_8_8888        ),
michael@0:     PIXMAN_STD_FAST_PATH    (SRC,  solid,    a8,       x8r8g8b8, mmx_composite_src_n_8_8888        ),
michael@0:     PIXMAN_STD_FAST_PATH    (SRC,  solid,    a8,       a8b8g8r8, mmx_composite_src_n_8_8888        ),
michael@0:     PIXMAN_STD_FAST_PATH    (SRC,  solid,    a8,       x8b8g8r8, mmx_composite_src_n_8_8888        ),
michael@0:     PIXMAN_STD_FAST_PATH    (SRC,  a8r8g8b8, null,     a8r8g8b8, mmx_composite_copy_area           ),
michael@0:     PIXMAN_STD_FAST_PATH    (SRC,  a8b8g8r8, null,     a8b8g8r8, mmx_composite_copy_area           ),
michael@0:     PIXMAN_STD_FAST_PATH    (SRC,  a8r8g8b8, null,     x8r8g8b8, mmx_composite_copy_area           ),
michael@0:     PIXMAN_STD_FAST_PATH    (SRC,  a8b8g8r8, null,     x8b8g8r8, mmx_composite_copy_area           ),
michael@0:     PIXMAN_STD_FAST_PATH    (SRC,  x8r8g8b8, null,     x8r8g8b8, mmx_composite_copy_area           ),
michael@0:     PIXMAN_STD_FAST_PATH    (SRC,  x8b8g8r8, null,     x8b8g8r8, mmx_composite_copy_area           ),
michael@0:     PIXMAN_STD_FAST_PATH    (SRC,  r5g6b5,   null,     r5g6b5,   mmx_composite_copy_area           ),
michael@0:     PIXMAN_STD_FAST_PATH    (SRC,  b5g6r5,   null,     b5g6r5,   mmx_composite_copy_area           ),
michael@0: 
michael@0:     PIXMAN_STD_FAST_PATH    (IN,   a8,       null,     a8,       mmx_composite_in_8_8              ),
michael@0:     PIXMAN_STD_FAST_PATH    (IN,   solid,    a8,       a8,       mmx_composite_in_n_8_8            ),
michael@0: 
michael@0:     SIMPLE_BILINEAR_FAST_PATH (SRC, a8r8g8b8,          a8r8g8b8, mmx_8888_8888                     ),
michael@0:     SIMPLE_BILINEAR_FAST_PATH (SRC, a8r8g8b8,          x8r8g8b8, mmx_8888_8888                     ),
michael@0:     SIMPLE_BILINEAR_FAST_PATH (SRC, x8r8g8b8,          x8r8g8b8, mmx_8888_8888                     ),
michael@0:     SIMPLE_BILINEAR_FAST_PATH (SRC, a8b8g8r8,          a8b8g8r8, mmx_8888_8888                     ),
michael@0:     SIMPLE_BILINEAR_FAST_PATH (SRC, a8b8g8r8,          x8b8g8r8, mmx_8888_8888                     ),
michael@0:     SIMPLE_BILINEAR_FAST_PATH (SRC, x8b8g8r8,          x8b8g8r8, mmx_8888_8888                     ),
michael@0: 
michael@0:     SIMPLE_BILINEAR_FAST_PATH (OVER, a8r8g8b8,         x8r8g8b8, mmx_8888_8888                     ),
michael@0:     SIMPLE_BILINEAR_FAST_PATH (OVER, a8b8g8r8,         x8b8g8r8, mmx_8888_8888                     ),
michael@0:     SIMPLE_BILINEAR_FAST_PATH (OVER, a8r8g8b8,         a8r8g8b8, mmx_8888_8888                     ),
michael@0:     SIMPLE_BILINEAR_FAST_PATH (OVER, a8b8g8r8,         a8b8g8r8, mmx_8888_8888                     ),
michael@0: 
