diff options
| author | Timothy B. Terriberry <tterribe@xiph.org> | 2010-10-20 15:39:11 -0700 |
|---|---|---|
| committer | Johann <johannkoenig@google.com> | 2010-10-25 09:23:29 -0400 |
| commit | b71962fdc98c3fb9fdf95d74191452cd51cbc2b5 (patch) | |
| tree | d89b4c30c5f3a2f55e24711cd9f45963267065f2 | |
| parent | e81e30c25de3a75e8b02b38d55c65134874f83aa (diff) | |
| download | libvpx-b71962fdc98c3fb9fdf95d74191452cd51cbc2b5.tar libvpx-b71962fdc98c3fb9fdf95d74191452cd51cbc2b5.tar.gz libvpx-b71962fdc98c3fb9fdf95d74191452cd51cbc2b5.tar.bz2 libvpx-b71962fdc98c3fb9fdf95d74191452cd51cbc2b5.zip | |
Add runtime CPU detection support for ARM.
The primary goal is to allow a binary to be built which supports
NEON, but can fall back to non-NEON routines, since some Android
devices do not have NEON, even if they are otherwise ARMv7 (e.g.,
Tegra).
The configure-generated flags HAVE_ARMV7, etc., are used to decide
which versions of each function to build, and when
CONFIG_RUNTIME_CPU_DETECT is enabled, the correct version is chosen
at run time.
In order for this to work, the CFLAGS must be set to something
appropriate (e.g., without -mfpu=neon for ARMv7, and with
appropriate -march and -mcpu for even earlier configurations), or
the native C code will not be able to run.
The ASFLAGS must remain set for the most advanced instruction set
required at build time, since the ARM assembler will refuse to emit
them otherwise.
I have not attempted to make any changes to configure to do this
automatically.
Doing so will probably require the addition of new configure options.
Many of the hooks for RTCD on ARM were already there, but a lot of
the code had bit-rotted, and a good deal of the ARM-specific code
is not integrated into the RTCD structs at all.
I did not try to resolve the latter, merely to add the minimal amount
of protection around them to allow RTCD to work.
Those functions that were called based on an ifdef at the calling
site were expanded to check the RTCD flags at that site, but they
should be added to an RTCD struct somewhere in the future.
The functions invoked with global function pointers still are, but
these should be moved into an RTCD struct for thread safety (I
believe every platform currently supported has atomic pointer
stores, but this is not guaranteed).
The encoder's boolhuff functions did not even have _c and armv7
suffixes, and the correct version was resolved at link time.
The token packing functions did have appropriate suffixes, but the
version was selected with a define, with no associated RTCD struct.
However, for both of these, the only armv7 instruction they actually
used was rbit, and this was completely superfluous, so I reworked
them to avoid it.
The only non-ARMv4 instruction remaining in them is clz, which is
ARMv5 (not even ARMv5TE is required).
Considering that there are no ARM-specific configs which are not at
least ARMv5TE, I did not try to detect these at runtime, and simply
enable them for ARMv5 and above.
Finally, the NEON register saving code was completely non-reentrant,
since it saved the registers to a global, static variable.
I moved the storage for this onto the stack.
A single binary built with this code was tested on an ARM11 (ARMv6)
and a Cortex A8 (ARMv7 w/NEON), for both the encoder and decoder,
and produced identical output, while using the correct accelerated
functions on each.
I did not test on any earlier processors.
