michael@0: #include "SkXfermode.h"
michael@0: #include "SkXfermode_proccoeff.h"
michael@0: #include "SkColorPriv.h"
michael@0: 
michael@0: #include <arm_neon.h>
michael@0: #include "SkColor_opts_neon.h"
michael@0: #include "SkXfermode_opts_arm_neon.h"
michael@0: 
michael@0: #define SkAlphaMulAlpha(a, b)   SkMulDiv255Round(a, b)
michael@0: 
michael@0: 
michael@0: ////////////////////////////////////////////////////////////////////////////////
michael@0: // NEONized skia functions
michael@0: ////////////////////////////////////////////////////////////////////////////////
michael@0: 
michael@0: static inline uint8x8_t SkAlphaMulAlpha_neon8(uint8x8_t color, uint8x8_t alpha) {
michael@0:     uint16x8_t tmp;
michael@0:     uint8x8_t ret;
michael@0: 
michael@0:     tmp = vmull_u8(color, alpha);
michael@0:     tmp = vaddq_u16(tmp, vdupq_n_u16(128));
michael@0:     tmp = vaddq_u16(tmp, vshrq_n_u16(tmp, 8));
michael@0: 
michael@0:     ret = vshrn_n_u16(tmp, 8);
michael@0: 
michael@0:     return ret;
michael@0: }
michael@0: 
michael@0: static inline uint16x8_t SkAlphaMulAlpha_neon8_16(uint8x8_t color, uint8x8_t alpha) {
michael@0:     uint16x8_t ret;
michael@0: 
michael@0:     ret = vmull_u8(color, alpha);
michael@0:     ret = vaddq_u16(ret, vdupq_n_u16(128));
michael@0:     ret = vaddq_u16(ret, vshrq_n_u16(ret, 8));
michael@0: 
michael@0:     ret = vshrq_n_u16(ret, 8);
michael@0: 
michael@0:     return ret;
michael@0: }
michael@0: 
michael@0: static inline uint8x8_t SkDiv255Round_neon8_32_8(int32x4_t p1, int32x4_t p2) {
michael@0:     uint16x8_t tmp;
michael@0: 
michael@0:     tmp = vcombine_u16(vmovn_u32(vreinterpretq_u32_s32(p1)),
michael@0:                        vmovn_u32(vreinterpretq_u32_s32(p2)));
michael@0: 
michael@0:     tmp += vdupq_n_u16(128);
michael@0:     tmp += vshrq_n_u16(tmp, 8);
michael@0: 
michael@0:     return vshrn_n_u16(tmp, 8);
michael@0: }
michael@0: 
michael@0: static inline uint16x8_t SkDiv255Round_neon8_16_16(uint16x8_t prod) {
michael@0:     prod += vdupq_n_u16(128);
michael@0:     prod += vshrq_n_u16(prod, 8);
michael@0: 
michael@0:     return vshrq_n_u16(prod, 8);
michael@0: }
michael@0: 
michael@0: static inline uint8x8_t clamp_div255round_simd8_32(int32x4_t val1, int32x4_t val2) {
michael@0:     uint8x8_t ret;
michael@0:     uint32x4_t cmp1, cmp2;
michael@0:     uint16x8_t cmp16;
michael@0:     uint8x8_t cmp8, cmp8_1;
michael@0: 
michael@0:     // Test if <= 0
michael@0:     cmp1 = vcleq_s32(val1, vdupq_n_s32(0));
michael@0:     cmp2 = vcleq_s32(val2, vdupq_n_s32(0));
michael@0:     cmp16 = vcombine_u16(vmovn_u32(cmp1), vmovn_u32(cmp2));
michael@0:     cmp8_1 = vmovn_u16(cmp16);
michael@0: 
michael@0:     // Init to zero
michael@0:     ret = vdup_n_u8(0);
michael@0: 
michael@0:     // Test if >= 255*255
michael@0:     cmp1 = vcgeq_s32(val1, vdupq_n_s32(255*255));
michael@0:     cmp2 = vcgeq_s32(val2, vdupq_n_s32(255*255));
michael@0:     cmp16 = vcombine_u16(vmovn_u32(cmp1), vmovn_u32(cmp2));
michael@0:     cmp8 = vmovn_u16(cmp16);
michael@0: 
michael@0:     // Insert 255 where true
michael@0:     ret = vbsl_u8(cmp8, vdup_n_u8(255), ret);
michael@0: 
michael@0:     // Calc SkDiv255Round
michael@0:     uint8x8_t div = SkDiv255Round_neon8_32_8(val1, val2);
michael@0: 
michael@0:     // Insert where false and previous test false
michael@0:     cmp8 = cmp8 | cmp8_1;
michael@0:     ret = vbsl_u8(cmp8, ret, div);
michael@0: 
michael@0:     // Return the final combination
michael@0:     return ret;
michael@0: }
michael@0: 
michael@0: ////////////////////////////////////////////////////////////////////////////////
michael@0: // 1 pixel modeprocs
michael@0: ////////////////////////////////////////////////////////////////////////////////
michael@0: 
michael@0: //  kSrcATop_Mode,  //!< [Da, Sc * Da + (1 - Sa) * Dc]
michael@0: SkPMColor srcatop_modeproc_neon(SkPMColor src, SkPMColor dst) {
michael@0:     unsigned sa = SkGetPackedA32(src);
michael@0:     unsigned da = SkGetPackedA32(dst);
michael@0:     unsigned isa = 255 - sa;
michael@0: 
michael@0:     uint8x8_t vda, visa, vsrc, vdst;
michael@0: 
michael@0:     vda = vdup_n_u8(da);
michael@0:     visa = vdup_n_u8(isa);
michael@0: 
michael@0:     uint16x8_t vsrc_wide, vdst_wide;
michael@0:     vsrc_wide = vmull_u8(vda, vreinterpret_u8_u32(vdup_n_u32(src)));
michael@0:     vdst_wide = vmull_u8(visa, vreinterpret_u8_u32(vdup_n_u32(dst)));
michael@0: 
michael@0:     vsrc_wide += vdupq_n_u16(128);
michael@0:     vsrc_wide += vshrq_n_u16(vsrc_wide, 8);
michael@0: 
michael@0:     vdst_wide += vdupq_n_u16(128);
michael@0:     vdst_wide += vshrq_n_u16(vdst_wide, 8);
michael@0: 
michael@0:     vsrc = vshrn_n_u16(vsrc_wide, 8);
michael@0:     vdst = vshrn_n_u16(vdst_wide, 8);
michael@0: 
michael@0:     vsrc += vdst;
michael@0:     vsrc = vset_lane_u8(da, vsrc, 3);
michael@0: 
michael@0:     return vget_lane_u32(vreinterpret_u32_u8(vsrc), 0);
michael@0: }
michael@0: 
michael@0: //  kDstATop_Mode,  //!< [Sa, Sa * Dc + Sc * (1 - Da)]
michael@0: SkPMColor dstatop_modeproc_neon(SkPMColor src, SkPMColor dst) {
