ARM平台YUV轉RGB888 - Nova的專欄 - CSDN博客

前段時間，公司有個嵌入式的項目，我做的其中有一項是yuv視頻到rgb888圖像的轉換。我主要是參照一位牛人的博客做的：http://blog.csdn.net/housisong/archive/2007/10/31/1859084.aspx

YUV422p To RGB888 C語言版本介面說明：

view plaincopy to clipboardprint?·········10········20········30········40········50········60········70········80········90········100·······110·······120·······130·······140·······150

1. /****************************************YUV422P_To_RGB24.c**************************/
2. //模塊功能：將YUV422_PLANAR圖像數據轉換成RGB24格式
3. typedefunsignedcharBYTE;//[0..255]
4. /*
5. *介面說明：
6. *功能：構建查詢表，轉換模塊運行前必須進行的初始化操作
7. */
8. voidYUV422P_To_RGB24_init();
9. /*
10. *介面說明：
11. *功能：將YUV422P圖像數據轉換成RGB24格式
12. *參數：
13. *pY:YUV422P圖像數據Y的起始指針
14. *pU:YUV422P圖像數據U的起始指針
15. *pV:YUV422P圖像數據V的起始指針
16. *DstPic:轉換成的RGB24圖像數據的起始指針
17. *width:圖像寬度
18. *height:圖像高度
19. *返回值：成功返回0，失敗返回-1
20. *注意：DstPic所指向的緩衝區必須事先分配好，其大小應該為width*height*3
21. */
22. intYUV422P_To_RGB24(BYTE*pY,BYTE*pU,BYTE*pV,BYTE*DstPic,intwidth,intheight);

/****************************************YUV422P_To_RGB24.c**************************///模塊功能：將YUV422_PLANAR圖像數據轉換成RGB24格式typedef unsigned char BYTE; // [0..255]/** 介面說明：* 功能：構建查詢表，轉換模塊運行前必須進行的初始化操作*/void YUV422P_To_RGB24_init();/**介面說明：*功能：將YUV422P圖像數據轉換成RGB24格式*參數：* pY: YUV422P圖像數據Y的起始指針* pU: YUV422P圖像數據U的起始指針* pV: YUV422P圖像數據V的起始指針* DstPic: 轉換成的RGB24圖像數據的起始指針* width: 圖像寬度* height: 圖像高度*返回值：成功返回0，失敗返回-1*注意：DstPic所指向的緩衝區必須事先分配好，其大小應該為 width*height*3*/int YUV422P_To_RGB24(BYTE* pY, BYTE* pU, BYTE* pV, BYTE* DstPic, int width, int height);

實現：

view plaincopy to clipboardprint?·········10········20········30········40········50········60········70········80········90········100·······110·······120·······130·······140·······150