Change-Id: I45cbd63a614f4554c3b325c45d46c0806f009eaa
33 files changed, 841 insertions, 471 deletions
@@ -93,6 +93,7 @@ CODEC_SRCS-$(BUILD_LIBVPX) += vpx_ports/x86.h CODEC_SRCS-$(BUILD_LIBVPX) += vpx_ports/x86_abi_support.asm CODEC_SRCS-$(BUILD_LIBVPX) += vpx_ports/x86_cpuid.c endif +CODEC_SRCS-$(ARCH_ARM) += vpx_ports/arm_cpudetect.c CODEC_SRCS-$(ARCH_ARM) += $(BUILD_PFX)vpx_config.asm CODEC_EXPORTS-$(BUILD_LIBVPX) += vpx/exports_com CODEC_EXPORTS-$(CONFIG_ENCODERS) += vpx/exports_enc diff --git a/vp8/common/arm/arm_systemdependent.c b/vp8/common/arm/arm_systemdependent.c new file mode 100644 index 000000000..fe62fae13 --- /dev/null +++ b/vp8/common/arm/arm_systemdependent.c @@ -0,0 +1,134 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include "vpx_ports/config.h" +#include "vpx_ports/arm.h" +#include "g_common.h" +#include "pragmas.h" +#include "subpixel.h" +#include "loopfilter.h" +#include "recon.h" +#include "idct.h" +#include "onyxc_int.h" + +extern void (*vp8_build_intra_predictors_mby_ptr)(MACROBLOCKD *x); +extern void vp8_build_intra_predictors_mby(MACROBLOCKD *x); +extern void vp8_build_intra_predictors_mby_neon(MACROBLOCKD *x); + +extern void (*vp8_build_intra_predictors_mby_s_ptr)(MACROBLOCKD *x); +extern void vp8_build_intra_predictors_mby_s(MACROBLOCKD *x); +extern void vp8_build_intra_predictors_mby_s_neon(MACROBLOCKD *x); + +void vp8_arch_arm_common_init(VP8_COMMON *ctx) +{ +#if CONFIG_RUNTIME_CPU_DETECT + VP8_COMMON_RTCD *rtcd = &ctx->rtcd; + int flags = arm_cpu_caps(); + int has_edsp = flags & HAS_EDSP; + int has_media = flags & HAS_MEDIA; + int has_neon = flags & HAS_NEON; + rtcd->flags = flags; + + /* Override default functions with fastest ones for this CPU. */ +#if HAVE_ARMV6 + if (has_media) + { + rtcd->subpix.sixtap16x16 = vp8_sixtap_predict16x16_armv6; + rtcd->subpix.sixtap8x8 = vp8_sixtap_predict8x8_armv6; + rtcd->subpix.sixtap8x4 = vp8_sixtap_predict8x4_armv6; + rtcd->subpix.sixtap4x4 = vp8_sixtap_predict_armv6; + rtcd->subpix.bilinear16x16 = vp8_bilinear_predict16x16_armv6; + rtcd->subpix.bilinear8x8 = vp8_bilinear_predict8x8_armv6; + rtcd->subpix.bilinear8x4 = vp8_bilinear_predict8x4_armv6; + rtcd->subpix.bilinear4x4 = vp8_bilinear_predict4x4_armv6; + + rtcd->idct.idct1 = vp8_short_idct4x4llm_1_v6; + rtcd->idct.idct16 = vp8_short_idct4x4llm_v6_dual; + rtcd->idct.iwalsh1 = vp8_short_inv_walsh4x4_1_v6; + rtcd->idct.iwalsh16 = vp8_short_inv_walsh4x4_v6; + + rtcd->loopfilter.normal_mb_v = vp8_loop_filter_mbv_armv6; + rtcd->loopfilter.normal_b_v = vp8_loop_filter_bv_armv6; + rtcd->loopfilter.normal_mb_h = vp8_loop_filter_mbh_armv6; + rtcd->loopfilter.normal_b_h = vp8_loop_filter_bh_armv6; + rtcd->loopfilter.simple_mb_v = vp8_loop_filter_mbvs_armv6; + rtcd->loopfilter.simple_b_v = vp8_loop_filter_bvs_armv6; + rtcd->loopfilter.simple_mb_h = vp8_loop_filter_mbhs_armv6; + rtcd->loopfilter.simple_b_h = vp8_loop_filter_bhs_armv6; + + rtcd->recon.copy16x16 = vp8_copy_mem16x16_v6; + rtcd->recon.copy8x8 = vp8_copy_mem8x8_v6; + rtcd->recon.copy8x4 = vp8_copy_mem8x4_v6; + rtcd->recon.recon = vp8_recon_b_armv6; + rtcd->recon.recon2 = vp8_recon2b_armv6; + rtcd->recon.recon4 = vp8_recon4b_armv6; + } +#endif + +#if HAVE_ARMV7 + if (has_neon) + { + rtcd->subpix.sixtap16x16 = vp8_sixtap_predict16x16_neon; + rtcd->subpix.sixtap8x8 = vp8_sixtap_predict8x8_neon; + rtcd->subpix.sixtap8x4 = vp8_sixtap_predict8x4_neon; + rtcd->subpix.sixtap4x4 = vp8_sixtap_predict_neon; + rtcd->subpix.bilinear16x16 = vp8_bilinear_predict16x16_neon; + rtcd->subpix.bilinear8x8 = vp8_bilinear_predict8x8_neon; + rtcd->subpix.bilinear8x4 = vp8_bilinear_predict8x4_neon; + rtcd->subpix.bilinear4x4 = vp8_bilinear_predict4x4_neon; + + rtcd->idct.idct1 = vp8_short_idct4x4llm_1_neon; + rtcd->idct.idct16 = vp8_short_idct4x4llm_neon; + rtcd->idct.iwalsh1 = vp8_short_inv_walsh4x4_1_neon; + rtcd->idct.iwalsh16 = vp8_short_inv_walsh4x4_neon; + + rtcd->loopfilter.normal_mb_v = vp8_loop_filter_mbv_neon; + rtcd->loopfilter.normal_b_v = vp8_loop_filter_bv_neon; + rtcd->loopfilter.normal_mb_h = vp8_loop_filter_mbh_neon; + rtcd->loopfilter.normal_b_h = vp8_loop_filter_bh_neon; + rtcd->loopfilter.simple_mb_v = vp8_loop_filter_mbvs_neon; + rtcd->loopfilter.simple_b_v = vp8_loop_filter_bvs_neon; + rtcd->loopfilter.simple_mb_h = vp8_loop_filter_mbhs_neon; + rtcd->loopfilter.simple_b_h = vp8_loop_filter_bhs_neon; + + rtcd->recon.copy16x16 = vp8_copy_mem16x16_neon; + rtcd->recon.copy8x8 = vp8_copy_mem8x8_neon; + rtcd->recon.copy8x4 = vp8_copy_mem8x4_neon; + rtcd->recon.recon = vp8_recon_b_neon; + rtcd->recon.recon2 = vp8_recon2b_neon; + rtcd->recon.recon4 = vp8_recon4b_neon; + } +#endif + +#endif + +#if HAVE_ARMV6 +#if CONFIG_RUNTIME_CPU_DETECT + if (has_media) +#endif + { + vp8_build_intra_predictors_mby_ptr = vp8_build_intra_predictors_mby; + vp8_build_intra_predictors_mby_s_ptr = vp8_build_intra_predictors_mby_s; + } +#endif + +#if HAVE_ARMV7 +#if CONFIG_RUNTIME_CPU_DETECT + if (has_neon) +#endif + { + vp8_build_intra_predictors_mby_ptr = + vp8_build_intra_predictors_mby_neon; + vp8_build_intra_predictors_mby_s_ptr = + vp8_build_intra_predictors_mby_s_neon; + } +#endif +} diff --git a/vp8/common/arm/idct_arm.h b/vp8/common/arm/idct_arm.h index f28d7f649..8b8d17917 100644 --- a/vp8/common/arm/idct_arm.h +++ b/vp8/common/arm/idct_arm.h @@ -19,6 +19,7 @@ extern prototype_idct_scalar_add(vp8_dc_only_idct_add_v6); extern prototype_second_order(vp8_short_inv_walsh4x4_1_v6); extern prototype_second_order(vp8_short_inv_walsh4x4_v6); +#if !CONFIG_RUNTIME_CPU_DETECT #undef vp8_idct_idct1 #define vp8_idct_idct1 