michael@0:     unsigned sa = SkGetPackedA32(src);
michael@0:     unsigned da = SkGetPackedA32(dst);
michael@0:     unsigned ida = 255 - da;
michael@0: 
michael@0:     uint8x8_t vsa, vida, vsrc, vdst;
michael@0: 
michael@0:     vsa = vdup_n_u8(sa);
michael@0:     vida = vdup_n_u8(ida);
michael@0: 
michael@0:     uint16x8_t vsrc_wide, vdst_wide;
michael@0:     vsrc_wide = vmull_u8(vida, vreinterpret_u8_u32(vdup_n_u32(src)));
michael@0:     vdst_wide = vmull_u8(vsa, vreinterpret_u8_u32(vdup_n_u32(dst)));
michael@0: 
michael@0:     vsrc_wide += vdupq_n_u16(128);
michael@0:     vsrc_wide += vshrq_n_u16(vsrc_wide, 8);
michael@0: 
michael@0:     vdst_wide += vdupq_n_u16(128);
michael@0:     vdst_wide += vshrq_n_u16(vdst_wide, 8);
michael@0: 
michael@0:     vsrc = vshrn_n_u16(vsrc_wide, 8);
michael@0:     vdst = vshrn_n_u16(vdst_wide, 8);
michael@0: 
michael@0:     vsrc += vdst;
michael@0:     vsrc = vset_lane_u8(sa, vsrc, 3);
michael@0: 
michael@0:     return vget_lane_u32(vreinterpret_u32_u8(vsrc), 0);
michael@0: }
michael@0: 
michael@0: //  kXor_Mode   [Sa + Da - 2 * Sa * Da, Sc * (1 - Da) + (1 - Sa) * Dc]
michael@0: SkPMColor xor_modeproc_neon(SkPMColor src, SkPMColor dst) {
michael@0:     unsigned sa = SkGetPackedA32(src);
michael@0:     unsigned da = SkGetPackedA32(dst);
michael@0:     unsigned ret_alpha = sa + da - (SkAlphaMulAlpha(sa, da) << 1);
michael@0:     unsigned isa = 255 - sa;
michael@0:     unsigned ida = 255 - da;
michael@0: 
michael@0:     uint8x8_t vsrc, vdst, visa, vida;
michael@0:     uint16x8_t vsrc_wide, vdst_wide;
michael@0: 
michael@0:     visa = vdup_n_u8(isa);
michael@0:     vida = vdup_n_u8(ida);
michael@0:     vsrc = vreinterpret_u8_u32(vdup_n_u32(src));
michael@0:     vdst = vreinterpret_u8_u32(vdup_n_u32(dst));
michael@0: 
michael@0:     vsrc_wide = vmull_u8(vsrc, vida);
michael@0:     vdst_wide = vmull_u8(vdst, visa);
michael@0: 
michael@0:     vsrc_wide += vdupq_n_u16(128);
michael@0:     vsrc_wide += vshrq_n_u16(vsrc_wide, 8);
michael@0: 
michael@0:     vdst_wide += vdupq_n_u16(128);
michael@0:     vdst_wide += vshrq_n_u16(vdst_wide, 8);
michael@0: 
michael@0:     vsrc = vshrn_n_u16(vsrc_wide, 8);
michael@0:     vdst = vshrn_n_u16(vdst_wide, 8);
michael@0: 
michael@0:     vsrc += vdst;
michael@0: 
michael@0:     vsrc = vset_lane_u8(ret_alpha, vsrc, 3);
michael@0: 
michael@0:     return vget_lane_u32(vreinterpret_u32_u8(vsrc), 0);
michael@0: }
michael@0: 
michael@0: // kPlus_Mode
michael@0: SkPMColor plus_modeproc_neon(SkPMColor src, SkPMColor dst) {
michael@0:     uint8x8_t vsrc, vdst;
michael@0:     vsrc = vreinterpret_u8_u32(vdup_n_u32(src));
michael@0:     vdst = vreinterpret_u8_u32(vdup_n_u32(dst));
michael@0:     vsrc = vqadd_u8(vsrc, vdst);
michael@0: 
michael@0:     return vget_lane_u32(vreinterpret_u32_u8(vsrc), 0);
michael@0: }
michael@0: 
michael@0: // kModulate_Mode
michael@0: SkPMColor modulate_modeproc_neon(SkPMColor src, SkPMColor dst) {
michael@0:     uint8x8_t vsrc, vdst, vres;
michael@0:     uint16x8_t vres_wide;
michael@0: 
michael@0:     vsrc = vreinterpret_u8_u32(vdup_n_u32(src));
michael@0:     vdst = vreinterpret_u8_u32(vdup_n_u32(dst));
michael@0: 
michael@0:     vres_wide = vmull_u8(vsrc, vdst);
michael@0: 
michael@0:     vres_wide += vdupq_n_u16(128);
michael@0:     vres_wide += vshrq_n_u16(vres_wide, 8);
michael@0: 
michael@0:     vres = vshrn_n_u16(vres_wide, 8);
michael@0: 
michael@0:     return vget_lane_u32(vreinterpret_u32_u8(vres), 0);
michael@0: }
michael@0: 
michael@0: ////////////////////////////////////////////////////////////////////////////////
michael@0: // 8 pixels modeprocs
michael@0: ////////////////////////////////////////////////////////////////////////////////
michael@0: 
michael@0: uint8x8x4_t dstover_modeproc_neon8(uint8x8x4_t src, uint8x8x4_t dst) {
michael@0:     uint8x8x4_t ret;
michael@0:     uint16x8_t src_scale;
michael@0: 
michael@0:     src_scale = vsubw_u8(vdupq_n_u16(256), dst.val[NEON_A]);
michael@0: 
michael@0:     ret.val[NEON_A] = dst.val[NEON_A] + SkAlphaMul_neon8(src.val[NEON_A], src_scale);
michael@0:     ret.val[NEON_R] = dst.val[NEON_R] + SkAlphaMul_neon8(src.val[NEON_R], src_scale);
michael@0:     ret.val[NEON_G] = dst.val[NEON_G] + SkAlphaMul_neon8(src.val[NEON_G], src_scale);
michael@0:     ret.val[NEON_B] = dst.val[NEON_B] + SkAlphaMul_neon8(src.val[NEON_B], src_scale);
michael@0: 
michael@0:     return ret;
michael@0: }
michael@0: 
michael@0: uint8x8x4_t srcin_modeproc_neon8(uint8x8x4_t src, uint8x8x4_t dst) {
michael@0:     uint8x8x4_t ret;
michael@0:     uint16x8_t scale;
michael@0: 
michael@0:     scale = SkAlpha255To256_neon8(dst.val[NEON_A]);
michael@0: 
michael@0:     ret.val[NEON_A] = SkAlphaMul_neon8(src.val[NEON_A], scale);
michael@0:     ret.val[NEON_R] = SkAlphaMul_neon8(src.val[NEON_R], scale);
michael@0:     ret.val[NEON_G] = SkAlphaMul_neon8(src.val[NEON_G], scale);
michael@0:     ret.val[NEON_B] = SkAlphaMul_neon8(src.val[NEON_B], scale);
michael@0: 
michael@0:     return ret;
michael@0: }
michael@0: 
michael@0: uint8x8x4_t dstin_modeproc_neon8(uint8x8x4_t src, uint8x8x4_t dst) {
