diff options
| author | James Dong <jdong@google.com> | 2011-05-31 18:53:46 -0700 |
|---|---|---|
| committer | James Dong <jdong@google.com> | 2011-06-02 12:32:46 -0700 |
| commit | 0c1bc742181ded4930842b46e9507372f0b1b963 (patch) | |
| tree | c952bfcb03ff7cce5e0f91ad7d25c67a2fdd39cb /media/libstagefright/codecs/on2/h264dec/omxdl/arm_neon/vc/m4p10/src_gcc/omxVCM4P10_TransformDequantLumaDCFromPair_s.S | |
| parent | 92a746c3b18d035189f596ce32847bf26247aaca (diff) | |
| download | frameworks_av-0c1bc742181ded4930842b46e9507372f0b1b963.zip frameworks_av-0c1bc742181ded4930842b46e9507372f0b1b963.tar.gz frameworks_av-0c1bc742181ded4930842b46e9507372f0b1b963.tar.bz2 | |
Initial-checkin for ON2 Software AVC/H264 decoder
o when neon is present, the performance gain of On2 AVC software decoder
over PV software decoder is more than 30%.
o In addition, it fixes some known PV software decoder issues like missing
output frames
o allow both pv and on2 software avc to be available for easy comparision
o change output frames from 8 to 16
Change-Id: I567ad1842025ead7092f0c47e3513d6d9ca232dd
Diffstat (limited to 'media/libstagefright/codecs/on2/h264dec/omxdl/arm_neon/vc/m4p10/src_gcc/omxVCM4P10_TransformDequantLumaDCFromPair_s.S')
| -rw-r--r-- | media/libstagefright/codecs/on2/h264dec/omxdl/arm_neon/vc/m4p10/src_gcc/omxVCM4P10_TransformDequantLumaDCFromPair_s.S | 76 |
1 files changed, 76 insertions, 0 deletions
diff --git a/media/libstagefright/codecs/on2/h264dec/omxdl/arm_neon/vc/m4p10/src_gcc/omxVCM4P10_TransformDequantLumaDCFromPair_s.S b/media/libstagefright/codecs/on2/h264dec/omxdl/arm_neon/vc/m4p10/src_gcc/omxVCM4P10_TransformDequantLumaDCFromPair_s.S new file mode 100644 index 0000000..a3a0715 --- /dev/null +++ b/media/libstagefright/codecs/on2/h264dec/omxdl/arm_neon/vc/m4p10/src_gcc/omxVCM4P10_TransformDequantLumaDCFromPair_s.S @@ -0,0 +1,76 @@ +/* + * (c) Copyright 2007-2008 ARM Limited. All Rights Reserved. + * + */ + + .eabi_attribute 24, 1 + .eabi_attribute 25, 1 + + .arm + .fpu neon + .text + + .global armVCM4P10_InvTransformDequantLumaDC4x4 + .func armVCM4P10_InvTransformDequantLumaDC4x4 +armVCM4P10_InvTransformDequantLumaDC4x4: + PUSH {r4-r6,lr} + VPUSH {d8-d13} + VLD4.16 {d0,d1,d2,d3},[r0] + LDR r2, =armVCM4P10_QPDivTable + LDR r3, =armVCM4P10_VMatrixQPModTable + VADD.I16 d4,d0,d1 + VADD.I16 d5,d2,d3 + VSUB.I16 d6,d0,d1 + LDRSB r4,[r2,r1] + VSUB.I16 d7,d2,d3 + LDRSB r5,[r3,r1] + VADD.I16 d0,d4,d5 + VSUB.I16 d1,d4,d5 + VSUB.I16 d2,d6,d7 + LSL r5,r5,r4 + VADD.I16 d3,d6,d7 + VTRN.16 d0,d1 + VTRN.16 d2,d3 + VTRN.32 q0,q1 + VADD.I16 d4,d0,d1 + VADD.I16 d5,d2,d3 + VSUB.I16 d6,d0,d1 + VSUB.I16 d7,d2,d3 + VADD.I16 d0,d4,d5 + VSUB.I16 d1,d4,d5 + VSUB.I16 d2,d6,d7 + VADD.I16 d3,d6,d7 + VDUP.16 d5,r5 + VMOV.I32 q3,#0x2 + VMOV.I32 q4,#0x2 + VMOV.I32 q5,#0x2 + VMOV.I32 q6,#0x2 + VMLAL.S16 q3,d0,d5 + VMLAL.S16 q4,d1,d5 + VMLAL.S16 q5,d2,d5 + VMLAL.S16 q6,d3,d5 + VSHRN.I32 d0,q3,#2 + VSHRN.I32 d1,q4,#2 + VSHRN.I32 d2,q5,#2 + VSHRN.I32 d3,q6,#2 + VST1.16 {d0,d1,d2,d3},[r0] + VPOP {d8-d13} + POP {r4-r6,pc} + .endfunc + +.global omxVCM4P10_TransformDequantLumaDCFromPair +.func omxVCM4P10_TransformDequantLumaDCFromPair +omxVCM4P10_TransformDequantLumaDCFromPair: + PUSH {r4-r6,lr} + MOV r4,r1 + MOV r5,r2 + BL armVCM4P10_UnpackBlock4x4 + MOV r0,r4 + MOV r1,r5 + BL armVCM4P10_InvTransformDequantLumaDC4x4 + MOV r0,#0 + POP {r4-r6,pc} + .endfunc + + .end + |