vp8_short_idct4x4llm_1_v6 @@ -34,6 +35,7 @@ extern prototype_second_order(vp8_short_inv_walsh4x4_v6); #undef vp8_idct_iwalsh16 #define vp8_idct_iwalsh16 vp8_short_inv_walsh4x4_v6 #endif +#endif #if HAVE_ARMV7 extern prototype_idct(vp8_short_idct4x4llm_1_neon); @@ -42,6 +44,7 @@ extern prototype_idct_scalar_add(vp8_dc_only_idct_add_neon); extern prototype_second_order(vp8_short_inv_walsh4x4_1_neon); extern prototype_second_order(vp8_short_inv_walsh4x4_neon); +#if !CONFIG_RUNTIME_CPU_DETECT #undef vp8_idct_idct1 #define vp8_idct_idct1 vp8_short_idct4x4llm_1_neon @@ -57,5 +60,6 @@ extern prototype_second_order(vp8_short_inv_walsh4x4_neon); #undef vp8_idct_iwalsh16 #define vp8_idct_iwalsh16 vp8_short_inv_walsh4x4_neon #endif +#endif #endif diff --git a/vp8/common/arm/loopfilter_arm.h b/vp8/common/arm/loopfilter_arm.h index 6c3628ae9..cd62207d7 100644 --- a/vp8/common/arm/loopfilter_arm.h +++ b/vp8/common/arm/loopfilter_arm.h @@ -22,6 +22,7 @@ extern prototype_loopfilter_block(vp8_loop_filter_bvs_armv6); extern prototype_loopfilter_block(vp8_loop_filter_mbhs_armv6); extern prototype_loopfilter_block(vp8_loop_filter_bhs_armv6); +#if !CONFIG_RUNTIME_CPU_DETECT #undef vp8_lf_normal_mb_v #define vp8_lf_normal_mb_v vp8_loop_filter_mbv_armv6 @@ -46,6 +47,7 @@ extern prototype_loopfilter_block(vp8_loop_filter_bhs_armv6); #undef vp8_lf_simple_b_h #define vp8_lf_simple_b_h vp8_loop_filter_bhs_armv6 #endif +#endif #if HAVE_ARMV7 extern prototype_loopfilter_block(vp8_loop_filter_mbv_neon); @@ -57,6 +59,7 @@ extern prototype_loopfilter_block(vp8_loop_filter_bvs_neon); extern prototype_loopfilter_block(vp8_loop_filter_mbhs_neon); extern prototype_loopfilter_block(vp8_loop_filter_bhs_neon); +#if !CONFIG_RUNTIME_CPU_DETECT #undef vp8_lf_normal_mb_v #define vp8_lf_normal_mb_v vp8_loop_filter_mbv_neon @@ -81,5 +84,6 @@ extern prototype_loopfilter_block(vp8_loop_filter_bhs_neon); #undef vp8_lf_simple_b_h #define vp8_lf_simple_b_h vp8_loop_filter_bhs_neon #endif +#endif #endif diff --git a/vp8/common/arm/recon_arm.h b/vp8/common/arm/recon_arm.h index 18855a3c0..c30f6dc2d 100644 --- a/vp8/common/arm/recon_arm.h +++ b/vp8/common/arm/recon_arm.h @@ -21,6 +21,7 @@ extern prototype_copy_block(vp8_copy_mem8x8_v6); extern prototype_copy_block(vp8_copy_mem8x4_v6); extern prototype_copy_block(vp8_copy_mem16x16_v6); +#if !CONFIG_RUNTIME_CPU_DETECT #undef vp8_recon_recon #define vp8_recon_recon vp8_recon_b_armv6 @@ -39,6 +40,7 @@ extern prototype_copy_block(vp8_copy_mem16x16_v6); #undef vp8_recon_copy16x16 #define vp8_recon_copy16x16 vp8_copy_mem16x16_v6 #endif +#endif #if HAVE_ARMV7 extern prototype_recon_block(vp8_recon_b_neon); @@ -49,6 +51,7 @@ extern prototype_copy_block(vp8_copy_mem8x8_neon); extern prototype_copy_block(vp8_copy_mem8x4_neon); extern