michael@0:     uint8x8x4_t ret;
michael@0:     uint16x8_t scale;
michael@0: 
michael@0:     scale = SkAlpha255To256_neon8(src.val[NEON_A]);
michael@0: 
michael@0:     ret = SkAlphaMulQ_neon8(dst, scale);
michael@0: 
michael@0:     return ret;
michael@0: }
michael@0: 
michael@0: uint8x8x4_t srcout_modeproc_neon8(uint8x8x4_t src, uint8x8x4_t dst) {
michael@0:     uint8x8x4_t ret;
michael@0:     uint16x8_t scale = vsubw_u8(vdupq_n_u16(256), dst.val[NEON_A]);
michael@0: 
michael@0:     ret = SkAlphaMulQ_neon8(src, scale);
michael@0: 
michael@0:     return ret;
michael@0: }
michael@0: 
michael@0: uint8x8x4_t dstout_modeproc_neon8(uint8x8x4_t src, uint8x8x4_t dst) {
michael@0:     uint8x8x4_t ret;
michael@0:     uint16x8_t scale = vsubw_u8(vdupq_n_u16(256), src.val[NEON_A]);
michael@0: 
michael@0:     ret = SkAlphaMulQ_neon8(dst, scale);
michael@0: 
michael@0:     return ret;
michael@0: }
michael@0: 
michael@0: uint8x8x4_t srcatop_modeproc_neon8(uint8x8x4_t src, uint8x8x4_t dst) {
michael@0:     uint8x8x4_t ret;
michael@0:     uint8x8_t isa;
michael@0: 
michael@0:     isa = vsub_u8(vdup_n_u8(255), src.val[NEON_A]);
michael@0: 
michael@0:     ret.val[NEON_A] = dst.val[NEON_A];
michael@0:     ret.val[NEON_R] = SkAlphaMulAlpha_neon8(src.val[NEON_R], dst.val[NEON_A])
michael@0:                       + SkAlphaMulAlpha_neon8(dst.val[NEON_R], isa);
michael@0:     ret.val[NEON_G] = SkAlphaMulAlpha_neon8(src.val[NEON_G], dst.val[NEON_A])
michael@0:                       + SkAlphaMulAlpha_neon8(dst.val[NEON_G], isa);
michael@0:     ret.val[NEON_B] = SkAlphaMulAlpha_neon8(src.val[NEON_B], dst.val[NEON_A])
michael@0:                       + SkAlphaMulAlpha_neon8(dst.val[NEON_B], isa);
michael@0: 
michael@0:     return ret;
michael@0: }
michael@0: 
michael@0: uint8x8x4_t dstatop_modeproc_neon8(uint8x8x4_t src, uint8x8x4_t dst) {
michael@0:     uint8x8x4_t ret;
michael@0:     uint8x8_t ida;
michael@0: 
michael@0:     ida = vsub_u8(vdup_n_u8(255), dst.val[NEON_A]);
michael@0: 
michael@0:     ret.val[NEON_A] = src.val[NEON_A];
michael@0:     ret.val[NEON_R] = SkAlphaMulAlpha_neon8(src.val[NEON_R], ida)
michael@0:                       + SkAlphaMulAlpha_neon8(dst.val[NEON_R], src.val[NEON_A]);
michael@0:     ret.val[NEON_G] = SkAlphaMulAlpha_neon8(src.val[NEON_G], ida)
michael@0:                       + SkAlphaMulAlpha_neon8(dst.val[NEON_G], src.val[NEON_A]);
michael@0:     ret.val[NEON_B] = SkAlphaMulAlpha_neon8(src.val[NEON_B], ida)
michael@0:                       + SkAlphaMulAlpha_neon8(dst.val[NEON_B], src.val[NEON_A]);
michael@0: 
michael@0:     return ret;
michael@0: }
michael@0: 
michael@0: uint8x8x4_t xor_modeproc_neon8(uint8x8x4_t src, uint8x8x4_t dst) {
michael@0:     uint8x8x4_t ret;
michael@0:     uint8x8_t isa, ida;
michael@0:     uint16x8_t tmp_wide, tmp_wide2;
michael@0: 
michael@0:     isa = vsub_u8(vdup_n_u8(255), src.val[NEON_A]);
michael@0:     ida = vsub_u8(vdup_n_u8(255), dst.val[NEON_A]);
michael@0: 
michael@0:     // First calc alpha
michael@0:     tmp_wide = vmovl_u8(src.val[NEON_A]);
michael@0:     tmp_wide = vaddw_u8(tmp_wide, dst.val[NEON_A]);
michael@0:     tmp_wide2 = vshll_n_u8(SkAlphaMulAlpha_neon8(src.val[NEON_A], dst.val[NEON_A]), 1);
michael@0:     tmp_wide = vsubq_u16(tmp_wide, tmp_wide2);
michael@0:     ret.val[NEON_A] = vmovn_u16(tmp_wide);
michael@0: 
michael@0:     // Then colors
michael@0:     ret.val[NEON_R] = SkAlphaMulAlpha_neon8(src.val[NEON_R], ida)
michael@0:                       + SkAlphaMulAlpha_neon8(dst.val[NEON_R], isa);
michael@0:     ret.val[NEON_G] = SkAlphaMulAlpha_neon8(src.val[NEON_G], ida)
michael@0:                       + SkAlphaMulAlpha_neon8(dst.val[NEON_G], isa);
michael@0:     ret.val[NEON_B] = SkAlphaMulAlpha_neon8(src.val[NEON_B], ida)
michael@0:                       + SkAlphaMulAlpha_neon8(dst.val[NEON_B], isa);
michael@0: 
michael@0:     return ret;
michael@0: }
michael@0: 
michael@0: uint8x8x4_t plus_modeproc_neon8(uint8x8x4_t src, uint8x8x4_t dst) {
michael@0:     uint8x8x4_t ret;
michael@0: 
michael@0:     ret.val[NEON_A] = vqadd_u8(src.val[NEON_A], dst.val[NEON_A]);
michael@0:     ret.val[NEON_R] = vqadd_u8(src.val[NEON_R], dst.val[NEON_R]);
michael@0:     ret.val[NEON_G] = vqadd_u8(src.val[NEON_G], dst.val[NEON_G]);
michael@0:     ret.val[NEON_B] = vqadd_u8(src.val[NEON_B], dst.val[NEON_B]);
michael@0: 
michael@0:     return ret;
michael@0: }
michael@0: 
michael@0: uint8x8x4_t modulate_modeproc_neon8(uint8x8x4_t src, uint8x8x4_t dst) {
michael@0:     uint8x8x4_t ret;
michael@0: 
michael@0:     ret.val[NEON_A] = SkAlphaMulAlpha_neon8(src.val[NEON_A], dst.val[NEON_A]);
michael@0:     ret.val[NEON_R] = SkAlphaMulAlpha_neon8(src.val[NEON_R], dst.val[NEON_R]);
michael@0:     ret.val[NEON_G] = SkAlphaMulAlpha_neon8(src.val[NEON_G], dst.val[NEON_G]);
michael@0:     ret.val[NEON_B] = SkAlphaMulAlpha_neon8(src.val[NEON_B], dst.val[NEON_B]);
michael@0: 
michael@0:     return ret;
michael@0: }
michael@0: 
michael@0: static inline uint8x8_t srcover_color(uint8x8_t a, uint8x8_t b) {