1. //使用整數運算(定點數運算)來代替浮點運算
2. constintcsY_coeff_16=1.164383*(1<<16);
3. constintcsU_blue_16=2.017232*(1<<16);
4. constintcsU_green_16=(-0.391762)*(1<<16);
5. constintcsV_green_16=(-0.812968)*(1<<16);
6. constintcsV_red_16=1.596027*(1<<16);
7. //顏色查表
8. staticBYTE_color_table[256*3];
9. staticconstBYTE*color_table=&_color_table[256];
10. //查表
11. staticintYm_tableEx[256];
12. staticintUm_blue_tableEx[256];
13. staticintUm_green_tableEx[256];
14. staticintVm_green_tableEx[256];
15. staticintVm_red_tableEx[256];
16. //顏色飽和函數
17. inlinelongborder_color(longcolor){
18. if(color>255)
19. return255;
20. elseif(color<0)
21. return0;
22. else
23. returncolor;
24. }
25. //採用查找表進行計算時，必須運行的初始化函數
26. voidYUV422P_To_RGB24_init(){
27. inti;
28. for(i=0;i<256*3;++i)
29. _color_table[i]=border_color(i-256);
30. for(i=0;i<256;++i){
31. Ym_tableEx[i]=(csY_coeff_16*(i-16))>>16;
32. Um_blue_tableEx[i]=(csU_blue_16*(i-128))>>16;
33. Um_green_tableEx[i]=(csU_green_16*(i-128))>>16;
34. Vm_green_tableEx[i]=(csV_green_16*(i-128))>>16;
35. Vm_red_tableEx[i]=(csV_red_16*(i-128))>>16;
36. }
37. }
38. inlinevoidYUVToRGB24_Table(BYTE*p,constBYTEY0,constBYTEY1,
39. constBYTEU,constBYTEV){
40. intYe0=Ym_tableEx[Y0];
41. intYe1=Ym_tableEx[Y1];
42. intUe_blue=Um_blue_tableEx[U];
43. intUe_green=Um_green_tableEx[U];
44. intVe_green=Vm_green_tableEx[V];
45. intVe_red=Vm_red_tableEx[V];
46. intUeVe_green=Ue_green+Ve_green;
47. *p=color_table[(Ye0+Ve_red)];
48. *(p+1)=color_table[(Ye0+UeVe_green)];
49. *(p+2)=color_table[(Ye0+Ue_blue)];
50. *(p+3)=color_table[(Ye1+Ve_red)];
51. *(p+4)=color_table[(Ye1+UeVe_green)];
52. *(p+5)=color_table[(Ye1+Ue_blue)];
53. }
54. intYUV420P_To_RGB24(BYTE*pY,BYTE*pU,BYTE*pV,BYTE*DstPic,intwidth,
55. intheight){
56. inty,x,x_uv;
57. BYTE*pDstLine=DstPic;
58. if((width%2)!=0||(height%2)!=0)
59. return(-1);
60. for(y=0;y<height;++y){
61. //DECODE_PlanarYUV211_Common_line(pDstLine,pY,pU,pV,width);
62. for(x=0;x<width;x+=2){
63. x_uv=x>>1;
64. YUVToRGB24_Table(&pDstLine[x*3],pY[x],pY[x+1],pU[x_uv],
65. pV[x_uv]);
66. }
67. pDstLine+=width*3;//RGB888
68. pY+=width;//YUV422
69. if(y%2==1){
70. pU+=width/2;
71. pV+=width/2;
72. }
73. }
74. return0;
75. }

//使用整數運算(定點數運算)來代替浮點運算const int csY_coeff_16 = 1.164383 * (1 << 16);const int csU_blue_16 = 2.017232 * (1 << 16);const int csU_green_16 = (-0.391762) * (1 << 16);const int csV_green_16 = (-0.812968) * (1 << 16);const int csV_red_16 = 1.596027 * (1 << 16);//顏色查表static BYTE _color_table[256 * 3];static const BYTE* color_table = &_color_table[256];//查表static int Ym_tableEx[256];static int Um_blue_tableEx[256];static int Um_green_tableEx[256];static int Vm_green_tableEx[256];static int Vm_red_tableEx[256];//顏色飽和函數inline long border_color(long color) {if (color > 255)return 255;else if (color < 0)return 0;elsereturn color;}//採用查找表進行計算時，必須運行的初始化函數void YUV422P_To_RGB24_init() {int i;for (i = 0; i < 256 * 3; ++i)_color_table[i] = border_color(i - 256);for (i = 0; i < 256; ++i) {Ym_tableEx[i] = (csY_coeff_16 * (i - 16)) >> 16;Um_blue_tableEx[i] = (csU_blue_16 * (i - 128)) >> 16;Um_green_tableEx[i] = (csU_green_16 * (i - 128)) >> 16;Vm_green_tableEx[i] = (csV_green_16 * (i - 128)) >> 16;Vm_red_tableEx[i] = (csV_red_16 * (i - 128)) >> 16;}}inline void YUVToRGB24_Table(BYTE *p, const BYTE Y0, const BYTE Y1,const BYTE U, const BYTE V) {int Ye0 = Ym_tableEx[Y0];int Ye1 = Ym_tableEx[Y1];int Ue_blue = Um_blue_tableEx[U];int Ue_green = Um_green_tableEx[U];int Ve_green = Vm_green_tableEx[V];int Ve_red = Vm_red_tableEx[V];int UeVe_green = Ue_green + Ve_green;*p = color_table[(Ye0 + Ve_red)];*(p + 1) = color_table[(Ye0 + UeVe_green)];*(p + 2) = color_table[(Ye0 + Ue_blue)];*(p + 3) = color_table[(Ye1 + Ve_red)];*(p + 4) = color_table[(Ye1 + UeVe_green)];*(p + 5) = color_table[(Ye1 + Ue_blue)];}int YUV420P_To_RGB24(BYTE* pY, BYTE* pU, BYTE* pV, BYTE* DstPic, int width,int height) {int y, x, x_uv;BYTE* pDstLine = DstPic;if ((width % 2) != 0 || (height % 2) != 0)return (-1);for (y = 0; y < height; ++y) {//DECODE_PlanarYUV211_Common_line(pDstLine, pY, pU, pV,width);for (x = 0; x < width; x += 2) {x_uv = x >> 1;YUVToRGB24_Table(&pDstLine[x * 3], pY[x], pY[x + 1], pU[x_uv],pV[x_uv]);}pDstLine += width * 3; //RGB888pY += width; //YUV422if (y % 2 == 1) {pU += width / 2;pV += width / 2;}}return 0;}