michael@0:     SIMPLE_BILINEAR_A8_MASK_FAST_PATH (OVER, a8r8g8b8, x8r8g8b8, mmx_8888_8_8888                   ),
michael@0:     SIMPLE_BILINEAR_A8_MASK_FAST_PATH (OVER, a8b8g8r8, x8b8g8r8, mmx_8888_8_8888                   ),
michael@0:     SIMPLE_BILINEAR_A8_MASK_FAST_PATH (OVER, a8r8g8b8, a8r8g8b8, mmx_8888_8_8888                   ),
michael@0:     SIMPLE_BILINEAR_A8_MASK_FAST_PATH (OVER, a8b8g8r8, a8b8g8r8, mmx_8888_8_8888                   ),
michael@0: 
michael@0:     { PIXMAN_OP_NONE },
michael@0: };
michael@0: 
michael@0: pixman_implementation_t *
michael@0: _pixman_implementation_create_mmx (pixman_implementation_t *fallback)
michael@0: {
michael@0:     pixman_implementation_t *imp = _pixman_implementation_create (fallback, mmx_fast_paths);
michael@0: 
michael@0:     imp->combine_32[PIXMAN_OP_OVER] = mmx_combine_over_u;
michael@0:     imp->combine_32[PIXMAN_OP_OVER_REVERSE] = mmx_combine_over_reverse_u;
michael@0:     imp->combine_32[PIXMAN_OP_IN] = mmx_combine_in_u;
michael@0:     imp->combine_32[PIXMAN_OP_IN_REVERSE] = mmx_combine_in_reverse_u;
michael@0:     imp->combine_32[PIXMAN_OP_OUT] = mmx_combine_out_u;
michael@0:     imp->combine_32[PIXMAN_OP_OUT_REVERSE] = mmx_combine_out_reverse_u;
michael@0:     imp->combine_32[PIXMAN_OP_ATOP] = mmx_combine_atop_u;
michael@0:     imp->combine_32[PIXMAN_OP_ATOP_REVERSE] = mmx_combine_atop_reverse_u;
michael@0:     imp->combine_32[PIXMAN_OP_XOR] = mmx_combine_xor_u;
michael@0:     imp->combine_32[PIXMAN_OP_ADD] = mmx_combine_add_u;
michael@0:     imp->combine_32[PIXMAN_OP_SATURATE] = mmx_combine_saturate_u;
michael@0: 
michael@0:     imp->combine_32_ca[PIXMAN_OP_SRC] = mmx_combine_src_ca;
michael@0:     imp->combine_32_ca[PIXMAN_OP_OVER] = mmx_combine_over_ca;
michael@0:     imp->combine_32_ca[PIXMAN_OP_OVER_REVERSE] = mmx_combine_over_reverse_ca;
michael@0:     imp->combine_32_ca[PIXMAN_OP_IN] = mmx_combine_in_ca;
michael@0:     imp->combine_32_ca[PIXMAN_OP_IN_REVERSE] = mmx_combine_in_reverse_ca;
michael@0:     imp->combine_32_ca[PIXMAN_OP_OUT] = mmx_combine_out_ca;
michael@0:     imp->combine_32_ca[PIXMAN_OP_OUT_REVERSE] = mmx_combine_out_reverse_ca;
michael@0:     imp->combine_32_ca[PIXMAN_OP_ATOP] = mmx_combine_atop_ca;
michael@0:     imp->combine_32_ca[PIXMAN_OP_ATOP_REVERSE] = mmx_combine_atop_reverse_ca;
michael@0:     imp->combine_32_ca[PIXMAN_OP_XOR] = mmx_combine_xor_ca;
michael@0:     imp->combine_32_ca[PIXMAN_OP_ADD] = mmx_combine_add_ca;
michael@0: 
michael@0:     imp->blt = mmx_blt;
michael@0:     imp->fill = mmx_fill;
michael@0: 
michael@0:     imp->src_iter_init = mmx_src_iter_init;
michael@0: 
michael@0:     return imp;
michael@0: }
michael@0: 
michael@0: #endif /* USE_X86_MMX || USE_ARM_IWMMXT || USE_LOONGSON_MMI */