michael@0:     uint16x8_t tmp;
michael@0: 
michael@0:     tmp = vaddl_u8(a, b);
michael@0:     tmp -= SkAlphaMulAlpha_neon8_16(a, b);
michael@0: 
michael@0:     return vmovn_u16(tmp);
michael@0: }
michael@0: 
michael@0: uint8x8x4_t screen_modeproc_neon8(uint8x8x4_t src, uint8x8x4_t dst) {
michael@0:     uint8x8x4_t ret;
michael@0: 
michael@0:     ret.val[NEON_A] = srcover_color(src.val[NEON_A], dst.val[NEON_A]);
michael@0:     ret.val[NEON_R] = srcover_color(src.val[NEON_R], dst.val[NEON_R]);
michael@0:     ret.val[NEON_G] = srcover_color(src.val[NEON_G], dst.val[NEON_G]);
michael@0:     ret.val[NEON_B] = srcover_color(src.val[NEON_B], dst.val[NEON_B]);
michael@0: 
michael@0:     return ret;
michael@0: }
michael@0: 
michael@0: template <bool overlay>
michael@0: static inline uint8x8_t overlay_hardlight_color(uint8x8_t sc, uint8x8_t dc,
michael@0:                                                uint8x8_t sa, uint8x8_t da) {
michael@0:     /*
michael@0:      * In the end we're gonna use (rc + tmp) with a different rc
michael@0:      * coming from an alternative.
michael@0:      * The whole value (rc + tmp) can always be expressed as
michael@0:      * VAL = COM - SUB in the if case
michael@0:      * VAL = COM + SUB - sa*da in the else case
michael@0:      *
michael@0:      * with COM = 255 * (sc + dc)
michael@0:      * and  SUB = sc*da + dc*sa - 2*dc*sc
michael@0:      */
michael@0: 
michael@0:     // Prepare common subexpressions
michael@0:     uint16x8_t const255 = vdupq_n_u16(255);
michael@0:     uint16x8_t sc_plus_dc = vaddl_u8(sc, dc);
michael@0:     uint16x8_t scda = vmull_u8(sc, da);
michael@0:     uint16x8_t dcsa = vmull_u8(dc, sa);
michael@0:     uint16x8_t sada = vmull_u8(sa, da);
michael@0: 
michael@0:     // Prepare non common subexpressions
michael@0:     uint16x8_t dc2, sc2;
michael@0:     uint32x4_t scdc2_1, scdc2_2;
michael@0:     if (overlay) {
michael@0:         dc2 = vshll_n_u8(dc, 1);
michael@0:         scdc2_1 = vmull_u16(vget_low_u16(dc2), vget_low_u16(vmovl_u8(sc)));
michael@0:         scdc2_2 = vmull_u16(vget_high_u16(dc2), vget_high_u16(vmovl_u8(sc)));
michael@0:     } else {
michael@0:         sc2 = vshll_n_u8(sc, 1);
michael@0:         scdc2_1 = vmull_u16(vget_low_u16(sc2), vget_low_u16(vmovl_u8(dc)));
michael@0:         scdc2_2 = vmull_u16(vget_high_u16(sc2), vget_high_u16(vmovl_u8(dc)));
michael@0:     }
michael@0: 
michael@0:     // Calc COM
michael@0:     int32x4_t com1, com2;
michael@0:     com1 = vreinterpretq_s32_u32(
michael@0:                 vmull_u16(vget_low_u16(const255), vget_low_u16(sc_plus_dc)));
michael@0:     com2 = vreinterpretq_s32_u32(
michael@0:                 vmull_u16(vget_high_u16(const255), vget_high_u16(sc_plus_dc)));
michael@0: 
michael@0:     // Calc SUB
michael@0:     int32x4_t sub1, sub2;
michael@0:     sub1 = vreinterpretq_s32_u32(vaddl_u16(vget_low_u16(scda), vget_low_u16(dcsa)));
michael@0:     sub2 = vreinterpretq_s32_u32(vaddl_u16(vget_high_u16(scda), vget_high_u16(dcsa)));
michael@0:     sub1 = vsubq_s32(sub1, vreinterpretq_s32_u32(scdc2_1));
michael@0:     sub2 = vsubq_s32(sub2, vreinterpretq_s32_u32(scdc2_2));
michael@0: 
michael@0:     // Compare 2*dc <= da
michael@0:     uint16x8_t cmp;
michael@0: 
michael@0:     if (overlay) {
michael@0:         cmp = vcleq_u16(dc2, vmovl_u8(da));
michael@0:     } else {
michael@0:         cmp = vcleq_u16(sc2, vmovl_u8(sa));
michael@0:     }
michael@0: 
michael@0:     // Prepare variables
michael@0:     int32x4_t val1_1, val1_2;
michael@0:     int32x4_t val2_1, val2_2;
michael@0:     uint32x4_t cmp1, cmp2;
michael@0: 
michael@0:     cmp1 = vmovl_u16(vget_low_u16(cmp));
michael@0:     cmp1 |= vshlq_n_u32(cmp1, 16);
michael@0:     cmp2 = vmovl_u16(vget_high_u16(cmp));
michael@0:     cmp2 |= vshlq_n_u32(cmp2, 16);
michael@0: 
michael@0:     // Calc COM - SUB
michael@0:     val1_1 = com1 - sub1;
michael@0:     val1_2 = com2 - sub2;
michael@0: 
michael@0:     // Calc COM + SUB - sa*da
michael@0:     val2_1 = com1 + sub1;
michael@0:     val2_2 = com2 + sub2;
michael@0: 
michael@0:     val2_1 = vsubq_s32(val2_1, vreinterpretq_s32_u32(vmovl_u16(vget_low_u16(sada))));
michael@0:     val2_2 = vsubq_s32(val2_2, vreinterpretq_s32_u32(vmovl_u16(vget_high_u16(sada))));
michael@0: 
michael@0:     // Insert where needed
michael@0:     val1_1 = vbslq_s32(cmp1, val1_1, val2_1);
michael@0:     val1_2 = vbslq_s32(cmp2, val1_2, val2_2);
michael@0: 
michael@0:     // Call the clamp_div255round function
michael@0:     return clamp_div255round_simd8_32(val1_1, val1_2);
michael@0: }
michael@0: 
michael@0: static inline uint8x8_t overlay_color(uint8x8_t sc, uint8x8_t dc,
michael@0:                                       uint8x8_t sa, uint8x8_t da) {
michael@0:     return overlay_hardlight_color<true>(sc, dc, sa, da);
michael@0: }
michael@0: 
michael@0: uint8x8x4_t overlay_modeproc_neon8(uint8x8x4_t src, uint8x8x4_t dst) {
michael@0:     uint8x8x4_t ret;
michael@0: 
michael@0:     ret.val[NEON_A] = srcover_color(src.val[NEON_A], dst.val[NEON_A]);
michael@0:     ret.val[NEON_R] = overlay_color(src.val[NEON_R], dst.val[NEON_R],