經測試發現，在hi3512（arm 926ej-s,267MHz）平台上運行時，該yuv轉rgb模塊的速度不是很快，大概20幀/秒。為了提高效率，核心解碼模塊我採用了arm彙編，重寫了YUVToRGB24_Table模塊。

YUV420P_To_RGB24_asm.c代碼：

view plaincopy to clipboardprint?·········10········20········30········40········50········60········70········80········90········100·······110·······120·······130·······140·······150

1. externintYUVToRGB24_Assemble(unsignedchar*pDstLine,unsignedchar**yuv,intwidth);
2. //使用整數運算(定點數運算)來代替浮點運算
3. constintcsY_coeff_16=1.164383*(1<<16);
4. constintcsU_blue_16=2.017232*(1<<16);
5. constintcsU_green_16=(-0.391762)*(1<<16);
6. constintcsV_green_16=(-0.812968)*(1<<16);
7. constintcsV_red_16=1.596027*(1<<16);
8. //查表
9. intYm_tableEx[256];
10. intUm_blue_tableEx[256];
11. intUm_green_tableEx[256];
12. intVm_green_tableEx[256];
13. intVm_red_tableEx[256];
14. //採用查找表進行計算時，必須運行的初始化函數
15. voidYUV422P_To_RGB24_init()
16. {
17. inti;
18. for(i=0;i<256;++i)
19. {
20. Ym_tableEx[i]=(csY_coeff_16*(i-16))>>16;
21. Um_blue_tableEx[i]=(csU_blue_16*(i-128))>>16;
22. Um_green_tableEx[i]=(csU_green_16*(i-128))>>16;
23. Vm_green_tableEx[i]=(csV_green_16*(i-128))>>16;
24. Vm_red_tableEx[i]=(csV_red_16*(i-128))>>16;
25. }
26. }
27. intYUV420P_To_RGB24(BYTE*pY,BYTE*pU,BYTE*pV,BYTE*DstPic,intwidth,intheight)
28. {
29. inty;
30. BYTE*pDstLine=DstPic;
31. BYTE*yuv[3];
33. if((width%8)!=0)
34. return(-1);
35. yuv[0]=pY;
36. yuv[1]=pU;
37. yuv[2]=pV;
38. for(y=height;y>0;--y)
39. {
40. YUVToRGB24_Assemble(pDstLine,yuv,width);//decoderalinewithasmfunctioninYUVToRGB24_Assemble.s
41. pDstLine+=width*3;//RGB888
42. yuv[0]+=width;//YUV422
43. if(y%2==1){
44. yuv[1]+=width>>1;
45. yuv[2]+=width>>1;
46. }
47. }
48. return0;
49. }

