generic-library/vpx
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Clean specializes of idct functions

Browse Source 
Change-Id: I8bb660de47b5f97263ec381dc428db96e9c9a4b2
This commit is contained in:
Linfeng Zhang
2017-05-02 14:14:06 -07:00
parent 4412996d59
commit 0178d974e5
1 changed files with 46 additions and 47 deletions
Download Patch File Download Diff File Expand all files Collapse all files

93
vpx_dsp/vpx_dsp_rtcd_defs.pl
View File

@@ -594,6 +594,36 @@ add_proto qw/void vpx_iwht4x4_16_add/, "const tran_low_t *input, uint8_t *dest,
add_proto qw/void vpx_iwht4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int stride"; add_proto qw/void vpx_iwht4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int stride";
if (vpx_config("CONFIG_EMULATE_HARDWARE") ne "yes") {
specialize qw/vpx_idct4x4_16_add neon sse2/;
specialize qw/vpx_idct4x4_1_add neon sse2/;
specialize qw/vpx_idct8x8_64_add neon sse2 ssse3/;
specialize qw/vpx_idct8x8_12_add neon sse2 ssse3/;
specialize qw/vpx_idct8x8_1_add neon sse2/;
specialize qw/vpx_idct16x16_256_add neon sse2/;
specialize qw/vpx_idct16x16_38_add neon sse2/;
$vpx_idct16x16_38_add_sse2=vpx_idct16x16_256_add_sse2;
specialize qw/vpx_idct16x16_10_add neon sse2/;
specialize qw/vpx_idct16x16_1_add neon sse2/;
specialize qw/vpx_idct32x32_1024_add neon sse2 ssse3/;
specialize qw/vpx_idct32x32_135_add neon sse2 ssse3/;
$vpx_idct32x32_135_add_sse2=vpx_idct32x32_1024_add_sse2;
specialize qw/vpx_idct32x32_34_add neon sse2 ssse3/;
specialize qw/vpx_idct32x32_1_add neon sse2/;
} # !CONFIG_EMULATE_HARDWARE
if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") { if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
# Note as optimized versions of these functions are added we need to add a check to ensure # Note as optimized versions of these functions are added we need to add a check to ensure
# that when CONFIG_EMULATE_HARDWARE is on, it defaults to the C versions only. # that when CONFIG_EMULATE_HARDWARE is on, it defaults to the C versions only.
@@ -634,35 +664,6 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
add_proto qw/void vpx_highbd_iwht4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int stride, int bd"; add_proto qw/void vpx_highbd_iwht4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int stride, int bd";
if (vpx_config("CONFIG_EMULATE_HARDWARE") ne "yes") { if (vpx_config("CONFIG_EMULATE_HARDWARE") ne "yes") {
specialize qw/vpx_idct4x4_16_add neon sse2/;
specialize qw/vpx_idct4x4_1_add neon sse2/;
specialize qw/vpx_idct8x8_64_add neon sse2 ssse3/;
specialize qw/vpx_idct8x8_12_add neon sse2 ssse3/;
specialize qw/vpx_idct8x8_1_add neon sse2/;
specialize qw/vpx_idct16x16_256_add neon sse2/;
specialize qw/vpx_idct16x16_38_add neon sse2/;
$vpx_idct16x16_38_add_sse2=vpx_idct16x16_256_add_sse2;
specialize qw/vpx_idct16x16_10_add neon sse2/;
specialize qw/vpx_idct16x16_1_add neon sse2/;
specialize qw/vpx_idct32x32_1024_add neon sse2 ssse3/;