michael@0:                                     src.val[NEON_A], dst.val[NEON_A]);
michael@0:     ret.val[NEON_G] = overlay_color(src.val[NEON_G], dst.val[NEON_G],
michael@0:                                     src.val[NEON_A], dst.val[NEON_A]);
michael@0:     ret.val[NEON_B] = overlay_color(src.val[NEON_B], dst.val[NEON_B],
michael@0:                                     src.val[NEON_A], dst.val[NEON_A]);
michael@0: 
michael@0:     return ret;
michael@0: }
michael@0: 
michael@0: template <bool lighten>
michael@0: static inline uint8x8_t lighten_darken_color(uint8x8_t sc, uint8x8_t dc,
michael@0:                                              uint8x8_t sa, uint8x8_t da) {
michael@0:     uint16x8_t sd, ds, cmp, tmp, tmp2;
michael@0: 
michael@0:     // Prepare
michael@0:     sd = vmull_u8(sc, da);
michael@0:     ds = vmull_u8(dc, sa);
michael@0: 
michael@0:     // Do test
michael@0:     if (lighten) {
michael@0:         cmp = vcgtq_u16(sd, ds);
michael@0:     } else {
michael@0:         cmp = vcltq_u16(sd, ds);
michael@0:     }
michael@0: 
michael@0:     // Assign if
michael@0:     tmp = vaddl_u8(sc, dc);
michael@0:     tmp2 = tmp;
michael@0:     tmp -= SkDiv255Round_neon8_16_16(ds);
michael@0: 
michael@0:     // Calc else
michael@0:     tmp2 -= SkDiv255Round_neon8_16_16(sd);
michael@0: 
michael@0:     // Insert where needed
michael@0:     tmp = vbslq_u16(cmp, tmp, tmp2);
michael@0: 
michael@0:     return vmovn_u16(tmp);
michael@0: }
michael@0: 
michael@0: static inline uint8x8_t darken_color(uint8x8_t sc, uint8x8_t dc,
michael@0:                                      uint8x8_t sa, uint8x8_t da) {
michael@0:     return lighten_darken_color<false>(sc, dc, sa, da);
michael@0: }
michael@0: 
michael@0: uint8x8x4_t darken_modeproc_neon8(uint8x8x4_t src, uint8x8x4_t dst) {
michael@0:     uint8x8x4_t ret;
michael@0: 
michael@0:     ret.val[NEON_A] = srcover_color(src.val[NEON_A], dst.val[NEON_A]);
michael@0:     ret.val[NEON_R] = darken_color(src.val[NEON_R], dst.val[NEON_R],
michael@0:                                    src.val[NEON_A], dst.val[NEON_A]);
michael@0:     ret.val[NEON_G] = darken_color(src.val[NEON_G], dst.val[NEON_G],
michael@0:                                    src.val[NEON_A], dst.val[NEON_A]);
michael@0:     ret.val[NEON_B] = darken_color(src.val[NEON_B], dst.val[NEON_B],
michael@0:                                    src.val[NEON_A], dst.val[NEON_A]);
michael@0: 
michael@0:     return ret;
michael@0: }
michael@0: 
michael@0: static inline uint8x8_t lighten_color(uint8x8_t sc, uint8x8_t dc,
michael@0:                                       uint8x8_t sa, uint8x8_t da) {
michael@0:     return lighten_darken_color<true>(sc, dc, sa, da);
michael@0: }
michael@0: 
michael@0: uint8x8x4_t lighten_modeproc_neon8(uint8x8x4_t src, uint8x8x4_t dst) {
michael@0:     uint8x8x4_t ret;
michael@0: 
michael@0:     ret.val[NEON_A] = srcover_color(src.val[NEON_A], dst.val[NEON_A]);
michael@0:     ret.val[NEON_R] = lighten_color(src.val[NEON_R], dst.val[NEON_R],
michael@0:                                     src.val[NEON_A], dst.val[NEON_A]);
michael@0:     ret.val[NEON_G] = lighten_color(src.val[NEON_G], dst.val[NEON_G],
michael@0:                                     src.val[NEON_A], dst.val[NEON_A]);
michael@0:     ret.val[NEON_B] = lighten_color(src.val[NEON_B], dst.val[NEON_B],
michael@0:                                     src.val[NEON_A], dst.val[NEON_A]);
michael@0: 
michael@0:     return ret;
michael@0: }
michael@0: 
michael@0: static inline uint8x8_t hardlight_color(uint8x8_t sc, uint8x8_t dc,
michael@0:                                         uint8x8_t sa, uint8x8_t da) {
michael@0:     return overlay_hardlight_color<false>(sc, dc, sa, da);
michael@0: }
michael@0: 
michael@0: uint8x8x4_t hardlight_modeproc_neon8(uint8x8x4_t src, uint8x8x4_t dst) {
michael@0:     uint8x8x4_t ret;
michael@0: 
michael@0:     ret.val[NEON_A] = srcover_color(src.val[NEON_A], dst.val[NEON_A]);
michael@0:     ret.val[NEON_R] = hardlight_color(src.val[NEON_R], dst.val[NEON_R],
michael@0:                                       src.val[NEON_A], dst.val[NEON_A]);
michael@0:     ret.val[NEON_G] = hardlight_color(src.val[NEON_G], dst.val[NEON_G],
michael@0:                                       src.val[NEON_A], dst.val[NEON_A]);
michael@0:     ret.val[NEON_B] = hardlight_color(src.val[NEON_B], dst.val[NEON_B],
michael@0:                                       src.val[NEON_A], dst.val[NEON_A]);
michael@0: 
michael@0:     return ret;
michael@0: }
michael@0: 
michael@0: static inline uint8x8_t difference_color(uint8x8_t sc, uint8x8_t dc,
michael@0:                                          uint8x8_t sa, uint8x8_t da) {
michael@0:     uint16x8_t sd, ds, tmp;
michael@0:     int16x8_t val;
michael@0: 
michael@0:     sd = vmull_u8(sc, da);
michael@0:     ds = vmull_u8(dc, sa);
michael@0: 
michael@0:     tmp = vminq_u16(sd, ds);
michael@0:     tmp = SkDiv255Round_neon8_16_16(tmp);
michael@0:     tmp = vshlq_n_u16(tmp, 1);
michael@0: 
michael@0:     val = vreinterpretq_s16_u16(vaddl_u8(sc, dc));