extern int YUVToRGB24_Assemble(unsigned char *pDstLine, unsigned char **yuv, int width);//使用整數運算(定點數運算)來代替浮點運算const int csY_coeff_16 = 1.164383 * (1 << 16);const int csU_blue_16 = 2.017232 * (1 << 16);const int csU_green_16 = (-0.391762) * (1 << 16);const int csV_green_16 = (-0.812968) * (1 << 16);const int csV_red_16 = 1.596027 * (1 << 16);//查表int Ym_tableEx[256];int Um_blue_tableEx[256];int Um_green_tableEx[256];int Vm_green_tableEx[256];int Vm_red_tableEx[256];//採用查找表進行計算時，必須運行的初始化函數void YUV422P_To_RGB24_init(){int i;for (i = 0; i < 256; ++i){Ym_tableEx[i]=(csY_coeff_16 * (i - 16) )>>16;Um_blue_tableEx[i]=(csU_blue_16 * (i - 128) )>>16;Um_green_tableEx[i]=(csU_green_16 * (i - 128) )>>16;Vm_green_tableEx[i]=(csV_green_16 * (i - 128) )>>16;Vm_red_tableEx[i]=(csV_red_16 * (i - 128) )>>16;}}int YUV420P_To_RGB24(BYTE* pY, BYTE* pU, BYTE* pV, BYTE* DstPic, int width, int height){int y;BYTE* pDstLine = DstPic;BYTE* yuv[3];if ((width % 8)!=0)return(-1);yuv[0] = pY;yuv[1] = pU;yuv[2] = pV;for (y = height; y > 0; --y){YUVToRGB24_Assemble(pDstLine, yuv, width); //decoder a line with asm function in YUVToRGB24_Assemble.spDstLine += width * 3; //RGB888yuv[0] += width; //YUV422if(y % 2 == 1) {yuv[1] += width >> 1;yuv[2] += width >> 1;}}return 0;}

arm彙編核心解碼模塊：

view plaincopy to clipboardprint?