prototype_copy_block(vp8_copy_mem16x16_neon); +#if !CONFIG_RUNTIME_CPU_DETECT #undef vp8_recon_recon #define vp8_recon_recon vp8_recon_b_neon @@ -67,5 +70,6 @@ extern prototype_copy_block(vp8_copy_mem16x16_neon); #undef vp8_recon_copy16x16 #define vp8_recon_copy16x16 vp8_copy_mem16x16_neon #endif +#endif #endif diff --git a/vp8/common/arm/subpixel_arm.h b/vp8/common/arm/subpixel_arm.h index 53600e547..6288538d0 100644 --- a/vp8/common/arm/subpixel_arm.h +++ b/vp8/common/arm/subpixel_arm.h @@ -22,6 +22,7 @@ extern prototype_subpixel_predict(vp8_bilinear_predict8x8_armv6); extern prototype_subpixel_predict(vp8_bilinear_predict8x4_armv6); extern prototype_subpixel_predict(vp8_bilinear_predict4x4_armv6); +#if !CONFIG_RUNTIME_CPU_DETECT #undef vp8_subpix_sixtap16x16 #define vp8_subpix_sixtap16x16 vp8_sixtap_predict16x16_armv6 @@ -46,6 +47,7 @@ extern prototype_subpixel_predict(vp8_bilinear_predict4x4_armv6); #undef vp8_subpix_bilinear4x4 #define vp8_subpix_bilinear4x4 vp8_bilinear_predict4x4_armv6 #endif +#endif #if HAVE_ARMV7 extern prototype_subpixel_predict(vp8_sixtap_predict16x16_neon); @@ -57,6 +59,7 @@ extern prototype_subpixel_predict(vp8_bilinear_predict8x8_neon); extern prototype_subpixel_predict(vp8_bilinear_predict8x4_neon); extern prototype_subpixel_predict(vp8_bilinear_predict4x4_neon); +#if !CONFIG_RUNTIME_CPU_DETECT #undef vp8_subpix_sixtap16x16 #define vp8_subpix_sixtap16x16 vp8_sixtap_predict16x16_neon @@ -81,5 +84,6 @@ extern prototype_subpixel_predict(vp8_bilinear_predict4x4_neon); #undef vp8_subpix_bilinear4x4 #define vp8_subpix_bilinear4x4 vp8_bilinear_predict4x4_neon #endif +#endif #endif diff --git a/vp8/common/arm/systemdependent.c b/vp8/common/arm/systemdependent.c deleted file mode 100644 index 1eed97e02..000000000 --- a/vp8/common/arm/systemdependent.c +++ /dev/null @@ -1,149 +0,0 @@ -/* - * Copyright (c) 2010 The WebM project authors. All Rights Reserved. - * - * Use of this source code is governed by a BSD-style license - * that can be found in the LICENSE file in the root of the source - * tree. An additional intellectual property rights grant can be found - * in the file PATENTS. All contributing project authors may - * be found in the AUTHORS file in the root of the source tree. - */ - - -#include "vpx_ports/config.h" -#include "g_common.h" -#include "pragmas.h" -#include "subpixel.h" -#include "loopfilter.h" -#include "recon.h" -#include "idct.h" -#include "onyxc_int.h" - -void (*vp8_build_intra_predictors_mby_ptr)(MACROBLOCKD *x); -extern void vp8_build_intra_predictors_mby(MACROBLOCKD *x); -extern void vp8_build_intra_predictors_mby_neon(MACROBLOCKD *x); - -void (*vp8_build_intra_predictors_mby_s_ptr)(MACROBLOCKD *x); -extern void vp8_build_intra_predictors_mby_s(MACROBLOCKD *x); -extern void