michael@0: 
michael@0:     val -= vreinterpretq_s16_u16(tmp);
michael@0: 
michael@0:     val = vmaxq_s16(val, vdupq_n_s16(0));
michael@0:     val = vminq_s16(val, vdupq_n_s16(255));
michael@0: 
michael@0:     return vmovn_u16(vreinterpretq_u16_s16(val));
michael@0: }
michael@0: 
michael@0: uint8x8x4_t difference_modeproc_neon8(uint8x8x4_t src, uint8x8x4_t dst) {
michael@0:     uint8x8x4_t ret;
michael@0: 
michael@0:     ret.val[NEON_A] = srcover_color(src.val[NEON_A], dst.val[NEON_A]);
michael@0:     ret.val[NEON_R] = difference_color(src.val[NEON_R], dst.val[NEON_R],
michael@0:                                        src.val[NEON_A], dst.val[NEON_A]);
michael@0:     ret.val[NEON_G] = difference_color(src.val[NEON_G], dst.val[NEON_G],
michael@0:                                        src.val[NEON_A], dst.val[NEON_A]);
michael@0:     ret.val[NEON_B] = difference_color(src.val[NEON_B], dst.val[NEON_B],
michael@0:                                        src.val[NEON_A], dst.val[NEON_A]);
michael@0: 
michael@0:     return ret;
michael@0: }
michael@0: 
michael@0: static inline uint8x8_t exclusion_color(uint8x8_t sc, uint8x8_t dc,
michael@0:                                         uint8x8_t sa, uint8x8_t da) {
michael@0:     /* The equation can be simplified to 255(sc + dc) - 2 * sc * dc */
michael@0: 
michael@0:     uint16x8_t sc_plus_dc, scdc, const255;
michael@0:     int32x4_t term1_1, term1_2, term2_1, term2_2;
michael@0: 
michael@0:     /* Calc (sc + dc) and (sc * dc) */
michael@0:     sc_plus_dc = vaddl_u8(sc, dc);
michael@0:     scdc = vmull_u8(sc, dc);
michael@0: 
michael@0:     /* Prepare constants */
michael@0:     const255 = vdupq_n_u16(255);
michael@0: 
michael@0:     /* Calc the first term */
michael@0:     term1_1 = vreinterpretq_s32_u32(
michael@0:                 vmull_u16(vget_low_u16(const255), vget_low_u16(sc_plus_dc)));
michael@0:     term1_2 = vreinterpretq_s32_u32(
michael@0:                 vmull_u16(vget_high_u16(const255), vget_high_u16(sc_plus_dc)));
michael@0: 
michael@0:     /* Calc the second term */
michael@0:     term2_1 = vreinterpretq_s32_u32(vshll_n_u16(vget_low_u16(scdc), 1));
michael@0:     term2_2 = vreinterpretq_s32_u32(vshll_n_u16(vget_high_u16(scdc), 1));
michael@0: 
michael@0:     return clamp_div255round_simd8_32(term1_1 - term2_1, term1_2 - term2_2);
michael@0: }
michael@0: 
michael@0: uint8x8x4_t exclusion_modeproc_neon8(uint8x8x4_t src, uint8x8x4_t dst) {
michael@0:     uint8x8x4_t ret;
michael@0: 
michael@0:     ret.val[NEON_A] = srcover_color(src.val[NEON_A], dst.val[NEON_A]);
michael@0:     ret.val[NEON_R] = exclusion_color(src.val[NEON_R], dst.val[NEON_R],
michael@0:                                       src.val[NEON_A], dst.val[NEON_A]);
michael@0:     ret.val[NEON_G] = exclusion_color(src.val[NEON_G], dst.val[NEON_G],
michael@0:                                       src.val[NEON_A], dst.val[NEON_A]);
michael@0:     ret.val[NEON_B] = exclusion_color(src.val[NEON_B], dst.val[NEON_B],
michael@0:                                       src.val[NEON_A], dst.val[NEON_A]);
michael@0: 
michael@0:     return ret;
michael@0: }
michael@0: 
michael@0: static inline uint8x8_t blendfunc_multiply_color(uint8x8_t sc, uint8x8_t dc,
michael@0:                                                  uint8x8_t sa, uint8x8_t da) {
michael@0:     uint32x4_t val1, val2;
michael@0:     uint16x8_t scdc, t1, t2;
michael@0: 
michael@0:     t1 = vmull_u8(sc, vdup_n_u8(255) - da);
michael@0:     t2 = vmull_u8(dc, vdup_n_u8(255) - sa);
michael@0:     scdc = vmull_u8(sc, dc);
michael@0: 
michael@0:     val1 = vaddl_u16(vget_low_u16(t1), vget_low_u16(t2));
michael@0:     val2 = vaddl_u16(vget_high_u16(t1), vget_high_u16(t2));
michael@0: 
michael@0:     val1 = vaddw_u16(val1, vget_low_u16(scdc));
michael@0:     val2 = vaddw_u16(val2, vget_high_u16(scdc));
michael@0: 
michael@0:     return clamp_div255round_simd8_32(
michael@0:                 vreinterpretq_s32_u32(val1), vreinterpretq_s32_u32(val2));
michael@0: }
michael@0: 
michael@0: uint8x8x4_t multiply_modeproc_neon8(uint8x8x4_t src, uint8x8x4_t dst) {
michael@0:     uint8x8x4_t ret;
michael@0: 
michael@0:     ret.val[NEON_A] = srcover_color(src.val[NEON_A], dst.val[NEON_A]);
michael@0:     ret.val[NEON_R] = blendfunc_multiply_color(src.val[NEON_R], dst.val[NEON_R],
michael@0:                                                src.val[NEON_A], dst.val[NEON_A]);
michael@0:     ret.val[NEON_G] = blendfunc_multiply_color(src.val[NEON_G], dst.val[NEON_G],
michael@0:                                                src.val[NEON_A], dst.val[NEON_A]);
michael@0:     ret.val[NEON_B] = blendfunc_multiply_color(src.val[NEON_B], dst.val[NEON_B],
michael@0:                                                src.val[NEON_A], dst.val[NEON_A]);
michael@0: 
michael@0:     return ret;
michael@0: }
michael@0: 
michael@0: ////////////////////////////////////////////////////////////////////////////////
michael@0: 
michael@0: typedef uint8x8x4_t (*SkXfermodeProcSIMD)(uint8x8x4_t src, uint8x8x4_t dst);
michael@0: 