1. .text
2. .macroloadua
3. adrr1,UM_BLUE
4. ldrr1,[r1]
5. ldrr9,[r1,a,lsl#2]
6. adrr1,UM_GREEN
7. ldrr1,[r1]
8. ldrr10,[r1,a,lsl#2]
9. .endm
11. .macroloadva
12. adrr1,VM_RED
13. ldrr1,[r1]
14. ldrr11,[r1,a,lsl#2]
15. adrr1,VM_GREEN
16. ldrr1,[r1]
17. ldrr12,[r1,a,lsl#2]
18. .endm
19. .macrobound_r0
20. cmpr0,#0x00
21. movltr0,#0x00
22. cmpr0,#255
23. movgtr0,#255
24. .endm
25. .globlYUVToRGB24_Assemble
26. @r0=pDstLine;r1=yuv;r2=width
27. YUVToRGB24_Assemble:
28. stmdbsp!,{r4,r5,r6,r7,r8,r9,r10,r11,r12,lr}
30. ldmiar1,{r1,r3,r4}@r1=y;r3=u;r4=v;
32. movr5,#0@r5=rowcounter
33. hloop:
34. ldrr6,[r1],#0x04@loady;4bytes
35. ldrhr7,[r3],#0x02@loadu;2bytes
36. ldrhr8,[r4],#0x02@loadv;2bytes
38. stmdbsp!,{r0,r1,r2,r3,r4,r5}
40. @tempregister:r0,r1rgbdata:r2,r3,r4
41. @ye:r5ue_blue:r9ue_green:r10ve_red:r11ve_green:r12
43. movr2,#0
44. movr3,#0
45. movr4,#0
47. @loadue_bule0,ue_green0
48. movr0,r7
49. andr0,r0,#0xFF
50. loadur0
52. @loadve_red0,ve_green0
53. movr0,r8
54. andr0,r0,#0xFF
55. loadvr0
57. @loadye0
58. movr0,r6
59. andr0,r0,#0xFF
60. adrr1,YM
61. ldrr1,[r1]
62. ldrr5,[r1,r0,lsl#2]
64. @r0=ye0+ve_red0
65. addr0,r5,r11
66. bound_r0
68. orrr2,r2,r0
70. @g0=ye0+ue_green0+ve_green0
71. addsr0,r10,r12
72. adcr0,r0,r5
73. bound_r0
75. orrr2,r2,r0,lsl#8
77. @b0=ye0+ue_blue0
78. addr0,r5,r9
79. bound_r0
81. orrr2,r2,r0,lsl#16
83. @loadye1
84. movr0,r6,lsr#8
85. andr0,r0,#0xFF
86. ldrr5,[r1,r0,lsl#2]
88. @r1=ye1+ve_red0
89. addr0,r5,r11
90. bound_r0
92. orrr2,r2,r0,lsl#24
94. @g1=ye1+ue_green0+ve_green0
95. addsr0,r10,r12
96. adcr0,r0,r5
97. bound_r0
99. orrr3,r3,r0
101. @b1=ye1+ue_blue0
102. addr0,r5,r9
103. bound_r0
105. orrr3,r3,r0,lsl#8
107. @loadue_bule1,ue_green1
108. movr0,r7,lsr#8
109. andr0,r0,#0xFF
110. loadur0
112. @loadve_red1,ve_green1
113. movr0,r8,lsr#8
114. andr0,r0,#0xFF
115. loadvr0
117. @loadye2
118. movr0,r6,lsr#16
119. andr0,r0,#0xFF
120. adrr1,YM
121. ldrr1,[r1]
122. ldrr5,[r1,r0,lsl#2]
124. @r2=ye2+ve_red1
125. addr0,r5,r11
126. bound_r0
128. orrr3,r3,r0,lsl#16
130. @g2=ye2+ue_green1+ve_green1
131. addsr0,r10,r12
132. adcr0,r0,r5
133. bound_r0
135. addr3,r3,r0,lsl#24
137. @b2=ye2+ue_blue1
138. addr0,r5,r9
139. bound_r0
141. orrr4,r4,r0
143. @loadye3
144. movr0,r6,lsr#24
145. andr0,r0,#0xFF
146. ldrr5,[r1,r0,lsl#2]
148. @r3=ye3+ve_red1
149. addr0,r5,r11
150. bound_r0
152. orrr4,r4,r0,lsl#8
154. @g3=ye3+ue_green1+ve_green1
155. addsr0,r10,r12
156. adcr0,r0,r5
157. bound_r0
159. orrr4,r4,r0,lsl#16
161. @b3=ye3+ue_blue1
162. addr0,r5,r9
163. bound_r0
165. orrr4,r4,r0,lsl#24
167. movr10,r2
168. movr11,r3
169. movr12,r4
171. ldmiasp!,{r0,r1,r2,r3,r4,r5}
173. stmiar0!,{r10,r11,r12}
175. addr5,r5,#4
176. cmpr5,r2
177. blohloop
178. ldmiasp!,{r4,r5,r6,r7,r8,r9,r10,r11,r12,pc}@exit
180. @tables
181. YM:.longYm_tableEx
182. UM_BLUE:.longUm_blue_tableEx
183. UM_GREEN:.longUm_green_tableEx
184. VM_GREEN:.longVm_green_tableEx
185. VM_RED:.longVm_red_tableEx

