diff options
Diffstat (limited to 'media/libstagefright/codecs/on2/h264dec/source/arm11_asm')
10 files changed, 2884 insertions, 0 deletions
diff --git a/media/libstagefright/codecs/on2/h264dec/source/arm11_asm/h264bsd_interpolate_chroma_hor.s b/media/libstagefright/codecs/on2/h264dec/source/arm11_asm/h264bsd_interpolate_chroma_hor.s new file mode 100755 index 0000000..634a484 --- /dev/null +++ b/media/libstagefright/codecs/on2/h264dec/source/arm11_asm/h264bsd_interpolate_chroma_hor.s @@ -0,0 +1,298 @@ +; Copyright (C) 2009 The Android Open Source Project +; +; Licensed under the Apache License, Version 2.0 (the "License"); +; you may not use this file except in compliance with the License. +; You may obtain a copy of the License at +; +; http://www.apache.org/licenses/LICENSE-2.0 +; +; Unless required by applicable law or agreed to in writing, software +; distributed under the License is distributed on an "AS IS" BASIS, +; WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +; See the License for the specific language governing permissions and +; limitations under the License. + +;------------------------------------------------------------------------------- +;-- +;-- Abstract : ARMv6 optimized version of h264bsdInterpolateChromaHor function +;-- +;------------------------------------------------------------------------------- + + + IF :DEF: H264DEC_WINASM + ;// We dont use REQUIRE8 and PRESERVE8 for winasm + ELSE + REQUIRE8 + PRESERVE8 + ENDIF + + AREA |.text|, CODE + + +;// h264bsdInterpolateChromaHor register allocation + +ref RN 0 +ptrA RN 0 + +mb RN 1 +block RN 1 + +x0 RN 2 +count RN 2 + +y0 RN 3 +valX RN 3 + +width RN 4 + +height RN 5 +tmp7 RN 5 + +chrPW RN 6 +tmp8 RN 6 + +tmp1 RN 7 +chrPH RN 7 + +tmp2 RN 8 + +tmp3 RN 9 + +tmp4 RN 10 + +tmp5 RN 11 + +tmp6 RN 12 + +c32 RN 14 +xFrac RN 14 + +;// Function exports and imports + + IMPORT h264bsdFillBlock + + EXPORT h264bsdInterpolateChromaHor + +;// Function arguments +;// +;// u8 *ref, : 0xc4 +;// u8 *predPartChroma, : 0xc8 +;// i32 x0, : 0xcc +;// i32 y0, : 0xd0 +;// u32 width, : 0xf8 +;// u32 height, : 0xfc +;// u32 xFrac, : 0x100 +;// u32 chromaPartWidth, : 0x104 +;// u32 chromaPartHeight : 0x108 + +h264bsdInterpolateChromaHor + STMFD sp!, {r0-r11,lr} + SUB sp, sp, #0xc4 + + LDR chrPW, [sp, #0x104] ;// chromaPartWidth + LDR width, [sp, #0xf8] ;// width + CMP x0, #0 + BLT do_fill + + ADD tmp6, x0, chrPW ;// tmp6 = x0+ chromaPartWidth + ADD tmp6, tmp6, #1 ;// tmp6 = x0 + chromaPartWidth + 1 + CMP tmp6, width ;// x0+chromaPartWidth+1 > width + BHI do_fill + + CMP y0, #0 + BLT do_fill + LDR chrPH, [sp, #0x108] ;// chromaPartHeight + LDR height, [sp, #0xfc] ;// height + ADD tmp6, y0, chrPH ;// tmp6 = y0 + chromaPartHeight + CMP tmp6, height + BLS skip_fill + +do_fill + LDR chrPH, [sp, #0x108] ;// chromaPartHeight + LDR height, [sp, #0xfc] ;// height + ADD tmp8, chrPW, #1 ;// tmp8 = chromaPartWidth+1 + MOV tmp2, tmp8 ;// tmp2 = chromaPartWidth+1 + STMIA sp,{width,height,tmp8,chrPH,tmp2} + ADD block, sp, #0x1c ;// block + BL h264bsdFillBlock + + LDR x0, [sp, #0xcc] + LDR y0, [sp, #0xd0] + LDR ref, [sp, #0xc4] ;// ref + STMIA sp,{width,height,tmp8,chrPH,tmp2} + ADD block, sp, #0x1c ;// block + MLA ref, height, width, ref ;// ref += width * height; + MLA block, chrPH, tmp8, block;// block + (chromaPH)*(chromaPW+1) + BL h264bsdFillBlock + + MOV x0, #0 ;// x0 = 0 + MOV y0, #0 ;// y0 = 0 + STR x0, [sp, #0xcc] + STR y0, [sp, #0xd0] + ADD ref, sp, #0x1c ;// ref = block + STR ref, [sp, #0xc4] ;// ref + + STR chrPH, [sp, #0xfc] ;// height + STR tmp8, [sp, #0xf8] ;// width + MOV width, tmp8 + SUB chrPW, chrPW, #1 + +skip_fill + MLA tmp3, y0, width, x0 ;// tmp3 = y0*width+x0 + LDR xFrac, [sp, #0x100] ;// xFrac + ADD ptrA, ref, tmp3 ;// ptrA = ref + y0*width+x0 + RSB valX, xFrac, #8 ;// valX = 8-xFrac + + LDR mb, [sp, #0xc8] ;// predPartChroma + + + ;// pack values to count register + ;// [31:28] loop_x (chromaPartWidth-1) + ;// [27:24] loop_y (chromaPartHeight-1) + ;// [23:20] chromaPartWidth-1 + ;// [19:16] chromaPartHeight-1 + ;// [15:00] nothing + + SUB tmp2, chrPH, #1 ;// chromaPartHeight-1 + SUB tmp1, chrPW, #1 ;// chromaPartWidth-1 + ADD count, count, tmp2, LSL #16 ;// chromaPartHeight-1 + ADD count, count, tmp2, LSL #24 ;// loop_y + ADD count, count, tmp1, LSL #20 ;// chromaPartWidth-1 + AND tmp2, count, #0x00F00000 ;// loop_x + PKHBT valX, valX, xFrac, LSL #16 ;// |xFrac|valX | + MOV valX, valX, LSL #3 ;// multiply by 8 in advance + MOV c32, #32 + + + ;/////////////////////////////////////////////////////////////////////////// + ;// Cb + ;/////////////////////////////////////////////////////////////////////////// + + ;// 2x2 pels per iteration + ;// bilinear vertical interpolation + +loop1_y + ADD count, count, tmp2, LSL #8 + LDRB tmp1, [ptrA, width] + LDRB tmp2, [ptrA], #1 + +loop1_x + LDRB tmp3, [ptrA, width] + LDRB tmp4, [ptrA], #1 + + PKHBT tmp5, tmp1, tmp3, LSL #16 + PKHBT tmp6, tmp2, tmp4, LSL #16 + + LDRB tmp1, [ptrA, width] + LDRB tmp2, [ptrA], #1 + + SMLAD tmp5, tmp5, valX, c32 ;// multiply + SMLAD tmp6, tmp6, valX, c32 ;// multiply + + PKHBT tmp7, tmp3, tmp1, LSL #16 + PKHBT tmp8, tmp4, tmp2, LSL #16 + + SMLAD tmp7, tmp7, valX, c32 ;// multiply + SMLAD tmp8, tmp8, valX, c32 ;// multiply + + MOV tmp5, tmp5, LSR #6 ;// scale down + STRB tmp5, [mb,#8] ;// store row 2 col 1 + + MOV tmp6, tmp6, LSR #6 ;// scale down + STRB tmp6, [mb],#1 ;// store row 1 col 1 + + MOV tmp7, tmp7, LSR #6 ;// scale down + STRB tmp7, [mb,#8] ;// store row 2 col 2 + + MOV tmp8, tmp8, LSR #6 ;// scale down + STRB tmp8, [mb],#1 ;// store row 1 col 2 + + SUBS count, count, #2<<28 + BCS loop1_x + + AND tmp2, count, #0x00F00000 + + ADDS mb, mb, #16 + SBC mb, mb, tmp2, LSR #20 + ADD ptrA, ptrA, width, LSL #1 + SBC ptrA, ptrA, tmp2, LSR #20 + SUB ptrA, ptrA, #1 + + ADDS count, count, #0xE << 24 + BGE loop1_y + + ;/////////////////////////////////////////////////////////////////////////// + ;// Cr + ;/////////////////////////////////////////////////////////////////////////// + LDR height, [sp,#0xfc] ;// height + LDR ref, [sp, #0xc4] ;// ref + LDR tmp1, [sp, #0xd0] ;// y0 + LDR tmp2, [sp, #0xcc] ;// x0 + LDR mb, [sp, #0xc8] ;// predPartChroma + + ADD tmp1, height, tmp1 + MLA tmp3, tmp1, width, tmp2 + ADD ptrA, ref, tmp3 + ADD mb, mb, #64 + + AND count, count, #0x00FFFFFF + AND tmp1, count, #0x000F0000 + ADD count, count, tmp1, LSL #8 + AND tmp2, count, #0x00F00000 + + ;// 2x2 pels per iteration + ;// bilinear vertical interpolation +loop2_y + ADD count, count, tmp2, LSL #8 + LDRB tmp1, [ptrA, width] + LDRB tmp2, [ptrA], #1 + +loop2_x + LDRB tmp3, [ptrA, width] + LDRB tmp4, [ptrA], #1 + + PKHBT tmp5, tmp1, tmp3, LSL #16 + PKHBT tmp6, tmp2, tmp4, LSL #16 + + LDRB tmp1, [ptrA, width] + LDRB tmp2, [ptrA], #1 + + SMLAD tmp5, tmp5, valX, c32 ;// multiply + SMLAD tmp6, tmp6, valX, c32 ;// multiply + + PKHBT tmp7, tmp3, tmp1, LSL #16 + PKHBT tmp8, tmp4, tmp2, LSL #16 + + SMLAD tmp7, tmp7, valX, c32 ;// multiply + SMLAD tmp8, tmp8, valX, c32 ;// multiply + + MOV tmp5, tmp5, LSR #6 ;// scale down + STRB tmp5, [mb,#8] ;// store row 2 col 1 + + MOV tmp6, tmp6, LSR #6 ;// scale down + STRB tmp6, [mb],#1 ;// store row 1 col 1 + + MOV tmp7, tmp7, LSR #6 ;// scale down + STRB tmp7, [mb,#8] ;// store row 2 col 2 + + MOV tmp8, tmp8, LSR #6 ;// scale down + STRB tmp8, [mb],#1 ;// store row 1 col 2 + + SUBS count, count, #2<<28 + BCS loop2_x + + AND tmp2, count, #0x00F00000 + + ADDS mb, mb, #16 + SBC mb, mb, tmp2, LSR #20 + ADD ptrA, ptrA, width, LSL #1 + SBC ptrA, ptrA, tmp2, LSR #20 + SUB ptrA, ptrA, #1 + + ADDS count, count, #0xE << 24 + BGE loop2_y + + ADD sp,sp,#0xd4 + LDMFD sp!, {r4-r11,pc} + + END diff --git a/media/libstagefright/codecs/on2/h264dec/source/arm11_asm/h264bsd_interpolate_chroma_hor_ver.s b/media/libstagefright/codecs/on2/h264dec/source/arm11_asm/h264bsd_interpolate_chroma_hor_ver.s new file mode 100755 index 0000000..7420ad3 --- /dev/null +++ b/media/libstagefright/codecs/on2/h264dec/source/arm11_asm/h264bsd_interpolate_chroma_hor_ver.s @@ -0,0 +1,339 @@ +; Copyright (C) 2009 The Android Open Source Project +; +; Licensed under the Apache License, Version 2.0 (the "License"); +; you may not use this file except in compliance with the License. +; You may obtain a copy of the License at +; +; http://www.apache.org/licenses/LICENSE-2.0 +; +; Unless required by applicable law or agreed to in writing, software +; distributed under the License is distributed on an "AS IS" BASIS, +; WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +; See the License for the specific language governing permissions and +; limitations under the License. + +;------------------------------------------------------------------------------- +;-- +;-- Abstract : ARMv6 optimized version of h264bsdInterpolateChromaHorVer +;-- function +;-- +;------------------------------------------------------------------------------- + + + IF :DEF: H264DEC_WINASM + ;// We dont use REQUIRE8 and PRESERVE8 for winasm + ELSE + REQUIRE8 + PRESERVE8 + ENDIF + + AREA |.text|, CODE + + +;// h264bsdInterpolateChromaHorVer register allocation + +ref RN 0 +ptrA RN 0 + +mb RN 1 +block RN 1 + +x0 RN 2 +count RN 2 + +y0 RN 3 +valY RN 3 + +width RN 4 + +tmp4 RN 5 +height RN 5 + +tmp1 RN 6 + +tmp2 RN 7 + +tmp3 RN 8 + +valX RN 9 + +tmp5 RN 10 +chrPW RN 10 + +tmp6 RN 11 +chrPH RN 11 + +xFrac RN 12 + +c32 RN 14 +yFrac RN 14 + +;// function exports and imports + + IMPORT h264bsdFillBlock + + EXPORT h264bsdInterpolateChromaHorVer + +;// Function arguments +;// +;// u8 *ref, : 0xc4 +;// u8 *predPartChroma, : 0xc8 +;// i32 x0, : 0xcc +;// i32 y0, : 0xd0 +;// u32 width, : 0xf8 +;// u32 height, : 0xfc +;// u32 xFrac, : 0x100 +;// u32 yFrac, : 0x104 +;// u32 chromaPartWidth, : 0x108 +;// u32 chromaPartHeight : 0x10c + +h264bsdInterpolateChromaHorVer + STMFD sp!, {r0-r11,lr} + SUB sp, sp, #0xc4 + + LDR chrPW, [sp, #0x108] ;// chromaPartWidth + LDR xFrac, [sp, #0x100] ;// xFrac + LDR width, [sp, #0xf8] ;// width + CMP x0, #0 + BLT do_fill + + ADD tmp1, x0, chrPW ;// tmp1 = x0+ chromaPartWidth + ADD tmp1, tmp1, #1 ;// tmp1 = x0+ chromaPartWidth+1 + CMP tmp1, width ;// x0+chromaPartWidth+1 > width + BHI do_fill + + CMP y0, #0 + BLT do_fill + LDR chrPH, [sp, #0x10c] ;// chromaPartHeight + LDR height, [sp, #0xfc] ;// height + ADD tmp1, y0, chrPH ;// tmp1 = y0 + chromaPartHeight + ADD tmp1, tmp1, #1 ;// tmp1 = y0 + chromaPartHeight + 1 + CMP tmp1, height + BLS skip_fill + +do_fill + LDR chrPH, [sp, #0x10c] ;// chromaPartHeight + LDR height, [sp, #0xfc] ;// height + ADD tmp3, chrPW, #1 ;// tmp3 = chromaPartWidth+1 + ADD tmp1, chrPW, #1 ;// tmp1 = chromaPartWidth+1 + ADD tmp2, chrPH, #1 ;// tmp2 = chromaPartHeight+1 + STMIA sp,{width,height,tmp1,tmp2,tmp3} + ADD block, sp, #0x1c ;// block + BL h264bsdFillBlock + + LDR x0, [sp, #0xcc] + LDR y0, [sp, #0xd0] + LDR ref, [sp, #0xc4] ;// ref + STMIA sp,{width,height,tmp1,tmp2,tmp3} + ADD