specialize qw/vpx_idct32x32_135_add neon sse2 ssse3/;
# Need to add 135 eob idct32x32 implementations.
$vpx_idct32x32_135_add_sse2=vpx_idct32x32_1024_add_sse2;
specialize qw/vpx_idct32x32_34_add neon sse2 ssse3/;
specialize qw/vpx_idct32x32_1_add neon sse2/;
specialize qw/vpx_highbd_idct4x4_16_add neon sse2/; specialize qw/vpx_highbd_idct4x4_16_add neon sse2/;
specialize qw/vpx_highbd_idct8x8_64_add neon sse2/; specialize qw/vpx_highbd_idct8x8_64_add neon sse2/;
@@ -684,41 +685,39 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
} # !CONFIG_EMULATE_HARDWARE } # !CONFIG_EMULATE_HARDWARE
} else { } else {
if (vpx_config("CONFIG_EMULATE_HARDWARE") ne "yes") { if (vpx_config("CONFIG_EMULATE_HARDWARE") ne "yes") {
specialize qw/vpx_idct4x4_1_add sse2 neon dspr2 msa/; specialize qw/vpx_idct4x4_16_add dspr2 msa/;
specialize qw/vpx_idct4x4_16_add sse2 neon dspr2 msa/; specialize qw/vpx_idct4x4_1_add dspr2 msa/;
specialize qw/vpx_idct8x8_1_add sse2 neon dspr2 msa/; specialize qw/vpx_idct8x8_64_add dspr2 msa/;
specialize qw/vpx_idct8x8_64_add sse2 ssse3 neon dspr2 msa/; specialize qw/vpx_idct8x8_12_add dspr2 msa/;
specialize qw/vpx_idct8x8_12_add sse2 ssse3 neon dspr2 msa/; specialize qw/vpx_idct8x8_1_add dspr2 msa/;
specialize qw/vpx_idct16x16_1_add sse2 neon dspr2 msa/; specialize qw/vpx_idct16x16_256_add dspr2 msa/;
specialize qw/vpx_idct16x16_256_add sse2 neon dspr2 msa/; specialize qw/vpx_idct16x16_38_add dspr2 msa/;
specialize qw/vpx_idct16x16_38_add sse2 neon dspr2 msa/;
$vpx_idct16x16_38_add_sse2=vpx_idct16x16_256_add_sse2;
$vpx_idct16x16_38_add_dspr2=vpx_idct16x16_256_add_dspr2; $vpx_idct16x16_38_add_dspr2=vpx_idct16x16_256_add_dspr2;
$vpx_idct16x16_38_add_msa=vpx_idct16x16_256_add_msa; $vpx_idct16x16_38_add_msa=vpx_idct16x16_256_add_msa;
specialize qw/vpx_idct16x16_10_add sse2 neon dspr2 msa/; specialize qw/vpx_idct16x16_10_add dspr2 msa/;
specialize qw/vpx_idct32x32_1024_add sse2 ssse3 neon dspr2 msa/; specialize qw/vpx_idct16x16_1_add dspr2 msa/;
specialize qw/vpx_idct32x32_135_add sse2 ssse3 neon dspr2 msa/; specialize qw/vpx_idct32x32_1024_add dspr2 msa/;
$vpx_idct32x32_135_add_sse2=vpx_idct32x32_1024_add_sse2;
specialize qw/vpx_idct32x32_135_add dspr2 msa/;
$vpx_idct32x32_135_add_dspr2=vpx_idct32x32_1024_add_dspr2; $vpx_idct32x32_135_add_dspr2=vpx_idct32x32_1024_add_dspr2;
$vpx_idct32x32_135_add_msa=vpx_idct32x32_1024_add_msa; $vpx_idct32x32_135_add_msa=vpx_idct32x32_1024_add_msa;
specialize qw/vpx_idct32x32_34_add sse2 ssse3 neon dspr2 msa/; specialize qw/vpx_idct32x32_34_add dspr2 msa/;
specialize qw/vpx_idct32x32_1_add sse2 neon dspr2 msa/; specialize qw/vpx_idct32x32_1_add dspr2 msa/;
specialize qw/vpx_iwht4x4_1_add msa/;
specialize qw/vpx_iwht4x4_16_add msa sse2/; specialize qw/vpx_iwht4x4_16_add msa sse2/;
specialize qw/vpx_iwht4x4_1_add msa/;
} # !CONFIG_EMULATE_HARDWARE } # !CONFIG_EMULATE_HARDWARE
} # CONFIG_VP9_HIGHBITDEPTH } # CONFIG_VP9_HIGHBITDEPTH
} # CONFIG_VP9 } # CONFIG_VP9