.text.macro loadu aadr r1, UM_BLUEldr r1, [r1]ldr r9, [r1, a, lsl #2]adr r1, UM_GREENldr r1, [r1]ldr r10, [r1, a, lsl #2].endm.macro loadv aadr r1, VM_REDldr r1, [r1]ldr r11, [r1, a, lsl #2]adr r1, VM_GREENldr r1, [r1]ldr r12, [r1, a, lsl #2].endm.macro bound_r0cmp r0, #0x00movlt r0, #0x00cmp r0, #255movgt r0, #255.endm.globl YUVToRGB24_Assemble@ r0 = pDstLine; r1 = yuv; r2 = widthYUVToRGB24_Assemble:stmdb sp!, { r4, r5, r6, r7, r8, r9, r10, r11, r12, lr}ldmia r1, {r1, r3, r4} @r1 = y; r3 = u; r4 = v;mov r5, #0 @r5 = row counterhloop:ldr r6, [r1], #0x04 @load y; 4 bytesldrh r7, [r3], #0x02 @load u; 2 bytesldrh r8, [r4], #0x02 @load v; 2 bytesstmdb sp!, {r0, r1, r2, r3, r4, r5}@ temp register: r0,r1 rgbdata: r2,r3,r4@ ye:r5 ue_blue:r9 ue_green:r10 ve_red:r11 ve_green:r12mov r2, #0mov r3, #0mov r4, #0@ load ue_bule0, ue_green0mov r0, r7and r0, r0, #0xFFloadu r0@load ve_red0, ve_green0mov r0, r8and r0, r0, #0xFFloadv r0@load ye0mov r0, r6and r0, r0, #0xFFadr r1, YMldr r1, [r1]ldr r5, [r1, r0, lsl #2]@r0 = ye0+ve_red0add r0, r5, r11bound_r0orr r2, r2, r0@g0 = ye0+ue_green0+ve_green0adds r0, r10, r12adc r0, r0, r5bound_r0orr r2, r2, r0, lsl #8@b0 = ye0+ue_blue0add r0, r5, r9bound_r0orr r2, r2, r0, lsl #16@load ye1mov r0, r6, lsr #8and r0, r0, #0xFFldr r5, [r1, r0, lsl #2]@r1 = ye1+ve_red0add r0, r5, r11bound_r0orr r2, r2, r0, lsl #24@g1 = ye1+ue_green0+ve_green0adds r0, r10, r12adc r0, r0, r5bound_r0orr r3, r3, r0@b1 = ye1+ue_blue0add r0, r5, r9bound_r0orr r3, r3, r0, lsl #8@ load ue_bule1, ue_green1mov r0, r7, lsr #8and r0, r0, #0xFFloadu r0@load ve_red1, ve_green1mov r0, r8, lsr #8and r0, r0, #0xFFloadv r0@load ye2mov r0, r6, lsr #16and r0, r0, #0xFFadr r1, YMldr r1, [r1]ldr r5, [r1, r0, lsl #2]@r2 = ye2+ve_red1add r0, r5, r11bound_r0orr r3, r3, r0, lsl #16@g2 = ye2+ue_green1+ve_green1adds r0, r10, r12adc r0, r0, r5bound_r0add r3, r3, r0, lsl #24@b2 = ye2+ue_blue1add r0, r5, r9bound_r0orr r4, r4, r0@load ye3mov r0, r6, lsr #24and r0, r0, #0xFFldr r5, [r1, r0, lsl #2]@r3 = ye3+ve_red1add r0, r5, r11bound_r0orr r4, r4, r0, lsl #8@g3 = ye3+ue_green1+ve_green1adds r0, r10, r12adc r0, r0, r5bound_r0orr r4, r4, r0, lsl #16@b3 = ye3+ue_blue1add r0, r5, r9bound_r0orr r4, r4, r0, lsl #24mov r10, r2mov r11, r3mov r12, r4ldmia sp!, {r0, r1, r2, r3, r4, r5}stmia r0!, {r10, r11, r12}add r5, r5, #4cmp r5, r2blo hloopldmia sp!, {r4, r5, r6, r7, r8, r9, r10, r11, r12, pc} @exit@tablesYM : .long Ym_tableExUM_BLUE : .long Um_blue_tableExUM_GREEN: .long Um_green_tableExVM_GREEN: .long Vm_green_tableExVM_RED : .long Vm_red_tableEx

將核心模塊改成彙編後，解碼達到了50幀/秒，效率提高了60%，彙編果然強大，哈哈。

推薦閱讀：