block, sp, #0x1c ;// block + MLA ref, height, width, ref ;// ref += width * height; + MLA block, tmp2, tmp1, block;// block + (chromaPW+1)*(chromaPH+1) + BL h264bsdFillBlock + + MOV x0, #0 ;// x0 = 0 + MOV y0, #0 ;// y0 = 0 + STR x0, [sp, #0xcc] + STR y0, [sp, #0xd0] + ADD ref, sp, #0x1c ;// ref = block + STR ref, [sp, #0xc4] ;// ref + + STR tmp2, [sp, #0xfc] ;// height + STR tmp1, [sp, #0xf8] ;// width + MOV width, tmp1 + +skip_fill + MLA tmp3, y0, width, x0 ;// tmp3 = y0*width+x0 + LDR yFrac, [sp, #0x104] ;// yFrac + LDR xFrac, [sp, #0x100] + ADD ptrA, ref, tmp3 ;// ptrA = ref + y0*width+x0 + RSB valX, xFrac, #8 ;// valX = 8-xFrac + RSB valY, yFrac, #8 ;// valY = 8-yFrac + + LDR mb, [sp, #0xc8] ;// predPartChroma + + + ;// pack values to count register + ;// [31:28] loop_x (chromaPartWidth-1) + ;// [27:24] loop_y (chromaPartHeight-1) + ;// [23:20] chromaPartWidth-1 + ;// [19:16] chromaPartHeight-1 + ;// [15:00] nothing + + SUB tmp2, chrPH, #1 ;// chromaPartHeight-1 + SUB tmp1, chrPW, #1 ;// chromaPartWidth-1 + ADD count, count, tmp2, LSL #16 ;// chromaPartHeight-1 + ADD count, count, tmp2, LSL #24 ;// loop_y + ADD count, count, tmp1, LSL #20 ;// chromaPartWidth-1 + AND tmp2, count, #0x00F00000 ;// loop_x + PKHBT valY, valY, yFrac, LSL #16 ;// |yFrac|valY | + MOV c32, #32 + + + ;/////////////////////////////////////////////////////////////////////////// + ;// Cb + ;/////////////////////////////////////////////////////////////////////////// + + ;// 2x2 pels per iteration + ;// bilinear vertical and horizontal interpolation + +loop1_y + LDRB tmp1, [ptrA] + LDRB tmp3, [ptrA, width] + LDRB tmp5, [ptrA, width, LSL #1] + + PKHBT tmp1, tmp1, tmp3, LSL #16 ;// |t3|t1| + PKHBT tmp3, tmp3, tmp5, LSL #16 ;// |t5|t3| + + SMUAD tmp1, tmp1, valY ;// t1=(t1*valY + t3*yFrac) + SMUAD tmp3, tmp3, valY ;// t3=(t3*valY + t5*yFrac) + + ADD count, count, tmp2, LSL #8 +loop1_x + ;// first + LDRB tmp2, [ptrA, #1]! + LDRB tmp4, [ptrA, width] + LDRB tmp6, [ptrA, width, LSL #1] + + PKHBT tmp2, tmp2, tmp4, LSL #16 ;// |t4|t2| + PKHBT tmp4, tmp4, tmp6, LSL #16 ;// |t6|t4| + + SMUAD tmp2, tmp2, valY ;// t2=(t2*valY + t4*yFrac) + MLA tmp5, tmp1, valX, c32 ;// t5=t1*valX+32 + MLA tmp5, tmp2, xFrac, tmp5 ;// t5=t2*xFrac+t5 + + SMUAD tmp4, tmp4, valY ;// t4=(t4*valY + t6*yFrac) + MLA tmp6, tmp3, valX, c32 ;// t3=t3*valX+32 + MLA tmp6, tmp4, xFrac, tmp6 ;// t6=t4*xFrac+t6 + + MOV tmp6, tmp6, LSR #6 ;// scale down + STRB tmp6, [mb, #8] ;// store pixel + MOV tmp5, tmp5, LSR #6 ;// scale down + STRB tmp5, [mb], #1 ;// store pixel + + ;// second + LDRB tmp1, [ptrA, #1]! + LDRB tmp3, [ptrA, width] + LDRB tmp5, [ptrA, width, LSL #1] + + PKHBT tmp1, tmp1, tmp3, LSL #16 ;// |t3|t1| + PKHBT tmp3, tmp3, tmp5, LSL #16 ;// |t5|t3| + + SMUAD tmp1, tmp1, valY ;// t1=(t1*valY + t3*yFrac) + MLA tmp5, tmp1, xFrac, c32 ;// t1=t1*xFrac+32 + MLA tmp5, tmp2, valX, tmp5 ;// t5=t2*valX+t5 + + SMUAD tmp3, tmp3, valY ;// t3=(t3*valY + t5*yFrac) + MLA tmp6, tmp3, xFrac, c32 ;// t3=t3*xFrac+32 + MLA tmp6, tmp4, valX, tmp6 ;// t6=t4*valX+t6 + + MOV tmp6, tmp6, LSR #6 ;// scale down + STRB tmp6, [mb, #8] ;// store pixel + MOV tmp5, tmp5, LSR #6 ;// scale down + STRB tmp5, [mb], #1 ;// store pixel + + SUBS count, count, #2<<28 + BCS loop1_x + + AND tmp2, count, #0x00F00000 + + ADDS mb, mb, #16 + SBC mb, mb, tmp2, LSR #20 + ADD ptrA, ptrA, width, LSL #1 + SBC ptrA, ptrA, tmp2, LSR #20 + + ADDS count, count, #0xE << 24 + BGE loop1_y + + ;/////////////////////////////////////////////////////////////////////////// + ;// Cr + ;/////////////////////////////////////////////////////////////////////////// + LDR height, [sp,#0xfc] ;// height + LDR ref, [sp, #0xc4] ;// ref + LDR tmp1, [sp, #0xd0] ;// y0 + LDR tmp2, [sp, #0xcc] ;// x0 + LDR mb, [sp, #0xc8] ;// predPartChroma + + ADD tmp1, height, tmp1 + MLA tmp3, tmp1, width, tmp2 + ADD ptrA, ref, tmp3 + ADD mb, mb, #64 + + AND count, count, #0x00FFFFFF + AND tmp1, count, #0x000F0000 + ADD count, count, tmp1, LSL #8 + AND tmp2, count, #0x00F00000 + + ;// 2x2 pels per iteration + ;// bilinear vertical and horizontal interpolation +loop2_y + LDRB tmp1, [ptrA] + LDRB tmp3, [ptrA, width] + LDRB tmp5, [ptrA, width, LSL #1] + + PKHBT tmp1, tmp1, tmp3, LSL #16 ;// |t3|t1| + PKHBT tmp3, tmp3, tmp5, LSL #16 ;// |t5|t3| + + SMUAD tmp1, tmp1, valY ;// t1=(t1*valY + t3*yFrac) + SMUAD tmp3, tmp3, valY ;// t3=(t3*valY + t5*yFrac) + + ADD count, count, tmp2, LSL #8 +loop2_x + ;// first + LDRB tmp2, [ptrA, #1]! + LDRB tmp4, [ptrA, width] + LDRB tmp6, [ptrA, width, LSL #1] + + PKHBT tmp2, tmp2, tmp4, LSL #16 ;// |t4|t2| + PKHBT tmp4, tmp4, tmp6, LSL #16 ;// |t6|t4| + + SMUAD tmp2, tmp2, valY ;// t2=(t2*valY + t4*yFrac) + MLA tmp5, tmp1, valX, c32 ;// t5=t1*valX+32 + MLA tmp5, tmp2, xFrac, tmp5 ;// t5=t2*xFrac+t5 + + SMUAD tmp4, tmp4, valY ;// t4=(t4*valY + t6*yFrac) + MLA tmp6, tmp3, valX, c32 ;// t3=t3*valX+32 + MLA tmp6, tmp4, xFrac, tmp6 ;// t6=t4*xFrac+t6 + + MOV tmp6, tmp6, LSR #6 ;// scale down + STRB tmp6, [mb, #8] ;// store pixel + MOV tmp5, tmp5, LSR #6 ;// scale down + STRB tmp5, [mb], #1 ;// store pixel + + ;// second + LDRB tmp1, [ptrA, #1]! + LDRB tmp3, [ptrA, width] + LDRB tmp5, [ptrA, width, LSL #1] + + PKHBT tmp1, tmp1, tmp3, LSL #16 ;// |t3|t1| + PKHBT tmp3, tmp3, tmp5, LSL #16 ;// |t5|t3| + + SMUAD tmp1, tmp1, valY ;// t1=(t1*valY + t3*yFrac) + MLA tmp5, tmp1, xFrac, c32 ;// t1=t1*xFrac+32 + MLA tmp5, tmp2, valX, tmp5 ;// t5=t2*valX+t5 + + SMUAD tmp3, tmp3, valY ;// t3=(t3*valY + t5*yFrac) + MLA tmp6, tmp3, xFrac, c32 ;// t3=t3*xFrac+32 + MLA tmp6, tmp4, valX, tmp6 ;// t6=t4*valX+t6 + + MOV tmp6, tmp6, LSR #6 ;// scale down + STRB tmp6, [mb, #8] ;// store pixel + MOV tmp5, tmp5, LSR #6 ;// scale down + STRB tmp5, [mb], #1 ;// store pixel + + SUBS count, count, #2<<28 + BCS loop2_x + + AND tmp2, count, #0x00F00000 + + ADDS mb, mb, #16 + SBC mb, mb, tmp2, LSR #20 + ADD ptrA, ptrA, width, LSL #1 + SBC ptrA, ptrA, tmp2, LSR #20 + + ADDS count, count, #0xE << 24 + BGE loop2_y + + ADD sp,sp,#0xd4 + LDMFD sp!,{r4-r11,pc} + + END diff --git a/media/libstagefright/codecs/on2/h264dec/source/arm11_asm/h264bsd_interpolate_chroma_ver.s b/media/libstagefright/codecs/on2/h264dec/source/arm11_asm/h264bsd_interpolate_chroma_ver.s new file mode 100755 index 0000000..af9df1b --- /dev/null +++ b/media/libstagefright/codecs/on2/h264dec/source/arm11_asm/h264bsd_interpolate_chroma_ver.s @@ -0,0 +1,288 @@ +; Copyright (C) 2009 The Android Open Source Project +; +; Licensed under the Apache License, Version 2.0 (the "License"); +; you may not use this file except in compliance with the License. +; You may obtain a copy of the License at +; +; http://www.apache.org/licenses/LICENSE-2.0 +; +; Unless required by applicable law or agreed to in writing, software +; distributed under the License is distributed on an "AS IS" BASIS, +; WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +; See the License for the specific language governing permissions and +; limitations under the License. + +;------------------------------------------------------------------------------- +;-- +;-- Abstract : ARMv6 optimized version of h264bsdInterpolateChromaVer function +;-- +;------------------------------------------------------------------------------- + + + IF :DEF: H264DEC_WINASM + ;// We dont use REQUIRE8 and PRESERVE8 for winasm + ELSE + REQUIRE8 + PRESERVE8 + ENDIF + + AREA |.text|, CODE + +;// h264bsdInterpolateChromaVer register allocation + +ref RN 0 +ptrA RN 0 + +mb RN 1 +block RN 1 + +x0 RN 2 +count RN 2 + +y0 RN 3 +valY RN 3 + +width RN 4 + +height RN 5 +tmp7 RN 5 + +chrPW RN 6 +tmp8 RN 6 + +tmp1 RN 7 + +tmp2 RN 8 + +tmp3 RN 9 + +tmp4 RN 10 + +tmp5 RN 11 +chrPH RN 11 + +tmp6 RN 12 + +c32 RN 14 +yFrac RN 14 + +;// Function exports and imports + + IMPORT h264bsdFillBlock + + EXPORT h264bsdInterpolateChromaVer + +;// Function arguments +;// +;// u8 *ref, : 0xc4 +;// u8 *predPartChroma, : 0xc8 +;// i32 x0, : 0xcc +;// i32 y0, : 0xd0 +;// u32 width, : 0xf8 +;// u32 height, : 0xfc +;// u32 yFrac, : 0x100 +;// u32 chromaPartWidth, : 0x104 +;// u32 chromaPartHeight : 0x108 + +h264bsdInterpolateChromaVer + STMFD sp!, {r0-r11,lr} + SUB sp, sp, #0xc4 + + LDR chrPW, [sp, #0x104] ;// chromaPartWidth + LDR width, [sp, #0xf8] ;// width + CMP x0, #0 + BLT do_fill + + ADD tmp1, x0, chrPW ;// tmp1 = x0+ chromaPartWidth + CMP tmp1, width ;// x0+chromaPartWidth > width + BHI do_fill + + CMP y0, #0 + BLT do_fill + LDR chrPH, [sp, #0x108] ;// chromaPartHeight + LDR height, [sp, #0xfc] ;// height + ADD tmp1, y0, chrPH ;// tmp1 = y0 + chromaPartHeight + ADD tmp1, tmp1, #1 ;// tmp1 = y0 + chromaPartHeight + 1 + CMP tmp1, height + BLS skip_fill + +do_fill + LDR chrPH, [sp, #0x108] ;// chromaPartHeight + LDR height, [sp, #0xfc] ;// height + ADD tmp1, chrPH, #1 ;// tmp1 = chromaPartHeight+1 + MOV tmp2, chrPW ;// tmp2 = chromaPartWidth + STMIA sp,{width,height,chrPW,tmp1,tmp2} + ADD block, sp, #0x1c ;// block + BL h264bsdFillBlock + + LDR x0, [sp, #0xcc] + LDR y0, [sp, #0xd0] + LDR ref, [sp, #0xc4] ;// ref + STMIA sp,{width,height,chrPW,tmp1,tmp2} + ADD block, sp, #0x1c ;// block + MLA ref, height, width, ref ;// ref += width * height; + MLA block, chrPW, tmp1, block;// block + (chromaPW)*(chromaPH+1) + BL h264bsdFillBlock + + MOV x0, #0 ;// x0 = 0 + MOV y0, #0 ;// y0 = 0 + STR x0, [sp, #0xcc] + STR y0, [sp, #0xd0] + ADD ref, sp, #0x1c ;// ref = block + STR ref, [sp, #0xc4] ;// ref + + STR tmp1, [sp, #0xfc] ;// height + STR chrPW, [sp, #0xf8] ;// width + MOV width, chrPW + +skip_fill + MLA tmp3, y0, width, x0 ;// tmp3 = y0*width+x0 + LDR yFrac, [sp, #0x100] ;// yFrac + ADD ptrA, ref, tmp3 ;// ptrA = ref + y0*width+x0 + RSB valY, yFrac, #8 ;// valY = 8-yFrac + + LDR mb, [sp, #0xc8] ;// predPartChroma + + + ;// pack values to count register + ;// [31:28] loop_x (chromaPartWidth-1) + ;// [27:24] loop_y (chromaPartHeight-1) + ;// [23:20] chromaPartWidth-1 + ;// [19:16] chromaPartHeight-1 + ;// [15:00] nothing + + SUB tmp2, chrPH, #1 ;// chromaPartHeight-1 + SUB tmp1, chrPW, #1 ;// chromaPartWidth-1 + ADD count, count, tmp2, LSL #16 ;// chromaPartHeight-1 + ADD count, count, tmp2, LSL #24 ;// loop_y + ADD count, count, tmp1, LSL #20 ;// chromaPartWidth-1 + AND tmp2, count, #0x00F00000 ;// loop_x + PKHBT valY, valY, yFrac, LSL #16 ;// |yFrac|valY | + MOV valY, valY, LSL #3 ;// multiply by 8 in advance + MOV c32, #32 + + + ;/////////////////////////////////////////////////////////////////////////// + ;// Cb + ;/////////////////////////////////////////////////////////////////////////// + + ;// 2x2 pels per iteration + ;// bilinear vertical interpolation + +loop1_y + ADD count, count, tmp2, LSL #8 +loop1_x + ;// Process 2x2 block + LDRB tmp2, [ptrA,width] ;// 2 row, 1 col + LDRB tmp3, [ptrA,width, LSL #1] ;// 3 row, 1 col + LDRB tmp1, [ptrA],#1 ;// 1 row, 1 col + + LDRB tmp5, [ptrA,width] ;// 2 row, 2 col + LDRB tmp6, [ptrA,width, LSL #1] ;// 3 row, 2 col + LDRB tmp4, [ptrA],#1 ;// 1 row, 