michael@0: extern SkXfermodeProcSIMD gNEONXfermodeProcs[];
michael@0: 
michael@0: SkNEONProcCoeffXfermode::SkNEONProcCoeffXfermode(SkReadBuffer& buffer)
michael@0:         : INHERITED(buffer) {
michael@0:     fProcSIMD = reinterpret_cast<void*>(gNEONXfermodeProcs[this->getMode()]);
michael@0: }
michael@0: 
michael@0: void SkNEONProcCoeffXfermode::xfer32(SkPMColor dst[], const SkPMColor src[],
michael@0:                                      int count, const SkAlpha aa[]) const {
michael@0:     SkASSERT(dst && src && count >= 0);
michael@0: 
michael@0:     SkXfermodeProc proc = this->getProc();
michael@0:     SkXfermodeProcSIMD procSIMD = reinterpret_cast<SkXfermodeProcSIMD>(fProcSIMD);
michael@0:     SkASSERT(procSIMD != NULL);
michael@0: 
michael@0:     if (NULL == aa) {
michael@0:         // Unrolled NEON code
michael@0:         while (count >= 8) {
michael@0:             uint8x8x4_t vsrc, vdst, vres;
michael@0: 
michael@0: #if (__GNUC__ > 4) || ((__GNUC__ == 4) && (__GNUC_MINOR__ > 6))
michael@0:             asm volatile (
michael@0:                 "vld4.u8    %h[vsrc], [%[src]]!  \t\n"
michael@0:                 "vld4.u8    %h[vdst], [%[dst]]   \t\n"
michael@0:                 : [vsrc] "=w" (vsrc), [vdst] "=w" (vdst), [src] "+&r" (src)
michael@0:                 : [dst] "r" (dst)
michael@0:                 :
michael@0:             );
michael@0: #else
michael@0:             register uint8x8_t d0 asm("d0");
michael@0:             register uint8x8_t d1 asm("d1");
michael@0:             register uint8x8_t d2 asm("d2");
michael@0:             register uint8x8_t d3 asm("d3");
michael@0:             register uint8x8_t d4 asm("d4");
michael@0:             register uint8x8_t d5 asm("d5");
michael@0:             register uint8x8_t d6 asm("d6");
michael@0:             register uint8x8_t d7 asm("d7");
michael@0: 
michael@0:             asm volatile (
michael@0:                 "vld4.u8    {d0-d3},[%[src]]!;"
michael@0:                 "vld4.u8    {d4-d7},[%[dst]];"
michael@0:                 : "=w" (d0), "=w" (d1), "=w" (d2), "=w" (d3),
michael@0:                   "=w" (d4), "=w" (d5), "=w" (d6), "=w" (d7),
michael@0:                   [src] "+&r" (src)
michael@0:                 : [dst] "r" (dst)
michael@0:                 :
michael@0:             );
michael@0:             vsrc.val[0] = d0; vdst.val[0] = d4;
michael@0:             vsrc.val[1] = d1; vdst.val[1] = d5;
michael@0:             vsrc.val[2] = d2; vdst.val[2] = d6;
michael@0:             vsrc.val[3] = d3; vdst.val[3] = d7;
michael@0: #endif
michael@0: 
michael@0:             vres = procSIMD(vsrc, vdst);
michael@0: 
michael@0:             vst4_u8((uint8_t*)dst, vres);
michael@0: 
michael@0:             count -= 8;
michael@0:             dst += 8;
michael@0:         }
michael@0:         // Leftovers
michael@0:         for (int i = 0; i < count; i++) {
michael@0:             dst[i] = proc(src[i], dst[i]);
michael@0:         }
michael@0:     } else {
michael@0:         for (int i = count - 1; i >= 0; --i) {
michael@0:             unsigned a = aa[i];
michael@0:             if (0 != a) {
michael@0:                 SkPMColor dstC = dst[i];
michael@0:                 SkPMColor C = proc(src[i], dstC);
michael@0:                 if (a != 0xFF) {
michael@0:                     C = SkFourByteInterp_neon(C, dstC, a);
michael@0:                 }
michael@0:                 dst[i] = C;
michael@0:             }
michael@0:         }
michael@0:     }
michael@0: }
michael@0: 
michael@0: void SkNEONProcCoeffXfermode::xfer16(uint16_t* SK_RESTRICT dst,
michael@0:                                      const SkPMColor* SK_RESTRICT src, int count,
michael@0:                                      const SkAlpha* SK_RESTRICT aa) const {
michael@0:     SkASSERT(dst && src && count >= 0);
michael@0: 
michael@0:     SkXfermodeProc proc = this->getProc();
michael@0:     SkXfermodeProcSIMD procSIMD = reinterpret_cast<SkXfermodeProcSIMD>(fProcSIMD);
michael@0:     SkASSERT(procSIMD != NULL);
michael@0: 
michael@0:     if (NULL == aa) {
michael@0:         while(count >= 8) {
michael@0:             uint16x8_t vdst, vres16;
michael@0:             uint8x8x4_t vdst32, vsrc, vres;
michael@0: 
michael@0:             vdst = vld1q_u16(dst);
michael@0: 
michael@0: #if (__GNUC__ > 4) || ((__GNUC__ == 4) && (__GNUC_MINOR__ > 6))
michael@0:             asm volatile (
michael@0:                 "vld4.u8    %h[vsrc], [%[src]]!  \t\n"
michael@0:                 : [vsrc] "=w" (vsrc), [src] "+&r" (src)
michael@0:                 : :
michael@0:             );
michael@0: #else
michael@0:             register uint8x8_t d0 asm("d0");
michael@0:             register uint8x8_t d1 asm("d1");
michael@0:             register uint8x8_t d2 asm("d2");
michael@0:             register uint8x8_t d3 asm("d3");
michael@0: 
michael@0:             asm volatile (
michael@0:                 "vld4.u8    {d0-d3},[%[src]]!;"
michael@0:                 : "=w" (d0), "=w" (d1), "=w" (d2), "=w" (d3),
michael@0:                   [src] "+&r" (src)
michael@0:                 : :
