1/* libs/pixelflinger/col32cb16blend_neon.S
2 *
3 * Copyright (C) 2009 The Android Open Source Project
4 *
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at
8 *
9 *      http://www.apache.org/licenses/LICENSE-2.0
10 *
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
16 */
17
18
19    .text
20    .balign 4
21
22    .global scanline_col32cb16blend_neon
23
24//
25// This function alpha blends a fixed color into a destination scanline, using
26// the formula:
27//
28//     d = s + (((a + (a >> 7)) * d) >> 8)
29//
30// where d is the destination pixel,
31//       s is the source color,
32//       a is the alpha channel of the source color.
33//
34// The NEON implementation processes 16 pixels per iteration. The remaining 0 - 15
35// pixels are processed in ARM code.
36//
37
38// r0 = destination buffer pointer
39// r1 = color pointer
40// r2 = count
41
42
43scanline_col32cb16blend_neon:
44    push        {r4-r11, lr}                    // stack ARM regs
45
46    vmov.u16    q15, #256                       // create alpha constant
47    movs        r3, r2, lsr #4                  // calc. sixteens iterations
48    vmov.u16    q14, #0x1f                      // create blue mask
49
50    beq         2f                              // if r3 == 0, branch to singles
51
52    vld4.8      {d0[], d2[], d4[], d6[]}, [r1]  // load color into four registers
53                                                //  split and duplicate them, such that
54                                                //  d0 = 8 equal red values
55                                                //  d2 = 8 equal green values
56                                                //  d4 = 8 equal blue values
57                                                //  d6 = 8 equal alpha values
58    vshll.u8    q0, d0, #5                      // shift up red and widen
59    vshll.u8    q1, d2, #6                      // shift up green and widen
60    vshll.u8    q2, d4, #5                      // shift up blue and widen
61
62    vshr.u8     d7, d6, #7                      // extract top bit of alpha
63    vaddl.u8    q3, d6, d7                      // add top bit into alpha
64    vsub.u16    q3, q15, q3                     // invert alpha
65
661:
67    // This loop processes 16 pixels per iteration. In the comments, references to
68    // the first eight pixels are suffixed with "0" (red0, green0, blue0),
69    // the second eight are suffixed "1".
70                                                // q8  = dst red0
71                                                // q9  = dst green0
72                                                // q10 = dst blue0
73                                                // q13 = dst red1
74                                                // q12 = dst green1
75                                                // q11 = dst blue1
76
77    vld1.16     {d20, d21, d22, d23}, [r0]      // load 16 dest pixels
78    vshr.u16    q8, q10, #11                    // shift dst red0 to low 5 bits
79    pld         [r0, #63]                       // preload next dest pixels
80    vshl.u16    q9, q10, #5                     // shift dst green0 to top 6 bits
81    vand        q10, q10, q14                   // extract dst blue0
82    vshr.u16    q9, q9, #10                     // shift dst green0 to low 6 bits
83    vmul.u16    q8, q8, q3                      // multiply dst red0 by src alpha
84    vshl.u16    q12, q11, #5                    // shift dst green1 to top 6 bits
85    vmul.u16    q9, q9, q3                      // multiply dst green0 by src alpha
86    vshr.u16    q13, q11, #11                   // shift dst red1 to low 5 bits
87    vmul.u16    q10, q10, q3                    // multiply dst blue0 by src alpha
88    vshr.u16    q12, q12, #10                   // shift dst green1 to low 6 bits
89    vand        q11, q11, q14                   // extract dst blue1
90    vadd.u16    q8, q8, q0                      // add src red to dst red0
91    vmul.u16    q13, q13, q3                    // multiply dst red1 by src alpha
92    vadd.u16    q9, q9, q1                      // add src green to dst green0
93    vmul.u16    q12, q12, q3                    // multiply dst green1 by src alpha
94    vadd.u16    q10, q10, q2                    // add src blue to dst blue0
95    vmul.u16    q11, q11, q3                    // multiply dst blue1 by src alpha
96    vshr.u16    q8, q8, #8                      // shift down red0
97    vadd.u16    q13, q13, q0                    // add src red to dst red1
98    vshr.u16    q9, q9, #8                      // shift down green0
99    vadd.u16    q12, q12, q1                    // add src green to dst green1
100    vshr.u16    q10, q10, #8                    // shift down blue0
101    vadd.u16    q11, q11, q2                    // add src blue to dst blue1
102    vsli.u16    q10, q9, #5                     // shift & insert green0 into blue0
103    vshr.u16    q13, q13, #8                    // shift down red1
104    vsli.u16    q10, q8, #11                    // shift & insert red0 into blue0
105    vshr.u16    q12, q12, #8                    // shift down green1
106    vshr.u16    q11, q11, #8                    // shift down blue1
107    subs        r3, r3, #1                      // decrement loop counter
108    vsli.u16    q11, q12, #5                    // shift & insert green1 into blue1
109    vsli.u16    q11, q13, #11                   // shift & insert red1 into blue1
110
111    vst1.16     {d20, d21, d22, d23}, [r0]!     // write 16 pixels back to dst
112    bne         1b                              // if count != 0, loop
113
1142:
115    ands        r3, r2, #15                     // calc. single iterations
116    beq         4f                              // if r3 == 0, exit
117
118    ldr         r4, [r1]                        // load source color
119    mov         r5, r4, lsr #24                 // shift down alpha
120    add         r5, r5, r5, lsr #7              // add in top bit
121    rsb         r5, r5, #256                    // invert alpha
122    and         r11, r4, #0xff                  // extract red
123    ubfx        r12, r4, #8, #8                 // extract green
124    ubfx        r4, r4, #16, #8                 // extract blue
125    mov         r11, r11, lsl #5                // prescale red
126    mov         r12, r12, lsl #6                // prescale green
127    mov         r4, r4, lsl #5                  // prescale blue
128
1293:
130    ldrh        r8, [r0]                        // load dest pixel
131    subs        r3, r3, #1                      // decrement loop counter
132    mov         r6, r8, lsr #11                 // extract dest red
133    ubfx        r7, r8, #5, #6                  // extract dest green
134    and         r8, r8, #0x1f                   // extract dest blue
135
136    smlabb      r6, r6, r5, r11                 // dest red * alpha + src red
137    smlabb      r7, r7, r5, r12                 // dest green * alpha + src green
138    smlabb      r8, r8, r5, r4                  // dest blue * alpha + src blue
139
140    mov         r6, r6, lsr #8                  // shift down red
141    mov         r7, r7, lsr #8                  // shift down green
142    mov         r6, r6, lsl #11                 // shift red into 565
143    orr         r6, r7, lsl #5                  // shift green into 565
144    orr         r6, r8, lsr #8                  // shift blue into 565
145
146    strh        r6, [r0], #2                    // store pixel to dest, update ptr
147    bne         3b                              // if count != 0, loop
1484:
149
150    pop         {r4-r11, pc}                    // return
151
152
153
154