2 col + + PKHBT tmp1, tmp1, tmp2, LSL #16 ;// |B|A| + PKHBT tmp2, tmp2, tmp3, LSL #16 ;// |C|B| + PKHBT tmp4, tmp4, tmp5, LSL #16 ;// |B|A| + + SMLAD tmp7, tmp2, valY, c32 ;// multiply + PKHBT tmp5, tmp5, tmp6, LSL #16 ;// |C|B| + SMLAD tmp2, tmp1, valY, c32 ;// multiply + SMLAD tmp8, tmp5, valY, c32 ;// multiply + SMLAD tmp5, tmp4, valY, c32 ;// multiply + + MOV tmp7, tmp7, LSR #6 ;// scale down + STRB tmp7, [mb,#8] ;// store row 2 col 1 + MOV tmp2, tmp2, LSR #6 ;// scale down + STRB tmp2, [mb],#1 ;// store row 1 col 1 + + MOV tmp8, tmp8, LSR #6 ;// scale down + STRB tmp8, [mb,#8] ;// store row 2 col 2 + MOV tmp5, tmp5, LSR #6 ;// scale down + STRB tmp5, [mb],#1 ;// store row 1 col 2 + + + SUBS count, count, #2<<28 + BCS loop1_x + + AND tmp2, count, #0x00F00000 + + ADDS mb, mb, #16 + SBC mb, mb, tmp2, LSR #20 + ADD ptrA, ptrA, width, LSL #1 + SBC ptrA, ptrA, tmp2, LSR #20 + + ADDS count, count, #0xE << 24 + BGE loop1_y + + ;/////////////////////////////////////////////////////////////////////////// + ;// Cr + ;/////////////////////////////////////////////////////////////////////////// + LDR height, [sp,#0xfc] ;// height + LDR ref, [sp, #0xc4] ;// ref + LDR tmp1, [sp, #0xd0] ;// y0 + LDR tmp2, [sp, #0xcc] ;// x0 + LDR mb, [sp, #0xc8] ;// predPartChroma + + ADD tmp1, height, tmp1 + MLA tmp3, tmp1, width, tmp2 + ADD ptrA, ref, tmp3 + ADD mb, mb, #64 + + AND count, count, #0x00FFFFFF + AND tmp1, count, #0x000F0000 + ADD count, count, tmp1, LSL #8 + AND tmp2, count, #0x00F00000 + + ;// 2x2 pels per iteration + ;// bilinear vertical interpolation +loop2_y + ADD count, count, tmp2, LSL #8 +loop2_x + ;// Process 2x2 block + LDRB tmp2, [ptrA,width] ;// 2 row, 1 col + LDRB tmp3, [ptrA,width, LSL #1] ;// 3 row, 1 col + LDRB tmp1, [ptrA],#1 ;// 1 row, 1 col + + LDRB tmp5, [ptrA,width] ;// 2 row, 2 col + LDRB tmp6, [ptrA,width, LSL #1] ;// 3 row, 2 col + LDRB tmp4, [ptrA],#1 ;// 1 row, 2 col + + PKHBT tmp1, tmp1, tmp2, LSL #16 ;// |B|A| + PKHBT tmp2, tmp2, tmp3, LSL #16 ;// |C|B| + PKHBT tmp4, tmp4, tmp5, LSL #16 ;// |B|A| + + SMLAD tmp7, tmp2, valY, c32 ;// multiply + PKHBT tmp5, tmp5, tmp6, LSL #16 ;// |C|B| + SMLAD tmp2, tmp1, valY, c32 ;// multiply + SMLAD tmp8, tmp5, valY, c32 ;// multiply + SMLAD tmp5, tmp4, valY, c32 ;// multiply + + MOV tmp7, tmp7, LSR #6 ;// scale down + STRB tmp7, [mb,#8] ;// store row 2 col 1 + MOV tmp2, tmp2, LSR #6 ;// scale down + STRB tmp2, [mb],#1 ;// store row 1 col 1 + + MOV tmp8, tmp8, LSR #6 ;// scale down + STRB tmp8, [mb,#8] ;// store row 2 col 2 + MOV tmp5, tmp5, LSR #6 ;// scale down + STRB tmp5, [mb],#1 ;// store row 1 col 2 + + + SUBS count, count, #2<<28 + BCS loop2_x + + AND tmp2, count, #0x00F00000 + + ADDS mb, mb, #16 + SBC mb, mb, tmp2, LSR #20 + ADD ptrA, ptrA, width, LSL #1 + SBC ptrA, ptrA, tmp2, LSR #20 + + ADDS count, count, #0xE << 24 + BGE loop2_y + + ADD sp,sp,#0xd4 + LDMFD sp!, {r4-r11,pc} + + END diff --git a/media/libstagefright/codecs/on2/h264dec/source/arm11_asm/h264bsd_interpolate_hor_half.s b/media/libstagefright/codecs/on2/h264dec/source/arm11_asm/h264bsd_interpolate_hor_half.s new file mode 100755 index 0000000..93968b6 --- /dev/null +++ b/media/libstagefright/codecs/on2/h264dec/source/arm11_asm/h264bsd_interpolate_hor_half.s @@ -0,0 +1,251 @@ +; Copyright (C) 2009 The Android Open Source Project +; +; Licensed under the Apache License, Version 2.0 (the "License"); +; you may not use this file except in compliance with the License. +; You may obtain a copy of the License at +; +; http://www.apache.org/licenses/LICENSE-2.0 +; +; Unless required by applicable law or agreed to in writing, software +; distributed under the License is distributed on an "AS IS" BASIS, +; WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +; See the License for the specific language governing permissions and +; limitations under the License. + +;------------------------------------------------------------------------------- +;-- +;-- Abstract : ARMv6 optimized version of h264bsdInterpolateHorHalf function +;-- +;------------------------------------------------------------------------------- + + + IF :DEF: H264DEC_WINASM + ;// We dont use REQUIRE8 and PRESERVE8 for winasm + ELSE + REQUIRE8 + PRESERVE8 + ENDIF + + AREA |.text|, CODE + +;// h264bsdInterpolateHorHalf register allocation + +ref RN 0 + +mb RN 1 +buff RN 1 + +count RN 2 +x0 RN 2 + +y0 RN 3 +x_2_0 RN 3 + +width RN 4 +x_3_1 RN 4 + +height RN 5 +x_6_4 RN 5 + +partW RN 6 +x_7_5 RN 6 + +partH RN 7 +tmp1 RN 7 + +tmp2 RN 8 + +tmp3 RN 9 + +tmp4 RN 10 + +mult_20_01 RN 11 +mult_20_m5 RN 12 + +plus16 RN 14 + + +;// function exports and imports + + IMPORT h264bsdFillBlock + + EXPORT h264bsdInterpolateHorHalf + +;// Horizontal filter approach +;// +;// Basic idea in horizontal filtering is to adjust coefficients +;// like below. Calculation is done with 16-bit maths. +;// +;// Reg x_2_0 x_3_1 x_6_4 x_7_5 x_2_0 +;// [ 2 0 ] [ 3 1 ] [ 6 4 ] [ 7 5 ] [ 10 8 ] ... +;// y_0 = 20 1 20 -5 -5 1 +;// y_1 = -5 20 1 1 20 -5 +;// y_2 = 1 -5 -5 20 1 20 +;// y_3 = 1 20 -5 -5 20 1 + + +h264bsdInterpolateHorHalf + STMFD sp!, {r0-r11, lr} + SUB sp, sp, #0x1e4 + + CMP x0, #0 + BLT do_fill ;// (x0 < 0) + LDR partW, [sp,#0x220] ;// partWidth + ADD tmp4, x0, partW ;// (x0+partWidth) + ADD tmp4, tmp4, #5 ;// (y0+partW+5) + LDR width, [sp,#0x218] ;// width + CMP tmp4, width + BHI do_fill ;// (x0+partW)>width + + CMP y0, #0 + BLT do_fill ;// (y0 < 0) + LDR partH, [sp,#0x224] ;// partHeight + ADD tmp2, y0, partH ;// (y0+partHeight) + LDR height, [sp,#0x21c] ;// height + CMP tmp2, height + BLS skip_fill ;// no overfill needed + + +do_fill + LDR partH, [sp,#0x224] ;// partHeight + LDR height, [sp,#0x21c] ;// height + LDR partW, [sp,#0x220] ;// partWidth + ADD tmp4, partW, #5 ;// tmp4 = partW + 5; + STMIB sp, {height, tmp4} ;// sp+4 = height, sp+8 = partWidth+5 + STR partH, [sp,#0xc] ;// sp+c = partHeight + STR tmp4, [sp,#0x10] ;// sp+10 = partWidth+5 + LDR width, [sp,#0x218] ;// width + STR width, [sp,#0] ;// sp+0 = width + ADD buff, sp, #0x28 ;// buff = p1[21*21/4+1] + BL h264bsdFillBlock + + MOV x0, #0 + STR x0,[sp,#0x1ec] ;// x0 = 0 + STR x0,[sp,#0x1f0] ;// y0 = 0 + ADD ref,sp,#0x28 ;// ref = p1 + STR tmp4, [sp,#0x218] ;// width = partWidth+5 + + +skip_fill + LDR x0 ,[sp,#0x1ec] ;// x0 + LDR y0 ,[sp,#0x1f0] ;// y0 + LDR width, [sp,#0x218] ;// width + MLA tmp2, width, y0, x0 ;// y0*width+x0 + ADD ref, ref, tmp2 ;// ref += y0*width+x0 + ADD ref, ref, #8 ;// ref = ref+8 + LDR mb, [sp, #0x1e8] ;// mb + + ;// pack values to count register + ;// [31:28] loop_x (partWidth-1) + ;// [27:24] loop_y (partHeight-1) + ;// [23:20] partWidth-1 + ;// [19:16] partHeight-1 + ;// [15:00] width + MOV count, width + SUB partW, partW, #1; + SUB partH, partH, #1; + ADD tmp2, partH, partW, LSL #4 + ADD count, count, tmp2, LSL #16 + + + LDR mult_20_01, = 0x00140001 + LDR mult_20_m5, = 0x0014FFFB + MOV plus16, #16 + AND tmp1, count, #0x000F0000 ;// partHeight-1 + AND tmp3, count, #0x00F00000 ;// partWidth-1 + ADD count, count, tmp1, LSL #8 +loop_y + LDR x_3_1, [ref, #-8] + ADD count, count, tmp3, LSL #8 + LDR x_7_5, [ref, #-4] + UXTB16 x_2_0, x_3_1 + UXTB16 x_3_1, x_3_1, ROR #8 + UXTB16 x_6_4, x_7_5 + +loop_x + UXTB16 x_7_5, x_7_5, ROR #8 + + SMLAD tmp1, x_2_0, mult_20_01, plus16 + SMLATB tmp3, x_2_0, mult_20_01, plus16 + SMLATB tmp2, x_2_0, mult_20_m5, plus16 + SMLATB tmp4, x_3_1, mult_20_01, plus16 + + SMLAD tmp1, x_3_1, mult_20_m5, tmp1 + SMLATB tmp3, x_3_1, mult_20_m5, tmp3 + SMLAD tmp2, x_3_1, mult_20_01, tmp2 + LDR x_3_1, [ref], #4 + SMLAD tmp4, x_6_4, mult_20_m5, tmp4 + + SMLABB tmp1, x_6_4, mult_20_m5, tmp1 + SMLADX tmp3, x_6_4, mult_20_m5, tmp3 + SMLADX tmp2, x_6_4, mult_20_01, tmp2 + SMLADX tmp4, x_7_5, mult_20_m5, tmp4 + + SMLABB tmp1, x_7_5, mult_20_01, tmp1 + UXTB16 x_2_0, x_3_1 + SMLABB tmp2, x_7_5, mult_20_m5, tmp2 + SMLADX tmp3, x_7_5, mult_20_01, tmp3 + SMLABB tmp4, x_2_0, mult_20_01, tmp4 + + MOV tmp2, tmp2, ASR #5 + MOV tmp1, tmp1, ASR #5 + PKHBT tmp2, tmp2, tmp4, LSL #(16-5) + PKHBT tmp1, tmp1, tmp3, LSL #(16-5) + USAT16 tmp2, #8, tmp2 + USAT16 tmp1, #8, tmp1 + + SUBS count, count, #4<<28 + ORR tmp1, tmp1, tmp2, LSL #8 + STR tmp1, [mb], #4 + BCC next_y + + UXTB16 x_3_1, x_3_1, ROR #8 + + SMLAD tmp1, x_6_4, mult_20_01, plus16 + SMLATB tmp3, x_6_4, mult_20_01, plus16 + SMLATB tmp2, x_6_4, mult_20_m5, plus16 + SMLATB tmp4, x_7_5, mult_20_01, plus16 + + SMLAD tmp1, x_7_5, mult_20_m5, tmp1 + SMLATB tmp3, x_7_5, mult_20_m5, tmp3 + SMLAD tmp2, x_7_5, mult_20_01, tmp2 + LDR x_7_5, [ref], #4 + SMLAD tmp4, x_2_0, mult_20_m5, tmp4 + + SMLABB tmp1, x_2_0, mult_20_m5, tmp1 + SMLADX tmp3, x_2_0, mult_20_m5, tmp3 + SMLADX tmp2, x_2_0, mult_20_01, tmp2 + SMLADX tmp4, x_3_1, mult_20_m5, tmp4 + + SMLABB tmp1, x_3_1, mult_20_01, tmp1 + UXTB16 x_6_4, x_7_5 + SMLABB tmp2, x_3_1, mult_20_m5, tmp2 + SMLADX tmp3, x_3_1, mult_20_01, tmp3 + SMLABB tmp4, x_6_4, mult_20_01, tmp4 + + MOV tmp2, tmp2, ASR #5 + MOV tmp1, tmp1, ASR #5 + PKHBT tmp2, tmp2, tmp4, LSL #(16-5) + PKHBT tmp1, tmp1, tmp3, LSL #(16-5) + USAT16 tmp2, #8, tmp2 + USAT16 tmp1, #8, tmp1 + + SUBS count, count, #4<<28 + ORR tmp1, tmp1, tmp2, LSL #8 + STR tmp1, [mb], #4 + BCS loop_x + +next_y + AND tmp3, count, #0x00F00000 ;// partWidth-1 + SMLABB ref, count, mult_20_01, ref ;// +width + ADDS mb, mb, #16 ;// +16, Carry=0 + SBC mb, mb, tmp3, LSR #20 ;// -(partWidth-1)-1 + SBC ref, ref, tmp3, LSR #20 ;// -(partWidth-1)-1 + ADDS count, count, #(1<<28)-(1<<24) + BGE loop_y + + ADD sp,sp,#0x1f4 + LDMFD sp!, {r4-r11, pc} + + END + diff --git a/media/libstagefright/codecs/on2/h264dec/source/arm11_asm/h264bsd_interpolate_hor_quarter.s b/media/libstagefright/codecs/on2/h264dec/source/arm11_asm/h264bsd_interpolate_hor_quarter.s new file mode 100755 index 0000000..de243d4 --- /dev/null +++ b/media/libstagefright/codecs/on2/h264dec/source/arm11_asm/h264bsd_interpolate_hor_quarter.s @@ -0,0 +1,273 @@ +; Copyright (C) 2009 The Android Open Source Project +; +; Licensed under the Apache License, Version 2.0 (the "License"); +; you may not use this file except in compliance with the License. +; You may obtain a copy of the License at +; +; http://www.apache.org/licenses/LICENSE-2.0 +; +; Unless required by applicable law or agreed to in writing, software +; distributed under the License is distributed on an "AS IS" BASIS, +; WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +; See the License for the specific language governing permissions and +; limitations under the License. + +;------------------------------------------------------------------------------- +;-- +;-- Abstract : ARMv6 optimized version of h264bsdInterpolateHorQuarter function +;-- +;------------------------------------------------------------------------------- + + + IF :DEF: H264DEC_WINASM + ;// We dont use REQUIRE8 and PRESERVE8 for winasm + ELSE + REQUIRE8 + PRESERVE8 + ENDIF + + AREA |.text|, CODE + +;// h264bsdInterpolateHorQuarter register