michael@0:             );
michael@0:             vsrc.val[0] = d0;
michael@0:             vsrc.val[1] = d1;
michael@0:             vsrc.val[2] = d2;
michael@0:             vsrc.val[3] = d3;
michael@0: #endif
michael@0: 
michael@0:             vdst32 = SkPixel16ToPixel32_neon8(vdst);
michael@0:             vres = procSIMD(vsrc, vdst32);
michael@0:             vres16 = SkPixel32ToPixel16_neon8(vres);
michael@0: 
michael@0:             vst1q_u16(dst, vres16);
michael@0: 
michael@0:             count -= 8;
michael@0:             dst += 8;
michael@0:         }
michael@0:         for (int i = 0; i < count; i++) {
michael@0:             SkPMColor dstC = SkPixel16ToPixel32(dst[i]);
michael@0:             dst[i] = SkPixel32ToPixel16_ToU16(proc(src[i], dstC));
michael@0:         }
michael@0:     } else {
michael@0:         for (int i = count - 1; i >= 0; --i) {
michael@0:             unsigned a = aa[i];
michael@0:             if (0 != a) {
michael@0:                 SkPMColor dstC = SkPixel16ToPixel32(dst[i]);
michael@0:                 SkPMColor C = proc(src[i], dstC);
michael@0:                 if (0xFF != a) {
michael@0:                     C = SkFourByteInterp_neon(C, dstC, a);
michael@0:                 }
michael@0:                 dst[i] = SkPixel32ToPixel16_ToU16(C);
michael@0:             }
michael@0:         }
michael@0:     }
michael@0: }
michael@0: 
michael@0: #ifndef SK_IGNORE_TO_STRING
michael@0: void SkNEONProcCoeffXfermode::toString(SkString* str) const {
michael@0:     this->INHERITED::toString(str);
michael@0: }
michael@0: #endif
michael@0: 
michael@0: ////////////////////////////////////////////////////////////////////////////////
michael@0: 
michael@0: SkXfermodeProcSIMD gNEONXfermodeProcs[] = {
michael@0:     NULL, // kClear_Mode
michael@0:     NULL, // kSrc_Mode
michael@0:     NULL, // kDst_Mode
michael@0:     NULL, // kSrcOver_Mode
michael@0:     dstover_modeproc_neon8,
michael@0:     srcin_modeproc_neon8,
michael@0:     dstin_modeproc_neon8,
michael@0:     srcout_modeproc_neon8,
michael@0:     dstout_modeproc_neon8,
michael@0:     srcatop_modeproc_neon8,
michael@0:     dstatop_modeproc_neon8,
michael@0:     xor_modeproc_neon8,
michael@0:     plus_modeproc_neon8,
michael@0:     modulate_modeproc_neon8,
michael@0:     screen_modeproc_neon8,
michael@0: 
michael@0:     overlay_modeproc_neon8,
michael@0:     darken_modeproc_neon8,
michael@0:     lighten_modeproc_neon8,
michael@0:     NULL, // kColorDodge_Mode
michael@0:     NULL, // kColorBurn_Mode
michael@0:     hardlight_modeproc_neon8,
michael@0:     NULL, // kSoftLight_Mode
michael@0:     difference_modeproc_neon8,
michael@0:     exclusion_modeproc_neon8,
michael@0:     multiply_modeproc_neon8,
michael@0: 
michael@0:     NULL, // kHue_Mode
michael@0:     NULL, // kSaturation_Mode
michael@0:     NULL, // kColor_Mode
michael@0:     NULL, // kLuminosity_Mode
michael@0: };
michael@0: 
michael@0: SK_COMPILE_ASSERT(
michael@0:     SK_ARRAY_COUNT(gNEONXfermodeProcs) == SkXfermode::kLastMode + 1,
michael@0:     mode_count_arm
michael@0: );
michael@0: 
michael@0: SkXfermodeProc gNEONXfermodeProcs1[] = {
michael@0:     NULL, // kClear_Mode
michael@0:     NULL, // kSrc_Mode
michael@0:     NULL, // kDst_Mode
michael@0:     NULL, // kSrcOver_Mode
michael@0:     NULL, // kDstOver_Mode
michael@0:     NULL, // kSrcIn_Mode
michael@0:     NULL, // kDstIn_Mode
michael@0:     NULL, // kSrcOut_Mode
michael@0:     NULL, // kDstOut_Mode
michael@0:     srcatop_modeproc_neon,
michael@0:     dstatop_modeproc_neon,
michael@0:     xor_modeproc_neon,
michael@0:     plus_modeproc_neon,
michael@0:     modulate_modeproc_neon,
michael@0:     NULL, // kScreen_Mode
michael@0: 
michael@0:     NULL, // kOverlay_Mode
michael@0:     NULL, // kDarken_Mode
michael@0:     NULL, // kLighten_Mode
michael@0:     NULL, // kColorDodge_Mode
michael@0:     NULL, // kColorBurn_Mode
michael@0:     NULL, // kHardLight_Mode
michael@0:     NULL, // kSoftLight_Mode
michael@0:     NULL, // kDifference_Mode
michael@0:     NULL, // kExclusion_Mode
michael@0:     NULL, // kMultiply_Mode
michael@0: 
michael@0:     NULL, // kHue_Mode
michael@0:     NULL, // kSaturation_Mode
michael@0:     NULL, // kColor_Mode
michael@0:     NULL, // kLuminosity_Mode
michael@0: };
michael@0: 
michael@0: SK_COMPILE_ASSERT(
michael@0:     SK_ARRAY_COUNT(gNEONXfermodeProcs1) == SkXfermode::kLastMode + 1,
michael@0:     mode1_count_arm
michael@0: );
michael@0: 
michael@0: SkProcCoeffXfermode* SkPlatformXfermodeFactory_impl_neon(const ProcCoeff& rec,
michael@0:                                                          SkXfermode::Mode mode) {
michael@0: 
michael@0:     void* procSIMD = reinterpret_cast<void*>(gNEONXfermodeProcs[mode]);
michael@0: 
michael@0:     if (procSIMD != NULL) {
michael@0:         return SkNEW_ARGS(SkNEONProcCoeffXfermode, (rec, mode, procSIMD));
michael@0:     }
michael@0:     return NULL;
michael@0: }
michael@0: 
michael@0: SkXfermodeProc SkPlatformXfermodeProcFactory_impl_neon(SkXfermode::Mode mode) {
michael@0:     return gNEONXfermodeProcs1[mode];
michael@0: }