allocation + +ref RN 0 + +mb RN 1 +buff RN 1 + +count RN 2 +x0 RN 2 + +y0 RN 3 +x_2_0 RN 3 + +width RN 4 +x_3_1 RN 4 + +height RN 5 +x_6_4 RN 5 + +partW RN 6 +x_7_5 RN 6 + +partH RN 7 +tmp1 RN 7 + +tmp2 RN 8 + +tmp3 RN 9 + +tmp4 RN 10 + +mult_20_01 RN 11 + +mult_20_m5 RN 12 + +plus16 RN 14 + + +;// function exports and imports + + IMPORT h264bsdFillBlock + + EXPORT h264bsdInterpolateHorQuarter + + +;// Horizontal filter approach +;// +;// Basic idea in horizontal filtering is to adjust coefficients +;// like below. Calculation is done with 16-bit maths. +;// +;// Reg x_2_0 x_3_1 x_6_4 x_7_5 x_2_0 +;// [ 2 0 ] [ 3 1 ] [ 6 4 ] [ 7 5 ] [ 10 8 ] ... +;// y_0 = 20 1 20 -5 -5 1 +;// y_1 = -5 20 1 1 20 -5 +;// y_2 = 1 -5 -5 20 1 20 +;// y_3 = 1 20 -5 -5 20 1 + + +h264bsdInterpolateHorQuarter + STMFD sp!, {r0-r11, lr} + SUB sp, sp, #0x1e4 + + CMP x0, #0 + BLT do_fill ;// (x0 < 0) + LDR partW, [sp,#0x220] ;// partWidth + ADD tmp4, x0, partW ;// (x0+partWidth) + ADD tmp4, tmp4, #5 ;// (y0+partW+5) + LDR width, [sp,#0x218] ;// width + CMP tmp4, width + BHI do_fill ;// (x0+partW)>width + + CMP y0, #0 + BLT do_fill ;// (y0 < 0) + LDR partH, [sp,#0x224] ;// partHeight + ADD tmp2, y0, partH ;// (y0+partHeight) + LDR height, [sp,#0x21c] ;// height + CMP tmp2, height + BLS skip_fill ;// no overfill needed + + +do_fill + LDR partH, [sp,#0x224] ;// partHeight + LDR height, [sp,#0x21c] ;// height + LDR partW, [sp,#0x220] ;// partWidth + ADD tmp4, partW, #5 ;// tmp4 = partW + 5; + STMIB sp, {height, tmp4} ;// sp+4 = height, sp+8 = partWidth+5 + STR partH, [sp,#0xc] ;// sp+c = partHeight + STR tmp4, [sp,#0x10] ;// sp+10 = partWidth+5 + LDR width, [sp,#0x218] ;// width + STR width, [sp,#0] ;// sp+0 = width + ADD buff, sp, #0x28 ;// buff = p1[21*21/4+1] + BL h264bsdFillBlock + + MOV x0, #0 + STR x0,[sp,#0x1ec] ;// x0 = 0 + STR x0,[sp,#0x1f0] ;// y0 = 0 + ADD ref,sp,#0x28 ;// ref = p1 + STR tmp4, [sp,#0x218] ;// width = partWidth+5 + + +skip_fill + LDR x0 ,[sp,#0x1ec] ;// x0 + LDR y0 ,[sp,#0x1f0] ;// y0 + LDR width, [sp,#0x218] ;// width + MLA tmp2, width, y0, x0 ;// y0*width+x0 + ADD ref, ref, tmp2 ;// ref += y0*width+x0 + ADD ref, ref, #8 ;// ref = ref+8 + LDR mb, [sp, #0x1e8] ;// mb + + ;// pack values to count register + ;// [31:28] loop_x (partWidth-1) + ;// [27:24] loop_y (partHeight-1) + ;// [23:20] partWidth-1 + ;// [19:16] partHeight-1 + ;// [15:00] width + MOV count, width + SUB partW, partW, #1; + SUB partH, partH, #1; + ADD tmp2, partH, partW, LSL #4 + ADD count, count, tmp2, LSL #16 + + + LDR mult_20_01, = 0x00140001 + LDR mult_20_m5, = 0x0014FFFB + MOV plus16, #16 + AND tmp1, count, #0x000F0000 ;// partHeight-1 + AND tmp3, count, #0x00F00000 ;// partWidth-1 + ADD count, count, tmp1, LSL #8 +loop_y + LDR x_3_1, [ref, #-8] + ADD count, count, tmp3, LSL #8 + LDR x_7_5, [ref, #-4] + UXTB16 x_2_0, x_3_1 + UXTB16 x_3_1, x_3_1, ROR #8 + UXTB16 x_6_4, x_7_5 + +loop_x + UXTB16 x_7_5, x_7_5, ROR #8 + + SMLAD tmp1, x_2_0, mult_20_01, plus16 + SMLATB tmp3, x_2_0, mult_20_01, plus16 + SMLATB tmp2, x_2_0, mult_20_m5, plus16 + SMLATB tmp4, x_3_1, mult_20_01, plus16 + + SMLAD tmp1, x_3_1, mult_20_m5, tmp1 + SMLATB tmp3, x_3_1, mult_20_m5, tmp3 + SMLAD tmp2, x_3_1, mult_20_01, tmp2 + LDR x_3_1, [ref], #4 + SMLAD tmp4, x_6_4, mult_20_m5, tmp4 + + SMLABB tmp1, x_6_4, mult_20_m5, tmp1 + SMLADX tmp3, x_6_4, mult_20_m5, tmp3 + SMLADX tmp2, x_6_4, mult_20_01, tmp2 + SMLADX tmp4, x_7_5, mult_20_m5, tmp4 + + SMLABB tmp1, x_7_5, mult_20_01, tmp1 + UXTB16 x_2_0, x_3_1 + SMLABB tmp2, x_7_5, mult_20_m5, tmp2 + SMLADX tmp3, x_7_5, mult_20_01, tmp3 + SMLABB tmp4, x_2_0, mult_20_01, tmp4 + + MOV tmp2, tmp2, ASR #5 + MOV tmp1, tmp1, ASR #5 + PKHBT tmp2, tmp2, tmp4, LSL #(16-5) + PKHBT tmp1, tmp1, tmp3, LSL #(16-5) + LDR tmp4, [sp, #0x228] + USAT16 tmp2, #8, tmp2 + USAT16 tmp1, #8, tmp1 + SUB tmp4, tmp4, #10 + + SUBS count, count, #4<<28 + LDR tmp3, [ref, tmp4] + ORR tmp1, tmp1, tmp2, LSL #8 + +;// quarter pel position + LDR tmp2, = 0x80808080 + MVN tmp3, tmp3 + UHSUB8 tmp1, tmp1, tmp3 + EOR tmp1, tmp1, tmp2 + STR tmp1, [mb], #4 + + BCC next_y + + UXTB16 x_3_1, x_3_1, ROR #8 + + SMLAD tmp1, x_6_4, mult_20_01, plus16 + SMLATB tmp3, x_6_4, mult_20_01, plus16 + SMLATB tmp2, x_6_4, mult_20_m5, plus16 + SMLATB tmp4, x_7_5, mult_20_01, plus16 + + SMLAD tmp1, x_7_5, mult_20_m5, tmp1 + SMLATB tmp3, x_7_5, mult_20_m5, tmp3 + SMLAD tmp2, x_7_5, mult_20_01, tmp2 + LDR x_7_5, [ref], #4 + SMLAD tmp4, x_2_0, mult_20_m5, tmp4 + + SMLABB tmp1, x_2_0, mult_20_m5, tmp1 + SMLADX tmp3, x_2_0, mult_20_m5, tmp3 + SMLADX tmp2, x_2_0, mult_20_01, tmp2 + SMLADX tmp4, x_3_1, mult_20_m5, tmp4 + + SMLABB tmp1, x_3_1, mult_20_01, tmp1 + UXTB16 x_6_4, x_7_5 + SMLABB tmp2, x_3_1, mult_20_m5, tmp2 + SMLADX tmp3, x_3_1, mult_20_01, tmp3 + SMLABB tmp4, x_6_4, mult_20_01, tmp4 + + MOV tmp2, tmp2, ASR #5 + MOV tmp1, tmp1, ASR #5 + PKHBT tmp2, tmp2, tmp4, LSL #(16-5) + PKHBT tmp1, tmp1, tmp3, LSL #(16-5) + LDR tmp4, [sp, #0x228] + USAT16 tmp2, #8, tmp2 + USAT16 tmp1, #8, tmp1 + SUB tmp4, tmp4, #10 + + SUBS count, count, #4<<28 + LDR tmp3, [ref, tmp4] + ORR tmp1, tmp1, tmp2, LSL #8 + +;// quarter pel + LDR tmp2, = 0x80808080 + MVN tmp3, tmp3 + UHSUB8 tmp1, tmp1, tmp3 + EOR tmp1, tmp1, tmp2 + + STR tmp1, [mb], #4 + BCS loop_x + +next_y + AND tmp3, count, #0x00F00000 ;// partWidth-1 + SMLABB ref, count, mult_20_01, ref ;// +width + ADDS mb, mb, #16 ;// +16, Carry=0 + SBC mb, mb, tmp3, LSR #20 ;// -(partWidth-1)-1 + SBC ref, ref, tmp3, LSR #20 ;// -(partWidth-1)-1 + ADDS count, count, #(1<<28)-(1<<24) + BGE loop_y + + ADD sp,sp,#0x1f4 + LDMFD sp!, {r4-r11, pc} + + END + diff --git a/media/libstagefright/codecs/on2/h264dec/source/arm11_asm/h264bsd_interpolate_hor_ver_quarter.s b/media/libstagefright/codecs/on2/h264dec/source/arm11_asm/h264bsd_interpolate_hor_ver_quarter.s new file mode 100755 index 0000000..1c79b39 --- /dev/null +++ b/media/libstagefright/codecs/on2/h264dec/source/arm11_asm/h264bsd_interpolate_hor_ver_quarter.s @@ -0,0 +1,536 @@ +; Copyright (C) 2009 The Android Open Source Project +; +; Licensed under the Apache License, Version 2.0 (the "License"); +; you may not use this file except in compliance with the License. +; You may obtain a copy of the License at +; +; http://www.apache.org/licenses/LICENSE-2.0 +; +; Unless required by applicable law or agreed to in writing, software +; distributed under the License is distributed on an "AS IS" BASIS, +; WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +; See the License for the specific language governing permissions and +; limitations under the License. + +;------------------------------------------------------------------------------- +;-- +;-- Abstract : ARMv6 optimized version of h264bsdInterpolateHorVerQuarter +;-- function +;-- +;------------------------------------------------------------------------------- + + + IF :DEF: H264DEC_WINASM + ;// We dont use REQUIRE8 and PRESERVE8 for winasm + ELSE + REQUIRE8 + PRESERVE8 + ENDIF + + AREA |.text|, CODE + +;// h264bsdInterpolateHorVerQuarter register allocation + +ref RN 0 + +mb RN 1 +buff RN 1 + +count RN 2 +x0 RN 2 + +y0 RN 3 +x_2_0 RN 3 +res RN 3 + +x_3_1 RN 4 +tmp1 RN 4 + +height RN 5 +x_6_4 RN 5 +tmp2 RN 5 + +partW RN 6 +x_7_5 RN 6 +tmp3 RN 6 + +partH RN 7 +tmp4 RN 7 + +tmp5 RN 8 + +tmp6 RN 9 + +tmpa RN 10 + +mult_20_01 RN 11 +tmpb RN 11 + +mult_20_m5 RN 12 +width RN 12 + +plus16 RN 14 + + +;// function exports and imports + + IMPORT h264bsdFillBlock + + EXPORT h264bsdInterpolateHorVerQuarter + +;// Horizontal filter approach +;// +;// Basic idea in horizontal filtering is to adjust coefficients +;// like below. Calculation is done with 16-bit maths. +;// +;// Reg x_2_0 x_3_1 x_6_4 x_7_5 x_2_0 +;// [ 2 0 ] [ 3 1 ] [ 6 4 ] [ 7 5 ] [ 10 8 ] ... +;// y_0 = 20 1 20 -5 -5 1 +;// y_1 = -5 20 1 1 20 -5 +;// y_2 = 1 -5 -5 20 1 20 +;// y_3 = 1 20 -5 -5 20 1 + + +h264bsdInterpolateHorVerQuarter + STMFD sp!, {r0-r11, lr} + SUB sp, sp, #0x1e4 + + CMP x0, #0 + BLT do_fill ;// (x0 < 0) + LDR partW, [sp,#0x220] ;// partWidth + LDR width, [sp,#0x218] ;// width + ADD tmpa, x0, partW ;// (x0+partWidth) + ADD tmpa, tmpa, #5 ;// (x0+partW+5) + CMP tmpa, width + BHI do_fill ;// (x0+partW)>width + + CMP y0, #0 + BLT do_fill ;// (y0 < 0) + LDR partH, [sp,#0x224] ;// partHeight + LDR height, [sp,#0x21c] ;// height + ADD tmp5, y0, partH ;// (y0+partHeight) + ADD tmp5, tmp5, #5 ;// (y0+partH+5) + CMP tmp5, height + BLS skip_fill ;// no overfill needed + + +do_fill + LDR partH, [sp,#0x224] ;// partHeight + LDR partW, [sp,#0x220] ;// partWidth + LDR height, [sp,#0x21c] ;// height + ADD tmp5, partH, #5 ;// tmp5 = partH + 5 + ADD tmpa, partW, #5 ;// tmpa = partW + 5 + STMIB sp, {height, tmpa} ;// sp+4 = height, sp+8 = partWidth+5 + LDR width, [sp,#0x218] ;// width + STR tmp5, [sp,#0xc] ;// sp+c = partHeight+5 + STR tmpa, [sp,#0x10] ;// sp+10 = partWidth+5 + STR width, [sp,#0] ;// sp+0 = width + ADD buff, sp, #0x28 ;// buff = p1[21*21/4+1] + BL h264bsdFillBlock + + MOV x0, #0 + STR x0,[sp,#0x1ec] ;// x0 = 0 + STR x0,[sp,#0x1f0] ;// y0 = 0 + ADD ref,sp,#0x28 ;// ref = p1 + STR tmpa, [sp,#0x218] ;// width = partWidth+5 + + +skip_fill + LDR x0 ,[sp,#0x1ec] ;// x0 + LDR y0 ,[sp,#0x1f0] ;// y0 + LDR width, [sp,#0x218] ;// width + LDR tmp6, [sp,#0x228] ;// horVerOffset + LDR mb, [sp, #0x1e8] ;// mb + MLA tmp5, width, y0, x0 ;// y0*width+x0 + ADD ref, ref, tmp5 ;// ref += y0*width+x0 + STR ref, [sp, #0x1e4] ;// store "ref" for vertical filtering + AND tmp6, tmp6, #2 ;// calculate ref for horizontal filter + MOV tmpa, #2 + ADD tmp6, tmpa, tmp6, LSR #1 + MLA ref, tmp6, width, ref + ADD ref, ref, #8 ;// ref = ref+8 + + ;// pack values to count register + ;// [31:28] loop_x (partWidth-1) + ;// [27:24] loop_y (partHeight-1) + ;// [23:20] partWidth-1 + ;// [19:16] partHeight-1 + ;// [15:00] width + MOV count, width + SUB partW, partW, #1; + SUB partH, partH, #1; + ADD tmp5, partH, partW, LSL #4 + ADD count, count, tmp5, LSL #16 + + + LDR mult_20_01, = 0x00140001 ;// constant multipliers + LDR mult_20_m5, = 0x0014FFFB ;// constant multipliers + MOV plus16, #16 ;// constant for add + AND tmp4, count, #0x000F0000 ;// partHeight-1 + AND tmp6, count, #0x00F00000 ;// partWidth-1 + ADD count, count, tmp4, LSL #8 ;// partH-1 to lower part of top byte + +;// HORIZONTAL PART + +loop_y_hor + LDR x_3_1, [ref, #-8] + ADD count, count, tmp6, LSL #8 ;// partW-1 to upper part of top byte + LDR x_7_5, [ref, #-4] + UXTB16 x_2_0, x_3_1 + UXTB16 x_3_1, x_3_1, ROR #8 + UXTB16 x_6_4, x_7_5 + +loop_x_hor + UXTB16 x_7_5, x_7_5, ROR #8 + + SMLAD tmp4, x_2_0, mult_20_01, plus16 + SMLATB tmp6, x_2_0, mult_20_01, plus16 + SMLATB tmp5, x_2_0, mult_20_m5, plus16 + SMLATB tmpa, x_3_1, mult_20_01, plus16 + + SMLAD tmp4, x_3_1, mult_20_m5, tmp4 + SMLATB tmp6, x_3_1, mult_20_m5, tmp6 + SMLAD tmp5, x_3_1, mult_20_01, tmp5 + LDR x_3_1, [ref], #4 + SMLAD tmpa, x_6_4, mult_20_m5, tmpa + + SMLABB tmp4, x_6_4, mult_20_m5, tmp4 + SMLADX tmp6, x_6_4, mult_20_m5, tmp6 + SMLADX tmp5, x_6_4, mult_20_01, tmp5 + SMLADX tmpa, x_7_5, mult_20_m5, tmpa + + SMLABB tmp4, x_7_5, mult_20_01, tmp4 + UXTB16 x_2_0, x_3_1 + SMLABB tmp5, x_7_5, mult_20_m5, tmp5 + SMLADX tmp6, x_7_5, mult_20_01, tmp6 + SMLABB tmpa, x_2_0, mult_20_01, tmpa + + MOV tmp5, tmp5, ASR #5 + MOV tmp4, tmp4, ASR #5 + PKHBT tmp5, tmp5, tmpa, LSL #(16-5) + PKHBT tmp4, tmp4, tmp6, LSL #(16-5) + USAT16 tmp5, #8, tmp5 + USAT16 tmp4, #8, tmp4 + + SUBS count, count, #4<<28 + ORR tmp4, tmp4, tmp5, LSL #8 + STR tmp4, [mb], #4 + BCC next_y_hor + + UXTB16 x_3_1, x_3_1, ROR #8 + + SMLAD tmp4, x_6_4, mult_20_01, plus16 + SMLATB tmp6, x_6_4, mult_20_01, plus16 + SMLATB tmp5, x_6_4, mult_20_m5, plus16 + SMLATB tmpa, x_7_5, mult_20_01, plus16 + + SMLAD tmp4, x_7_5, mult_20_m5, tmp4 + SMLATB tmp6, x_7_5, mult_20_m5, tmp6 + SMLAD tmp5, x_7_5, mult_20_01, tmp5 + LDR x_7_5, [ref], #4 + SMLAD tmpa, x_2_0, mult_20_m5, tmpa + + SMLABB tmp4, x_2_0, mult_20_m5, tmp4 + SMLADX tmp6, x_2_0, mult_20_m5, tmp6 + SMLADX tmp5, x_2_0, mult_20_01, tmp5 + SMLADX tmpa, x_3_1, mult_20_m5, tmpa + + SMLABB tmp4, x_3_1, mult_20_01, tmp4 + UXTB16 x_6_4, x_7_5 + SMLABB tmp5, x_3_1, mult_20_m5, tmp5 + SMLADX tmp6, x_3_1, mult_20_01, tmp6 + SMLABB tmpa, x_6_4, mult_20_01, tmpa + + MOV tmp5, tmp5, ASR #5 + MOV tmp4, tmp4, ASR #5 + PKHBT tmp5, tmp5, tmpa, LSL #(16-5) + PKHBT tmp4, tmp4, tmp6, LSL #(16-5) + USAT16 tmp5, #8, tmp5 + USAT16 tmp4, #8, tmp4 + + SUBS count, count, #4<<28 + ORR tmp4, tmp4, tmp5, LSL #8 + STR tmp4, [mb], #4 + BCS loop_x_hor + +next_y_hor + AND tmp6, count, #0x00F00000 ;// partWidth-1 + SMLABB ref, count, mult_20_01, ref ;// +width + ADDS mb, mb, #16 ;// +16, Carry=0 + SBC mb, mb, tmp6, LSR #20 ;// -(partWidth-1)-1 + SBC ref, ref, tmp6, LSR #20 ;// -(partWidth-1)-1 + ADDS count, count, #(1<<28)-(1<<24) ;// decrement counter (partW) + BGE loop_y_hor + + + +;// VERTICAL PART +;// +;// Approach to vertical interpolation +;// +;// Interpolation is done by using 32-bit loads and stores +;// and by using 16 bit arithmetic. 4x4 block is processed +;// in each round. +;// +;// |a_11|a_11|a_11|a_11|...|a_1n|a_1n|a_1n|a_1n| +;// |b_11|b_11|b_11|b_11|...|b_1n|b_1n|b_1n|b_1n| +;// |c_11|c_11|c_11|c_11|...|c_1n|c_1n|c_1n|c_1n| +;// |d_11|d_11|d_11|d_11|...|d_1n|d_1n|d_1n|d_1n| +;// .. +;// .. +;// |a_m1|a_m1|a_m1|a_m1|... +;// |b_m1|b_m1|b_m1|b_m1|... +;// |c_m1|c_m1|c_m1|c_m1|... +;// |d_m1|d_m1|d_m1|d_m1|... + +;// Approach to bilinear interpolation to quarter pel position. +;// 4 bytes are processed parallel +;// +;// algorithm (a+b+1)/2. Rouding upwards +1 can be achieved by +;// negating second operand to get one's complement (instead of 2's) +;// and using subtraction, EOR is used to correct sign. +;// +;// MVN b, b +;// UHSUB8 a, a, b +;// EOR a, a, 0x80808080 + + + LDR ref, [sp, #0x1e4] ;// ref + LDR tmpa, [sp, #0x228] ;// horVerOffset + LDR mb, [sp, #0x1e8] ;// mb + LDR width, [sp, #0x218] ;// width + ADD ref, ref, #2 ;// calculate correct position + AND tmpa, tmpa, #1 + ADD ref, ref, tmpa + LDR plus16, = 0x00100010 ;// +16 to lower and upperf halfwords + AND count, count, #0x00FFFFFF ;// partWidth-1 + + AND tmpa, count, #0x000F0000 ;// partHeight-1 + ADD count, count, tmpa, LSL #8 + +loop_y + ADD count, count, tmp6, LSL #8 ;// partWidth-1 + +loop_x + LDR tmp1, [ref], width ;// |a4|a3|a2|a1| + LDR tmp2, [ref], width ;// |c4|c3|c2|c1| + LDR tmp3, [ref], width ;// |g4|g3|g2|g1| + LDR tmp4, [ref], width ;// |m4|m3|m2|m1| + LDR tmp5, [ref], width ;// |r4|r3|r2|r1| + LDR tmp6, [ref], width ;// |t4|t3|t2|t1| + + ;// first four pixels + UXTB16 tmpa, tmp3 ;// |g3|g1| + UXTAB16 tmpa, tmpa, tmp4 ;// |g3+m3|g1+m1| + UXTB16 tmpb, tmp2 ;// |c3|c1| + ADD tmpa, tmpa, tmpa, LSL #2 ;// 5(G+M) + + UXTAB16 tmpb, tmpb, tmp5 ;// |c3+r3|c1+r1| + ADD tmpa, plus16, tmpa, LSL #2 ;// 16+20(G+M) + UXTAB16 tmpa, tmpa, tmp1 ;// 16+20(G+M)+A + UXTAB16 tmpa, tmpa, tmp6 ;// 16+20(G+M)+A+T + + ADD tmpb, tmpb, tmpb, LSL #2 ;// 5(C+R) + SSUB16 tmpa, tmpa, tmpb ;// 16+20(G+M)+(A+T)-5(C+R) + + USAT16 tmpb, #13, tmpa ;// saturate + LDR res, = 0x00FF00FF + UXTB16 tmpa, tmp3, ROR #8 ;// |g4|g2| + UXTAB16 tmpa, tmpa, tmp4, ROR #8 ;// |g4+m4|g2+m2| + AND res, res, tmpb, LSR #5 ;// mask and divide by 32 + + ADD tmpa, tmpa, tmpa, LSL #2 ;// 5(G+M) + UXTB16 tmpb, tmp2, ROR #8 ;// |c4|c2| + ADD tmpa, plus16, tmpa, LSL #2 ;// 16+20(G+M) + UXTAB16 tmpb, tmpb, tmp5, ROR #8 ;// |c4+r4|c2+r2| + UXTAB16 tmpa, tmpa, tmp1, ROR #8 ;// 16+20(G+M)+A + UXTAB16 tmpa, tmpa, tmp6, ROR #8 ;// 16+20(G+M)+A+T + + ADD tmpb, tmpb, tmpb, LSL #2 ;// 5(C+R) + SSUB16 tmpa, tmpa, tmpb ;// 16+20(G+M)+(A+T)-5(C+R) + + USAT16 tmpb, #13, tmpa ;// saturate + LDR tmp1, [mb] + LDR tmpa, = 0xFF00FF00 + MVN tmp1, tmp1 + AND tmpa, tmpa, tmpb, LSL #3 ;// mask and divede by 32 + ORR res, res, tmpa + + LDR tmpa, = 0x80808080 + UHSUB8 res, res, tmp1 ;// bilinear interpolation + LDR tmp1, [ref], width ;// load next row + EOR res, res, tmpa ;// correct sign + + STR res, [mb], #16 ;// next row (mb) + + + ;// tmp2 = |a4|a3|a2|a1| + ;// tmp3 = |c4|c3|c2|c1| + ;// tmp4 = |g4|g3|g2|g1| + ;// tmp5 = |m4|m3|m2|m1| + ;// tmp6 = |r4|r3|r2|r1| + ;// tmp1 = |t4|t3|t2|t1| + + ;// second four pixels + UXTB16 tmpa, tmp4 ;// |g3|g1| + UXTAB16 tmpa, tmpa, tmp5 ;// |g3+m3|g1+m1| + UXTB16 tmpb, tmp3 ;// |c3|c1| + ADD tmpa, tmpa, tmpa, LSL #2 ;// 5(G+M) + UXTAB16 tmpb, tmpb, tmp6 ;// |c3+r3|c1+r1| + ADD tmpa, plus16, tmpa, LSL #2 ;// 16+20(G+M) + UXTAB16 tmpa, tmpa, tmp2 ;// 16+20(G+M)+A + UXTAB16 tmpa, tmpa, tmp1 ;// 16+20(G+M)+A+T + + ADD tmpb, tmpb, tmpb, LSL #2 ;// 5(C+R) + SSUB16 tmpa, tmpa, tmpb ;// 16+20(G+M)+(A+T)-5(C+R) + + USAT16 tmpb, #13, tmpa ;// saturate + LDR res, = 0x00FF00FF + UXTB16 tmpa, tmp4, ROR #8 ;// |g4|g2| + UXTAB16 tmpa, tmpa, tmp5, ROR #8 ;// |g4+m4|g2+m2| + AND res, res, tmpb, LSR #5 ;// mask and divide by 32 + + ADD tmpa, tmpa, tmpa, LSL #2 ;// 5(G+M) + UXTB16 tmpb, tmp3, ROR #8 ;// |c4|c2| + ADD tmpa, plus16, tmpa, LSL #2 ;// 16+20(G+M) + UXTAB16 tmpb, tmpb, tmp6, ROR #8 ;// |c4+r4|c2+r2| + UXTAB16 tmpa, tmpa, tmp2, ROR #8 ;// 16+20(G+M)+A + UXTAB16 tmpa, tmpa, tmp1, ROR #8 ;// 16+20(G+M)+A+T + + ADD tmpb, tmpb, tmpb, LSL #2 ;// 5(C+R) + SSUB16 tmpa, tmpa, tmpb ;// 16+20(G+M)+(A+T)-5(C+R) + + USAT16 tmpb, #13, tmpa ;// saturate + LDR tmp2, [mb] + LDR tmpa, = 0xFF00FF00 + MVN tmp2, tmp2 + + AND tmpa, tmpa, tmpb, LSL #3 ;// mask and divide by 32 + ORR res, res, tmpa + LDR tmpa, = 0x80808080 + UHSUB8 res, res, tmp2 ;// bilinear interpolation + LDR tmp2, [ref], width ;// load next row + EOR res, res, tmpa ;// correct sign + STR res, [mb], #16 ;// next row + + ;// tmp3 = |a4|a3|a2|a1| + ;// tmp4 = |c4|c3|c2|c1| + ;// tmp5 = |g4|g3|g2|g1| + ;// tmp6 = |m4|m3|m2|m1| + ;// tmp1 = |r4|r3|r2|r1| + ;// tmp2 = |t4|t3|t2|t1| + + ;// third four pixels + UXTB16 tmpa, tmp5 ;// |g3|g1| + UXTAB16 tmpa, tmpa, tmp6 ;// |g3+m3|g1+m1| + UXTB16 tmpb, tmp4 ;// |c3|c1| + ADD tmpa, tmpa, tmpa, LSL #2 ;// 5(G+M) + UXTAB16 tmpb, tmpb, tmp1 ;// |c3+r3|c1+r1| + ADD tmpa, plus16, tmpa, LSL #2 ;// 16+20(G+M) + UXTAB16 tmpa, tmpa, tmp3 ;// 16+20(G+M)+A + UXTAB16 tmpa, tmpa, tmp2 ;// 16+20(G+M)+A+T + + ADD tmpb, tmpb, tmpb, LSL #2 ;// 5(C+R) + SSUB16 tmpa, tmpa, tmpb ;// 16+20(G+M)+(A+T)-5(C+R) + + USAT16 tmpb, #13, tmpa ;// saturate + LDR res, = 0x00FF00FF + UXTB16 tmpa, tmp5, ROR #8 ;// |g4|g2| + UXTAB16 tmpa, tmpa, tmp6, ROR #8 ;// |g4+m4|g2+m2| + AND res, res, tmpb, LSR #5 ;// mask and divide by 32 + + ADD tmpa, tmpa, tmpa, LSL #2 ;// 5(G+M) + UXTB16 tmpb, tmp4, ROR #8 ;// |c4|c2| + ADD tmpa, plus16, tmpa, LSL #2 ;// 16+20(G+M) + UXTAB16 tmpb, tmpb, tmp1, ROR #8 ;// |c4+r4|c2+r2| + UXTAB16 tmpa, tmpa, tmp3, ROR #8 ;// 16+20(G+M)+A + UXTAB16 tmpa, tmpa, tmp2, ROR #8 ;// 16+20(G+M)+A+T + + + ADD tmpb, tmpb, tmpb, LSL #2 ;// 5(C+R) + SSUB16 tmpa, tmpa, tmpb ;// 16+20(G+M)+(A+T)-5(C+R) + + USAT16 tmpb, #13, tmpa ;// saturate + LDR tmp3, [mb] + LDR tmpa, = 0xFF00FF00 + MVN tmp3, tmp3 + + AND tmpa, tmpa, tmpb, LSL #3 ;// mask and divide by 32 + ORR res, res, tmpa + LDR tmpa, = 0x80808080 + UHSUB8 res, res, tmp3 ;// bilinear interpolation + LDR tmp3, [ref] ;// load next row + EOR res, res, tmpa ;// correct sign + STR res, [mb], #16 ;// next row + + ;// tmp4 = |a4|a3|a2|a1| + ;// tmp5 = |c4|c3|c2|c1| + ;// tmp6 = |g4|g3|g2|g1| + ;// tmp1 = |m4|m3|m2|m1| + ;// tmp2 = |r4|r3|r2|r1| + ;// tmp3 = |t4|t3|t2|t1| + + ;// fourth four pixels + UXTB16 tmpa, tmp6 ;// |g3|g1| + UXTAB16 tmpa, tmpa, tmp1 ;// |g3+m3|g1+m1| + UXTB16 tmpb, tmp5 ;// |c3|c1| + ADD tmpa, tmpa, tmpa, LSL #2 ;// 5(G+M) + UXTAB16 tmpb, tmpb, tmp2 ;// |c3+r3|c1+r1| + ADD tmpa, plus16, tmpa, LSL #2 ;// 16+20(G+M) + UXTAB16 tmpa, tmpa, tmp4 ;// 16+20(G+M)+A + UXTAB16 tmpa, tmpa, tmp3 ;// 16+20(G+M)+A+T + + ADD tmpb, tmpb, tmpb, LSL #2 ;// 5(C+R) + SSUB16 tmpa, tmpa, tmpb ;// 16+20(G+M)+(A+T)-5(C+R) + + USAT16 tmpb, #13, tmpa ;// saturate + LDR res, = 0x00FF00FF + UXTB16 tmpa, tmp6, ROR #8 ;// |g4|g2| + UXTAB16 tmpa, tmpa, tmp1, ROR #8 ;// |g4+m4|g2+m2| + AND res, res, tmpb, LSR #5 ;// mask and divide by 32 + + ADD tmpa, tmpa, tmpa, LSL #2 ;// 5(G+M) + UXTB16 tmpb, tmp5, ROR #8 ;// |c4|c2| + ADD tmpa, plus16, tmpa, LSL #2 ;// 16+20(G+M) + UXTAB16 tmpb, tmpb, tmp2, ROR #8 ;// |c4+r4|c2+r2| + UXTAB16 tmpa, tmpa, tmp4, ROR #8 ;// 16+20(G+M)+A + UXTAB16 tmpa, tmpa, tmp3, ROR #8 ;// 16+20(G+M)+A+T + + ADD tmpb, tmpb, tmpb, LSL #2 ;// 5(C+R) + SSUB16 tmpa, tmpa, tmpb ;// 16+20(G+M)+(A+T)-5(C+R) + + USAT16 tmpb, #13, tmpa ;// saturate + LDR tmp5, [mb] + LDR tmp4, = 0xFF00FF00 + MVN tmp5, tmp5 + + AND tmpa, tmp4, tmpb, LSL #3 ;// mask and divide by 32 + ORR res, res, tmpa + LDR tmpa, = 0x80808080 + UHSUB8 res, res, tmp5 ;// bilinear interpolation + + ;// decrement loop_x counter + SUBS count, count, #4<<28 ;// decrement x loop counter + + ;// calculate "ref" address for next round + SUB ref, ref, width, LSL #3 ;// ref -= 8*width; + ADD ref, ref, #4 ;// next column (4 pixels) + + EOR res, res, tmpa ;// correct sign + STR res, [mb], #-44 + + BCS loop_x + + ADDS mb, mb, #64 ;// set Carry=0 + ADD ref, ref, width, LSL #2 ;// ref += 4*width + AND tmp6, count, #0x00F00000 ;// partWidth-1 + SBC ref, ref, tmp6, LSR #20 ;// -(partWidth-1)-1 + SBC mb, mb, tmp6, LSR #20 ;// -(partWidth-1)-1 + + ADDS count, count, #0xC << 24 ;// decrement y loop counter + BGE loop_y + + ADD sp, sp, #0x1f4 + LDMFD sp!, {r4-r11, pc} + + END diff --git a/media/libstagefright/codecs/on2/h264dec/source/arm11_asm/h264bsd_interpolate_mid_hor.s b/media/libstagefright/codecs/on2/h264dec/source/arm11_asm/h264bsd_interpolate_mid_hor.s new file mode 100755 index 0000000..a81aed7 --- /dev/null +++ b/media/libstagefright/codecs/on2/h264dec/source/arm11_asm/h264bsd_interpolate_mid_hor.s @@ -0,0 +1,163 @@ +; Copyright (C) 2009 The Android Open Source Project
+;
+; Licensed under the Apache License, Version 2.0 (the "License");
+; you may not use this file except in compliance with the License.
+; You may obtain a copy of the License at
+;
+; http://www.apache.org/licenses/LICENSE-2.0
+;
+; Unless required by applicable law or agreed to in writing, software
+; distributed under the License is distributed on an "AS IS" BASIS,
+; WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+; See the License for the specific language governing permissions and
+; limitations under the License.
+
+;-------------------------------------------------------------------------------
+;--
+;-- Abstract : ARMv6 optimized version horizontal part of
+;-- h264bsdInterpolateMid functions
+;--
+;-------------------------------------------------------------------------------
+
+
+ IF :DEF: H264DEC_WINASM
+ ;// We dont use REQUIRE8 and PRESERVE8 for winasm
+ ELSE
+ REQUIRE8
+ PRESERVE8
+ ENDIF
+
+ AREA |.text|, CODE
+
+
+;// Register allocation
+
+ref RN 0 ;// pointer to current position in reference image
+mb RN 1 ;// pointer to current position in interpolated mb
+count RN 2 ;// bit-packed width and count values
+
+x_2_0 RN 4
+x_3_1 RN 5
+x_6_4 RN 6
+x_7_5 RN 7
+
+tmp1 RN 8
+tmp2 RN 9
+tmp3 RN 10
+tmp4 RN 11
+
+mult_20_01 RN 12 ;// [20, 1]
+mult_20_m5 RN 14 ;// [20, -5]
+
+
+ EXPORT h264bsdInterpolateMidHorPart
+
+;// Horizontal filter approach
+;//
+;// Basic idea in horizontal filtering is to adjust coefficients
+;// like below. Calculation is done with 16-bit maths.
+;//
+;// Reg x_2_0 x_3_1 x_6_4 x_7_5 x_2_0
+;// [ 2 0 ] [ 3 1 ] [ 6 4 ] [ 7 5 ] [ 10 8 ] ...
+;// y_0 = 20 1 20 -5 -5 1
+;// y_1 = -5 20 1 1 20 -5
+;// y_2 = 1 -5 -5 20 1 20
+;// y_3 = 1 20 -5 -5 20 1
+
+
+h264bsdInterpolateMidHorPart
+ STMFD sp!, {r4-r11, lr}
+
+ ;// pack values to count register
+ ;// [31:28] loop_x (partWidth-1)
+ ;// [27:24] loop_y (partHeight-1)
+ ;// [23:20] partWidth-1
+ ;// [19:16] partHeight-1
+ ;// [15:00] width
+
+
+ LDR mult_20_01, = 0x00140001
+ LDR mult_20_m5, = 0x0014FFFB
+ AND tmp3, count, #0x000F0000 ;// partWidth-1
+loop_y
+ LDR x_3_1, [ref, #-8]
+ ADD count, count, tmp3, LSL #12
+ LDR x_7_5, [ref, #-4]
+ UXTB16 x_2_0, x_3_1
+ UXTB16 x_3_1, x_3_1, ROR #8
+ UXTB16 x_6_4, x_7_5
+
+loop_x
+ UXTB16 x_7_5, x_7_5, ROR #8
+
+ SMUAD tmp1, x_2_0, mult_20_01
+ SMULTB tmp2, x_2_0, mult_20_m5
+ SMULTB tmp3, x_2_0, mult_20_01
+ SMULTB tmp4, x_3_1, mult_20_01
+
+ SMLAD tmp1, x_3_1, mult_20_m5, tmp1
+ SMLAD tmp2, x_3_1, mult_20_01, tmp2
+ SMLATB tmp3, x_3_1, mult_20_m5, tmp3
+ LDR x_3_1, [ref], #4
+ SMLAD tmp4, x_6_4, mult_20_m5, tmp4
+
+ SMLABB tmp1, x_6_4, mult_20_m5, tmp1
+ SMLADX tmp2, x_6_4, mult_20_01, tmp2
+ SMLADX tmp3, x_6_4, mult_20_m5, tmp3
+ SMLADX tmp4, x_7_5, mult_20_m5, tmp4
+
+ SMLABB tmp1, x_7_5, mult_20_01, tmp1
+ SMLABB tmp2, x_7_5, mult_20_m5, tmp2
+ UXTB16 x_2_0, x_3_1
+ SMLADX tmp3, x_7_5, mult_20_01, tmp3
+ SMLABB tmp4, x_2_0, mult_20_01, tmp4
+
+ SUBS count, count, #4<<28
+ STR tmp1, [mb], #4
+ STR tmp2, [mb], #4
+ STR tmp3, [mb], #4
+ STR tmp4, [mb], #4
+ BCC next_y
+
+ UXTB16 x_3_1, x_3_1, ROR #8
+
+ SMUAD tmp1, x_6_4, mult_20_01
+ SMULTB tmp2, x_6_4, mult_20_m5
+ SMULTB tmp3, x_6_4, mult_20_01
+ SMULTB tmp4, x_7_5, mult_20_01
+
+ SMLAD tmp1, x_7_5, mult_20_m5, tmp1
+ SMLAD tmp2, x_7_5, mult_20_01, tmp2
+ SMLATB tmp3, x_7_5, mult_20_m5, tmp3
+ LDR x_7_5, [ref], #4
+ SMLAD tmp4, x_2_0, mult_20_m5, tmp4
+
+ SMLABB tmp1, x_2_0, mult_20_m5, tmp1
+ SMLADX tmp2, x_2_0, mult_20_01, tmp2
+ SMLADX tmp3, x_2_0, mult_20_m5, tmp3
+ SMLADX tmp4, x_3_1, mult_20_m5, tmp4
+
+ SMLABB tmp1, x_3_1, mult_20_01, tmp1
+ SMLABB tmp2, x_3_1, mult_20_m5, tmp2
+ UXTB16 x_6_4, x_7_5
+ SMLADX tmp3, x_3_1, mult_20_01, tmp3
+ SMLABB tmp4, x_6_4, mult_20_01, tmp4
+
+ SUBS count, count, #4<<28
+ STR tmp1, [mb], #4
+ STR tmp2, [mb], #4
+ STR tmp3, [mb], #4
+ STR tmp4, [mb], #4
+ BCS loop_x
+
+next_y
+ AND tmp3, count, #0x000F0000 ;// partWidth-1
+ SMLABB ref, count, mult_20_01, ref ;// +width
+ SBC ref, ref, tmp3, LSR #16 ;// -(partWidth-1)-1
+ ADDS count, count, #(1<<28)-(1<<20)
+ BGE loop_y
+
+ LDMFD sp!, {r4-r11, pc}
+
+ END
+
diff --git a/media/libstagefright/codecs/on2/h264dec/source/arm11_asm/h264bsd_interpolate_ver_half.s b/media/libstagefright/codecs/on2/h264dec/source/arm11_asm/h264bsd_interpolate_ver_half.s new file mode 100755 index 0000000..244fc6f --- /dev/null +++ b/media/libstagefright/codecs/on2/h264dec/source/arm11_asm/h264bsd_interpolate_ver_half.s @@ -0,0 +1,347 @@ +; Copyright (C) 2009 The Android Open Source Project +; +; Licensed under the Apache License, Version 2.0 (the "License"); +; you may not use this file except in compliance with the License. +; You may obtain a copy of the License at +; +; http://www.apache.org/licenses/LICENSE-2.0 +; +; Unless required by applicable law or agreed to in writing, software +; distributed under the License is distributed on an "AS IS" BASIS, +; WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +; See the License for the specific language governing permissions and +; limitations under the License. + +;------------------------------------------------------------------------------- +;-- +;-- Abstract : ARMv6 optimized version of h264bsdInterpolateVerHalf function +;-- +;------------------------------------------------------------------------------- + + + IF :DEF: H264DEC_WINASM + ;// We dont use REQUIRE8 and PRESERVE8 for winasm + ELSE + REQUIRE8 + PRESERVE8 + ENDIF + + AREA |.text|, CODE + +;// h264bsdInterpolateVerHalf register allocation + +ref RN 0 + +mb RN 1 +buff RN 1 + +count RN 2 +x0 RN 2 + +res RN 3 +y0 RN 3 + +tmp1 RN 4 + +tmp2 RN 5 +height RN 5 + +tmp3 RN 6 +partW RN 6 + +tmp4 RN 7 +partH RN 7 + +tmp5 RN 8 +tmp6 RN 9 + +tmpa RN 10 +tmpb RN 11 +width RN 12 + +plus16 RN 14 + + +;// function exports and imports + + IMPORT h264bsdFillBlock + + EXPORT h264bsdInterpolateVerHalf + +;// Approach to vertical interpolation +;// +;// Interpolation is done by using 32-bit loads and stores +;// and by using 16 bit arithmetic. 4x4 block is processed +;// in each round. +;// +;// |a_11|a_11|a_11|a_11|...|a_1n|a_1n|a_1n|a_1n| +;// |b_11|b_11|b_11|b_11|...|b_1n|b_1n|b_1n|b_1n| +;// |c_11|c_11|c_11|c_11|...|c_1n|c_1n|c_1n|c_1n| +;// |d_11|d_11|d_11|d_11|...|d_1n|d_1n|d_1n|d_1n| +;// .. +;// .. +;// |a_m1|a_m1|a_m1|a_m1|... +;// |b_m1|b_m1|b_m1|b_m1|... +;// |c_m1|c_m1|c_m1|c_m1|... +;// |d_m1|d_m1|d_m1|d_m1|... + +h264bsdInterpolateVerHalf + STMFD sp!, {r0-r11, lr} + SUB sp, sp, #0x1e4 + + CMP x0, #0 + BLT do_fill ;// (x0 < 0) + LDR partW, [sp,#0x220] ;// partWidth + ADD tmp5, x0, partW ;// (x0+partWidth) + LDR width, [sp,#0x218] ;// width + CMP tmp5, width + BHI do_fill ;// (x0+partW)>width + + CMP y0, #0 + BLT do_fill ;// (y0 < 0) + LDR partH, [sp,#0x224] ;// partHeight + ADD tmp6, y0, partH ;// (y0+partHeight) + ADD tmp6, tmp6, #5 ;// (y0+partH+5) + LDR height, [sp,#0x21c] ;// height + CMP tmp6, height + BLS skip_fill ;// no overfill needed + + +do_fill + LDR partH, [sp,#0x224] ;// partHeight + ADD tmp5, partH, #5 ;// r2 = partH + 5; + LDR height, [sp,#0x21c] ;// height + LDR partW, [sp,#0x220] ;// partWidth + STMIB sp, {height, partW} ;// sp+4 = height, sp+8 = partWidth + STR tmp5, [sp,#0xc] ;// sp+c partHeight+5 + STR partW, [sp,#0x10] ;// sp+10 = partWidth + LDR width, [sp,#0x218] ;// width + STR width, [sp,#0] ;// sp+0 = width + ADD buff, sp, #0x28 ;// buff = p1[21*21/4+1] + BL h264bsdFillBlock + + MOV x0, #0 + STR x0,[sp,#0x1ec] ;// x0 = 0 + STR x0,[sp,#0x1f0] ;// y0 = 0 + ADD ref,sp,#0x28 ;// ref = p1 + STR partW, [sp,#0x218] + + +skip_fill + LDR x0 ,[sp,#0x1ec] ;// x0 + LDR y0 ,[sp,#0x1f0] ;// y0 + LDR width, [sp,#0x218] ;// width + MLA tmp6, width, y0, x0 ;// y0*width+x0 + ADD ref, ref, tmp6 ;// ref += y0*width+x0 + LDR mb, [sp, #0x1e8] ;// mb + + ADD count, partW, partH, LSL #16 ;// |partH|partW| + LDR tmp5, = 0x00010001 + SSUB16 count, count, tmp5; ;// |partH-1|partW-1| + LDR plus16, = 0x00100010 + + AND tmp1, count, #0x000000FF ;// partWidth + + +loop_y + ADD count, count, tmp1, LSL #24 ;// partWidth-1 to top byte + +loop_x + LDR tmp1, [ref], width ;// |a4|a3|a2|a1| + LDR tmp2, [ref], width ;// |c4|c3|c2|c1| + LDR tmp3, [ref], width ;// |g4|g3|g2|g1| + LDR tmp4, [ref], width ;// |m4|m3|m2|m1| + LDR tmp5, [ref], width ;// |r4|r3|r2|r1| + LDR tmp6, [ref], width ;// |t4|t3|t2|t1| + + ;// first four pixels + UXTB16 tmpa, tmp3 ;// |g3|g1| + UXTAB16 tmpa, tmpa, tmp4 ;// |g3+m3|g1+m1| + UXTB16 tmpb, tmp2 ;// |c3|c1| + ADD tmpa, tmpa, tmpa, LSL #2 ;// 5(G+M) + + UXTAB16 tmpb, tmpb, tmp5 ;// |c3+r3|c1+r1| + ADD tmpa, plus16, tmpa, LSL #2 ;// 16+20(G+M) + UXTAB16 tmpa, tmpa, tmp1 ;// 16+20(G+M)+A + UXTAB16 tmpa, tmpa, tmp6 ;// 16+20(G+M)+A+T + + ADD tmpb, tmpb, tmpb, LSL #2 ;// 5(C+R) + SSUB16 tmpa, tmpa, tmpb ;// 16+20(G+M)+(A+T)-5(C+R) + + USAT16 tmpb, #13, tmpa ;// saturate + LDR res, = 0x00FF00FF + UXTB16 tmpa, tmp3, ROR #8 ;// |g4|g2| + UXTAB16 tmpa, tmpa, tmp4, ROR #8 ;// |g4+m4|g2+m2| + AND res, res, tmpb, LSR #5 ;// mask and divide by 32 + + ADD tmpa, tmpa, tmpa, LSL #2 ;// 5(G+M) + UXTB16 tmpb, tmp2, ROR #8 ;// |c4|c2| + ADD tmpa, plus16, tmpa, LSL #2 ;// 16+20(G+M) + UXTAB16 tmpb, tmpb, tmp5, ROR #8 ;// |c4+r4|c2+r2| + UXTAB16 tmpa, tmpa, tmp1, ROR #8 ;// 16+20(G+M)+A + UXTAB16 tmpa, tmpa, tmp6, ROR #8 ;// 16+20(G+M)+A+T + + ADD tmpb, tmpb, tmpb, LSL #2 ;// 5(C+R) + SSUB16 tmpa, tmpa, tmpb ;// 16+20(G+M)+(A+T)-5(C+R) + + USAT16 tmpb, #13, tmpa ;// saturate + LDR tmp1, [ref], width + LDR tmpa, = 0xFF00FF00 + + AND tmpa, tmpa, tmpb, LSL #3 ;// mask and divede by 32 + ORR res, res, tmpa + STR res, [mb], #16 ;// next row (mb) + + ;// tmp2 = |a4|a3|a2|a1| + ;// tmp3 = |c4|c3|c2|c1| + ;// tmp4 = |g4|g3|g2|g1| + ;// tmp5 = |m4|m3|m2|m1| + ;// tmp6 = |r4|r3|r2|r1| + ;// tmp1 = |t4|t3|t2|t1| + + ;// second four pixels + UXTB16 tmpa, tmp4 ;// |g3|g1| + UXTAB16 tmpa, tmpa, tmp5 ;// |g3+m3|g1+m1| + UXTB16 tmpb, tmp3 ;// |c3|c1| + ADD tmpa, tmpa, tmpa, LSL #2 ;// 5(G+M) + UXTAB16 tmpb, tmpb, tmp6 ;// |c3+r3|c1+r1| + ADD tmpa, plus16, tmpa, LSL #2 ;// 16+20(G+M) + UXTAB16 tmpa, tmpa, tmp2 ;// 16+20(G+M)+A + UXTAB16 tmpa, tmpa, tmp1 ;// 16+20(G+M)+A+T + + ADD tmpb, tmpb, tmpb, LSL #2 ;// 5(C+R) + SSUB16 tmpa, tmpa, tmpb ;// 16+20(G+M)+(A+T)-5(C+R) + + USAT16 tmpb, #13, tmpa ;// saturate + LDR res, = 0x00FF00FF + UXTB16 tmpa, tmp4, ROR #8 ;// |g4|g2| + UXTAB16 tmpa, tmpa, tmp5, ROR #8 ;// |g4+m4|g2+m2| + AND res, res, tmpb, LSR #5 ;// mask and divide by 32 + + ADD tmpa, tmpa, tmpa, LSL #2 ;// 5(G+M) + UXTB16 tmpb, tmp3, ROR #8 ;// |c4|c2| + ADD tmpa, plus16, tmpa, LSL #2 ;// 16+20(G+M) + UXTAB16 tmpb, tmpb, tmp6, ROR #8 ;// |c4+r4|c2+r2| + UXTAB16 tmpa, tmpa, tmp2, ROR #8 ;// 16+20(G+M)+A + UXTAB16 tmpa, tmpa, tmp1, ROR #8 ;// 16+20(G+M)+A+T + + ADD tmpb, tmpb, tmpb, LSL #2 ;// 5(C+R) + SSUB16 tmpa, tmpa, tmpb ;// 16+20(G+M)+(A+T)-5(C+R) + + USAT16 tmpb, #13, tmpa ;// saturate + LDR tmp2, [ref], width + LDR tmpa, = 0xFF00FF00 + + AND tmpa, tmpa, tmpb, LSL #3 ;// mask and divide by 32 + ORR res, res, tmpa + STR res, [mb], #16 ;// next row + + ;// tmp3 = |a4|a3|a2|a1| + ;// tmp4 = |c4|c3|c2|c1| + ;// tmp5 = |g4|g3|g2|g1| + ;// tmp6 = |m4|m3|m2|m1| + ;// tmp1 = |r4|r3|r2|r1| + ;// tmp2 = |t4|t3|t2|t1| + + ;// third four pixels + UXTB16 tmpa, tmp5 ;// |g3|g1| + UXTAB16 tmpa, tmpa, tmp6 ;// |g3+m3|g1+m1| + UXTB16 tmpb, tmp4 ;// |c3|c1| + ADD tmpa, tmpa, tmpa, LSL #2 ;// 5(G+M) + UXTAB16 tmpb, tmpb, tmp1 ;// |c3+r3|c1+r1| + ADD tmpa, plus16, tmpa, LSL #2 ;// 16+20(G+M) + UXTAB16 tmpa, tmpa, tmp3 ;// 16+20(G+M)+A + UXTAB16 tmpa, tmpa, tmp2 ;// 16+20(G+M)+A+T + + ADD tmpb, tmpb, tmpb, LSL #2 ;// 5(C+R) + SSUB16 tmpa, tmpa, tmpb ;// 16+20(G+M)+(A+T)-5(C+R) + + USAT16 tmpb, #13, tmpa ;// saturate + LDR res, = 0x00FF00FF + UXTB16 tmpa, tmp5, ROR #8 ;// |g4|g2| + UXTAB16 tmpa, tmpa, tmp6, ROR #8 ;// |g4+m4|g2+m2| + AND res, res, tmpb, LSR #5 ;// mask and divide by 32 + + ADD tmpa, tmpa, tmpa, LSL #2 ;// 5(G+M) + UXTB16 tmpb, tmp4, ROR #8 ;// |c4|c2| + ADD tmpa, plus16, tmpa, LSL #2 ;// 16+20(G+M) + UXTAB16 tmpb, tmpb, tmp1, ROR #8 ;// |c4+r4|c2+r2| + UXTAB16 tmpa, tmpa, tmp3, ROR #8 ;// 16+20(G+M)+A + UXTAB16 tmpa, tmpa, tmp2, ROR #8 ;// 16+20(G+M)+A+T + + + ADD tmpb, tmpb, tmpb, LSL #2 ;// 5(C+R) + SSUB16 tmpa, tmpa, tmpb ;// 16+20(G+M)+(A+T)-5(C+R) + + USAT16 tmpb, #13, tmpa ;// saturate + LDR tmp3, [ref] + LDR tmpa, = 0xFF00FF00 + + ;// decrement loop_x counter + SUBS count, count, #4<<24 ;// (partWidth-1) -= 4; + + AND tmpa, tmpa, tmpb, LSL #3 ;// mask and divide by 32 + ORR res, res, tmpa + STR res, [mb], #16 ;// next row + + ;// tmp4 = |a4|a3|a2|a1| + ;// tmp5 = |c4|c3|c2|c1| + ;// tmp6 = |g4|g3|g2|g1| + ;// tmp1 = |m4|m3|m2|m1| + ;// tmp2 = |r4|r3|r2|r1| + ;// tmp3 = |t4|t3|t2|t1| + + ;// fourth four pixels + UXTB16 tmpa, tmp6 ;// |g3|g1| + UXTAB16 tmpa, tmpa, tmp1 ;// |g3+m3|g1+m1| + UXTB16 tmpb, tmp5 ;// |c3|c1| + ADD tmpa, tmpa, tmpa, LSL #2 ;// 5(G+M) + UXTAB16 tmpb, tmpb, tmp2 ;// |c3+r3|c1+r1| + ADD tmpa, plus16, tmpa, LSL #2 ;// 16+20(G+M) + UXTAB16 tmpa, tmpa, tmp4 ;// 16+20(G+M)+A + UXTAB16 tmpa, tmpa, tmp3 ;// 16+20(G+M)+A+T + + ADD tmpb, tmpb, tmpb, LSL #2 ;// 5(C+R) + SSUB16 tmpa, tmpa, tmpb ;// 16+20(G+M)+(A+T)-5(C+R) + + USAT16 tmpb, #13, tmpa ;// saturate + LDR res, = 0x00FF00FF + UXTB16 tmpa, tmp6, ROR #8 ;// |g4|g2| + UXTAB16 tmpa, tmpa, tmp1, ROR #8 ;// |g4+m4|g2+m2| + AND res, res, tmpb, LSR #5 ;// mask and divide by 32 + + ADD tmpa, tmpa, tmpa, LSL #2 ;// 5(G+M) + UXTB16 tmpb, tmp5, ROR #8 ;// |c4|c2| + ADD tmpa, plus16, tmpa, LSL #2 ;// 16+20(G+M) + UXTAB16 tmpb, tmpb, tmp2, ROR #8 ;// |c4+r4|c2+r2| + UXTAB16 tmpa, tmpa, tmp4, ROR #8 ;// 16+20(G+M)+A + UXTAB16 tmpa, tmpa, tmp3, ROR #8 ;// 16+20(G+M)+A+T + + ADD tmpb, tmpb, tmpb, LSL #2 ;// 5(C+R) + SSUB16 tmpa, tmpa, tmpb ;// 16+20(G+M)+(A+T)-5(C+R) + + USAT16 tmpb, #13, tmpa ;// saturate + LDR tmp4, = 0xFF00FF00 + + ;// calculate "ref" address for next round + SUB ref, ref, width, LSL #3 ;// ref -= 8*width; + ADD ref, ref, #4; ;// next column (4 pixels) + AND tmpa, tmp4, tmpb, LSL #3 ;// mask and divide by 32 + ORR res, res, tmpa + STR res, [mb], #-44 + + BCS loop_x + + ADDS count, count, #252<<16 ;// (partHeight-1) -= 4; + ADD ref, ref, width, LSL #2 ;// ref += 4*width + AND tmp1, count, #0x000000FF ;// partWidth-1 + ADD tmp2, tmp1, #1 ;// partWidth + SUB ref, ref, tmp2 ;// ref -= partWidth + ADD mb, mb, #64; + SUB mb, mb, tmp2; ;// mb -= partWidth + BGE loop_y + + ADD sp,sp,#0x1f4 + LDMFD sp!, {r4-r11, pc} + + END diff --git a/media/libstagefright/codecs/on2/h264dec/source/arm11_asm/h264bsd_interpolate_ver_quarter.s b/media/libstagefright/codecs/on2/h264dec/source/arm11_asm/h264bsd_interpolate_ver_quarter.s new file mode 100755 index 0000000..5266c85 --- /dev/null +++ b/media/libstagefright/codecs/on2/h264dec/source/arm11_asm/h264bsd_interpolate_ver_quarter.s @@ -0,0 +1,374 @@ +; Copyright (C) 2009 The Android Open Source Project +; +; Licensed under the Apache License, Version 2.0 (the "License"); +; you may not use this file except in compliance with the License. +; You may obtain a copy of the License at +; +; http://www.apache.org/licenses/LICENSE-2.0 +; +; Unless required by applicable law or agreed to in writing, software +; distributed under the License is distributed on an "AS IS" BASIS, +; WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +; See the License for the specific language governing permissions and +; limitations under the License. + +;------------------------------------------------------------------------------- +;-- +;-- Abstract : ARMv6 optimized version of h264bsdInterpolateVerQuarter function +;-- +;------------------------------------------------------------------------------- + + IF :DEF: H264DEC_WINASM + ;// We dont use REQUIRE8 and PRESERVE8 for winasm + ELSE + REQUIRE8 + PRESERVE8 + ENDIF + + AREA |.text|, CODE + +;// h264bsdInterpolateVerQuarter register allocation + +ref RN 0 + +mb RN 1 +buff RN 1 + +count RN 2 +x0 RN 2 + +res RN 3 +y0 RN 3 + +tmp1 RN 4 + +tmp2 RN 5 +height RN 5 + +tmp3 RN 6 +partW RN 6 + +tmp4 RN 7 +partH RN 7 + +tmp5 RN 8 +tmp6 RN 9 + +tmpa RN 10 +tmpb RN 11 +width RN 12 + +plus16 RN 14 + + +;// function exports and imports + + IMPORT h264bsdFillBlock + + EXPORT h264bsdInterpolateVerQuarter + +;// Approach to vertical interpolation +;// +;// Interpolation is done by using 32-bit loads and stores +;// and by using 16 bit arithmetic. 4x4 block is processed +;// in each round. +;// +;// |a_11|a_11|a_11|a_11|...|a_1n|a_1n|a_1n|a_1n| +;// |b_11|b_11|b_11|b_11|...|b_1n|b_1n|b_1n|b_1n| +;// |c_11|c_11|c_11|c_11|...|c_1n|c_1n|c_1n|c_1n| +;// |d_11|d_11|d_11|d_11|...|d_1n|d_1n|d_1n|d_1n| +;// .. +;// .. +;// |a_m1|a_m1|a_m1|a_m1|... +;// |b_m1|b_m1|b_m1|b_m1|... +;// |c_m1|c_m1|c_m1|c_m1|... +;// |d_m1|d_m1|d_m1|d_m1|... + +h264bsdInterpolateVerQuarter + STMFD sp!, {r0-r11, lr} + SUB sp, sp, #0x1e4 + + CMP x0, #0 + BLT do_fill ;// (x0 < 0) + LDR partW, [sp,#0x220] ;// partWidth + ADD tmp5, x0, partW ;// (x0+partWidth) + LDR width, [sp,#0x218] ;// width + CMP tmp5, width + BHI do_fill ;// (x0+partW)>width + + CMP y0, #0 + BLT do_fill ;// (y0 < 0) + LDR partH, [sp,#0x224] ;// partHeight + ADD tmp6, y0, partH ;// (y0+partHeight) + ADD tmp6, tmp6, #5 ;// (y0+partH+5) + LDR height, [sp,#0x21c] ;// height + CMP tmp6, height + BLS skip_fill ;// no overfill needed + + +do_fill + LDR partH, [sp,#0x224] ;// partHeight + ADD tmp5, partH, #5 ;// r2 = partH + 5; + LDR height, [sp,#0x21c] ;// height + LDR partW, [sp,#0x220] ;// partWidth + STMIB sp, {height, partW} ;// sp+4 = height, sp+8 = partWidth + STR tmp5, [sp,#0xc] ;// sp+c partHeight+5 + STR partW, [sp,#0x10] ;// sp+10 = partWidth + LDR width, [sp,#0x218] ;// width + STR width, [sp,#0] ;// sp+0 = width + ADD buff, sp, #0x28 ;// buff = p1[21*21/4+1] + BL h264bsdFillBlock + + MOV x0, #0 + STR x0,[sp,#0x1ec] ;// x0 = 0 + STR x0,[sp,#0x1f0] ;// y0 = 0 + ADD ref,sp,#0x28 ;// ref = p1 + STR partW, [sp,#0x218] + + +skip_fill + LDR x0 ,[sp,#0x1ec] ;// x0 + LDR y0 ,[sp,#0x1f0] ;// y0 + LDR width, [sp,#0x218] ;// width + MLA tmp6, width, y0, x0 ;// y0*width+x0 + ADD ref, ref, tmp6 ;// ref += y0*width+x0 + LDR mb, [sp, #0x1e8] ;// mb + + ADD count, partW, partH, LSL #8 ;// |xx|xx|partH|partW| + LDR tmp5, = 0x00010100 + RSB count, tmp5, count, LSL #8 ;// |xx|partH-1|partW-1|xx| + LDR tmp2, [sp, #0x228] ;// verOffset + ADD count, count, tmp2 ;// |xx|partH-1|partW-1|verOffset| + LDR plus16, = 0x00100010 + + AND tmp1, count, #0x0000FF00 ;// partWidth + + +loop_y + ADD count, count, tmp1, LSL #16 ;// partWidth-1 to top byte + +loop_x + LDR tmp1, [ref], width ;// |a4|a3|a2|a1| + LDR tmp2, [ref], width ;// |c4|c3|c2|c1| + LDR tmp3, [ref], width ;// |g4|g3|g2|g1| + LDR tmp4, [ref], width ;// |m4|m3|m2|m1| + LDR tmp5, [ref], width ;// |r4|r3|r2|r1| + LDR tmp6, [ref], width ;// |t4|t3|t2|t1| + + ;// first four pixels + UXTB16 tmpa, tmp3 ;// |g3|g1| + UXTAB16 tmpa, tmpa, tmp4 ;// |g3+m3|g1+m1| + UXTB16 tmpb, tmp2 ;// |c3|c1| + ADD tmpa, tmpa, tmpa, LSL #2 ;// 5(G+M) + + UXTAB16 tmpb, tmpb, tmp5 ;// |c3+r3|c1+r1| + ADD tmpa, plus16, tmpa, LSL #2 ;// 16+20(G+M) + UXTAB16 tmpa, tmpa, tmp1 ;// 16+20(G+M)+A + UXTAB16 tmpa, tmpa, tmp6 ;// 16+20(G+M)+A+T + + ADD tmpb, tmpb, tmpb, LSL #2 ;// 5(C+R) + SSUB16 tmpa, tmpa, tmpb ;// 16+20(G+M)+(A+T)-5(C+R) + + USAT16 tmpb, #13, tmpa ;// saturate + LDR res, = 0x00FF00FF + UXTB16 tmpa, tmp3, ROR #8 ;// |g4|g2| + UXTAB16 tmpa, tmpa, tmp4, ROR #8 ;// |g4+m4|g2+m2| + AND res, res, tmpb, LSR #5 ;// mask and divide by 32 + + ADD tmpa, tmpa, tmpa, LSL #2 ;// 5(G+M) + UXTB16 tmpb, tmp2, ROR #8 ;// |c4|c2| + ADD tmpa, plus16, tmpa, LSL #2 ;// 16+20(G+M) + UXTAB16 tmpb, tmpb, tmp5, ROR #8 ;// |c4+r4|c2+r2| + UXTAB16 tmpa, tmpa, tmp1, ROR #8 ;// 16+20(G+M)+A + UXTAB16 tmpa, tmpa, tmp6, ROR #8 ;// 16+20(G+M)+A+T + + ADD tmpb, tmpb, tmpb, LSL #2 ;// 5(C+R) + SSUB16 tmpa, tmpa, tmpb ;// 16+20(G+M)+(A+T)-5(C+R) + + USAT16 tmpb, #13, tmpa ;// saturate + MOVS tmp1, count, LSL #31 ;// update flags (verOffset) + LDR tmpa, = 0xFF00FF00 + MVNEQ tmp1, tmp3 ;// select verOffset=0 + MVNNE tmp1, tmp4 ;// select verOffset=1 + AND tmpa, tmpa, tmpb, LSL #3 ;// mask and divede by 32 + ORR res, res, tmpa + + LDR tmpa, = 0x80808080 + UHSUB8 res, res, tmp1 ;// bilinear interpolation + LDR tmp1, [ref], width ;// load next row + EOR res, res, tmpa ;// correct sign + + STR res, [mb], #16 ;// next row (mb) + + + ;// tmp2 = |a4|a3|a2|a1| + ;// tmp3 = |c4|c3|c2|c1| + ;// tmp4 = |g4|g3|g2|g1| + ;// tmp5 = |m4|m3|m2|m1| + ;// tmp6 = |r4|r3|r2|r1| + ;// tmp1 = |t4|t3|t2|t1| + + ;// second four pixels + UXTB16 tmpa, tmp4 ;// |g3|g1| + UXTAB16 tmpa, tmpa, tmp5 ;// |g3+m3|g1+m1| + UXTB16 tmpb, tmp3 ;// |c3|c1| + ADD tmpa, tmpa, tmpa, LSL #2 ;// 5(G+M) + UXTAB16 tmpb, tmpb, tmp6 ;// |c3+r3|c1+r1| + ADD tmpa, plus16, tmpa, LSL #2 ;// 16+20(G+M) + UXTAB16 tmpa, tmpa, tmp2 ;// 16+20(G+M)+A + UXTAB16 tmpa, tmpa, tmp1 ;// 16+20(G+M)+A+T + + ADD tmpb, tmpb, tmpb, LSL #2 ;// 5(C+R) + SSUB16 tmpa, tmpa, tmpb ;// 16+20(G+M)+(A+T)-5(C+R) + + USAT16 tmpb, #13, tmpa ;// saturate + LDR res, = 0x00FF00FF + UXTB16 tmpa, tmp4, ROR #8 ;// |g4|g2| + UXTAB16 tmpa, tmpa, tmp5, ROR #8 ;// |g4+m4|g2+m2| + AND res, res, tmpb, LSR #5 ;// mask and divide by 32 + + ADD tmpa, tmpa, tmpa, LSL #2 ;// 5(G+M) + UXTB16 tmpb, tmp3, ROR #8 ;// |c4|c2| + ADD tmpa, plus16, tmpa, LSL #2 ;// 16+20(G+M) + UXTAB16 tmpb, tmpb, tmp6, ROR #8 ;// |c4+r4|c2+r2| + UXTAB16 tmpa, tmpa, tmp2, ROR #8 ;// 16+20(G+M)+A + UXTAB16 tmpa, tmpa, tmp1, ROR #8 ;// 16+20(G+M)+A+T + + ADD tmpb, tmpb, tmpb, LSL #2 ;// 5(C+R) + SSUB16 tmpa, tmpa, tmpb ;// 16+20(G+M)+(A+T)-5(C+R) + + USAT16 tmpb, #13, tmpa ;// saturate + LDR tmpa, = 0xFF00FF00 + MVNEQ tmp2, tmp4 ;// select verOffset=0 + MVNNE tmp2, tmp5 ;// select verOffset=1 + + AND tmpa, tmpa, tmpb, LSL #3 ;// mask and divide by 32 + ORR res, res, tmpa + LDR tmpa, = 0x80808080 + UHSUB8 res, res, tmp2 ;// bilinear interpolation + LDR tmp2, [ref], width ;// load next row + EOR res, res, tmpa ;// correct sign + STR res, [mb], #16 ;// next row + + ;// tmp3 = |a4|a3|a2|a1| + ;// tmp4 = |c4|c3|c2|c1| + ;// tmp5 = |g4|g3|g2|g1| + ;// tmp6 = |m4|m3|m2|m1| + ;// tmp1 = |r4|r3|r2|r1| + ;// tmp2 = |t4|t3|t2|t1| + + ;// third four pixels + UXTB16 tmpa, tmp5 ;// |g3|g1| + UXTAB16 tmpa, tmpa, tmp6 ;// |g3+m3|g1+m1| + UXTB16 tmpb, tmp4 ;// |c3|c1| + ADD tmpa, tmpa, tmpa, LSL #2 ;// 5(G+M) + UXTAB16 tmpb, tmpb, tmp1 ;// |c3+r3|c1+r1| + ADD tmpa, plus16, tmpa, LSL #2 ;// 16+20(G+M) + UXTAB16 tmpa, tmpa, tmp3 ;// 16+20(G+M)+A + UXTAB16 tmpa, tmpa, tmp2 ;// 16+20(G+M)+A+T + + ADD tmpb, tmpb, tmpb, LSL #2 ;// 5(C+R) + SSUB16 tmpa, tmpa, tmpb ;// 16+20(G+M)+(A+T)-5(C+R) + + USAT16 tmpb, #13, tmpa ;// saturate + LDR res, = 0x00FF00FF + UXTB16 tmpa, tmp5, ROR #8 ;// |g4|g2| + UXTAB16 tmpa, tmpa, tmp6, ROR #8 ;// |g4+m4|g2+m2| + AND res, res, tmpb, LSR #5 ;// mask and divide by 32 + + ADD tmpa, tmpa, tmpa, LSL #2 ;// 5(G+M) + UXTB16 tmpb, tmp4, ROR #8 ;// |c4|c2| + ADD tmpa, plus16, tmpa, LSL #2 ;// 16+20(G+M) + UXTAB16 tmpb, tmpb, tmp1, ROR #8 ;// |c4+r4|c2+r2| + UXTAB16 tmpa, tmpa, tmp3, ROR #8 ;// 16+20(G+M)+A + UXTAB16 tmpa, tmpa, tmp2, ROR #8 ;// 16+20(G+M)+A+T + + + ADD tmpb, tmpb, tmpb, LSL #2 ;// 5(C+R) + SSUB16 tmpa, tmpa, tmpb ;// 16+20(G+M)+(A+T)-5(C+R) + + USAT16 tmpb, #13, tmpa ;// saturate + LDR tmpa, = 0xFF00FF00 + MVNEQ tmp3, tmp5 ;// select verOffset=0 + MVNNE tmp3, tmp6 ;// select verOffset=1 + + AND tmpa, tmpa, tmpb, LSL #3 ;// mask and divide by 32 + ORR res, res, tmpa + LDR tmpa, = 0x80808080 + UHSUB8 res, res, tmp3 ;// bilinear interpolation + LDR tmp3, [ref] ;// load next row + EOR res, res, tmpa ;// correct sign + STR res, [mb], #16 ;// next row + + ;// tmp4 = |a4|a3|a2|a1| + ;// tmp5 = |c4|c3|c2|c1| + ;// tmp6 = |g4|g3|g2|g1| + ;// tmp1 = |m4|m3|m2|m1| + ;// tmp2 = |r4|r3|r2|r1| + ;// tmp3 = |t4|t3|t2|t1| + + ;// fourth four pixels + UXTB16 tmpa, tmp6 ;// |g3|g1| + UXTAB16 tmpa, tmpa, tmp1 ;// |g3+m3|g1+m1| + UXTB16 tmpb, tmp5 ;// |c3|c1| + ADD tmpa, tmpa, tmpa, LSL #2 ;// 5(G+M) + UXTAB16 tmpb, tmpb, tmp2 ;// |c3+r3|c1+r1| + ADD tmpa, plus16, tmpa, LSL #2 ;// 16+20(G+M) + UXTAB16 tmpa, tmpa, tmp4 ;// 16+20(G+M)+A + UXTAB16 tmpa, tmpa, tmp3 ;// 16+20(G+M)+A+T + + ADD tmpb, tmpb, tmpb, LSL #2 ;// 5(C+R) + SSUB16 tmpa, tmpa, tmpb ;// 16+20(G+M)+(A+T)-5(C+R) + + USAT16 tmpb, #13, tmpa ;// saturate + LDR res, = 0x00FF00FF + UXTB16 tmpa, tmp6, ROR #8 ;// |g4|g2| + UXTAB16 tmpa, tmpa, tmp1, ROR #8 ;// |g4+m4|g2+m2| + AND res, res, tmpb, LSR #5 ;// mask and divide by 32 + + ADD tmpa, tmpa, tmpa, LSL #2 ;// 5(G+M) + UXTB16 tmpb, tmp5, ROR #8 ;// |c4|c2| + ADD tmpa, plus16, tmpa, LSL #2 ;// 16+20(G+M) + UXTAB16 tmpb, tmpb, tmp2, ROR #8 ;// |c4+r4|c2+r2| + UXTAB16 tmpa, tmpa, tmp4, ROR #8 ;// 16+20(G+M)+A + UXTAB16 tmpa, tmpa, tmp3, ROR #8 ;// 16+20(G+M)+A+T + + ADD tmpb, tmpb, tmpb, LSL #2 ;// 5(C+R) + SSUB16 tmpa, tmpa, tmpb ;// 16+20(G+M)+(A+T)-5(C+R) + + USAT16 tmpb, #13, tmpa ;// saturate + LDR tmp4, = 0xFF00FF00 + MVNEQ tmp5, tmp6 ;// select verOffset=0 + MVNNE tmp5, tmp1 ;// select verOffset=1 + + AND tmpa, tmp4, tmpb, LSL #3 ;// mask and divide by 32 + ORR res, res, tmpa + LDR tmpa, = 0x80808080 + UHSUB8 res, res, tmp5 ;// bilinear interpolation + + ;// decrement loop_x counter + SUBS count, count, #4<<24 ;// (partWidth-1) -= 4; + + ;// calculate "ref" address for next round + SUB ref, ref, width, LSL #3 ;// ref -= 8*width; + ADD ref, ref, #4; ;// next column (4 pixels) + + EOR res, res, tmpa ;// correct sign + STR res, [mb], #-44 + + BCS loop_x + + ADDS count, count, #252<<16 ;// (partHeight-1) -= 4; + ADD ref, ref, width, LSL #2 ;// ref += 4*width + AND tmp1, count, #0x0000FF00 ;// partWidth-1 + MOV tmp2, #1 + ADD tmp2, tmp2, tmp1, LSR #8 ;// partWidth + SUB ref, ref, tmp2 ;// ref -= partWidth + ADD mb, mb, #64; + SUB mb, mb, tmp2; ;// mb -= partWidth + BGE loop_y + + ADD sp,sp,#0x1f4 + LDMFD sp!, {r4-r11, pc} + + END diff --git a/media/libstagefright/codecs/on2/h264dec/source/arm11_asm/win_asm.bat b/media/libstagefright/codecs/on2/h264dec/source/arm11_asm/win_asm.bat new file mode 100644 index 0000000..1b8d88c --- /dev/null +++ b/media/libstagefright/codecs/on2/h264dec/source/arm11_asm/win_asm.bat @@ -0,0 +1,15 @@ +echo off +set ASMFLAGS= -checkreglist -CPU ARM1136 -PreDefine "H264DEC_WINASM SETL {TRUE}" +set ASM="D:\Program Files\Microsoft Visual Studio 8\VC\ce\bin\x86_arm\armasm" +echo on + +%ASM% %ASMFLAGS% h264bsd_interpolate_chroma_ver.s +%ASM% %ASMFLAGS% h264bsd_interpolate_chroma_hor.s +%ASM% %ASMFLAGS% h264bsd_interpolate_hor_half.s +%ASM% %ASMFLAGS% h264bsd_interpolate_hor_quarter.s +%ASM% %ASMFLAGS% h264bsd_interpolate_hor_ver_quarter.s +%ASM% %ASMFLAGS% h264bsd_interpolate_ver_half.s +%ASM% %ASMFLAGS% h264bsd_interpolate_ver_quarter.s + +rem %ASM% %ASMFLAGS% h264bsd_interpolate_chroma_hor_ver.s +rem %ASM% %ASMFLAGS% h264bsd_interpolate_mid_hor.s |