vp8_build_intra_predictors_mby_s_neon(MACROBLOCKD *x); - -void vp8_machine_specific_config(VP8_COMMON *ctx) -{ -#if CONFIG_RUNTIME_CPU_DETECT - VP8_COMMON_RTCD *rtcd = &ctx->rtcd; - -#if HAVE_ARMV7 - rtcd->subpix.sixtap16x16 = vp8_sixtap_predict16x16_neon; - rtcd->subpix.sixtap8x8 = vp8_sixtap_predict8x8_neon; - rtcd->subpix.sixtap8x4 = vp8_sixtap_predict8x4_neon; - rtcd->subpix.sixtap4x4 = vp8_sixtap_predict_neon; - rtcd->subpix.bilinear16x16 = vp8_bilinear_predict16x16_neon; - rtcd->subpix.bilinear8x8 = vp8_bilinear_predict8x8_neon; - rtcd->subpix.bilinear8x4 = vp8_bilinear_predict8x4_neon; - rtcd->subpix.bilinear4x4 = vp8_bilinear_predict4x4_neon; - - rtcd->idct.idct1 = vp8_short_idct4x4llm_1_neon; - rtcd->idct.idct16 = vp8_short_idct4x4llm_neon; - rtcd->idct.iwalsh1 = vp8_short_inv_walsh4x4_1_neon; - rtcd->idct.iwalsh16 = vp8_short_inv_walsh4x4_neon; - - rtcd->loopfilter.normal_mb_v = vp8_loop_filter_mbv_neon; - rtcd->loopfilter.normal_b_v = vp8_loop_filter_bv_neon; - rtcd->loopfilter.normal_mb_h = vp8_loop_filter_mbh_neon; - rtcd->loopfilter.normal_b_h = vp8_loop_filter_bh_neon; - rtcd->loopfilter.simple_mb_v = vp8_loop_filter_mbvs_neon; - rtcd->loopfilter.simple_b_v = vp8_loop_filter_bvs_neon; - rtcd->loopfilter.simple_mb_h = vp8_loop_filter_mbhs_neon; - rtcd->loopfilter.simple_b_h = vp8_loop_filter_bhs_neon; - - rtcd->recon.copy16x16 = vp8_copy_mem16x16_neon; - rtcd->recon.copy8x8 = vp8_copy_mem8x8_neon; - rtcd->recon.copy8x4 = vp8_copy_mem8x4_neon; - rtcd->recon.recon = vp8_recon_b_neon; - rtcd->recon.recon2 = vp8_recon2b_neon; - rtcd->recon.recon4 = vp8_recon4b_neon; -#elif HAVE_ARMV6 - - rtcd->subpix.sixtap16x16 = vp8_sixtap_predict16x16_armv6; - rtcd->subpix.sixtap8x8 = vp8_sixtap_predict8x8_armv6; - rtcd->subpix.sixtap8x4 = vp8_sixtap_predict8x4_armv6; - rtcd->subpix.sixtap4x4 = vp8_sixtap_predict_armv6; - rtcd->subpix.bilinear16x16 = vp8_bilinear_predict16x16_armv6; - rtcd->subpix.bilinear8x8 = vp8_bilinear_predict8x8_armv6; - rtcd->subpix.bilinear8x4 = vp8_bilinear_predict8x4_armv6; - rtcd->subpix.bilinear4x4 = vp8_bilinear_predict4x4_armv6; - - rtcd->idct.idct1 = vp8_short_idct4x4llm_1_v6; - rtcd->idct.idct16 = vp8_short_idct4x4llm_v6_dual; - rtcd->idct.iwalsh1 = vp8_short_inv_walsh4x4_1_armv6; - rtcd->idct.iwalsh16 = vp8_short_inv_walsh4x4_armv6; - - rtcd->loopfilter.normal_mb_v = vp8_loop_filter_mbv_armv6; - rtcd->loopfilter.normal_b_v = vp8_loop_filter_bv_armv6; - rtcd->loopfilter.normal_mb_h = vp8_loop_filter_mbh_armv6; - rtcd->loopfilter.normal_b_h = vp8_loop_filter_bh_armv6; - rtcd->loopfilter.simple_mb_v = vp8_loop_filter_mbvs_armv6; - rtcd->loopfilter.simple_b_v = vp8_loop_filter_bvs_armv6; - rtcd->loopfilter.simple_mb_h = vp8_loop_filter_mbhs_armv6; - rtcd->loopfilter.simple_b_h = vp8_loop_filter_bhs_armv6; - - rtcd->recon.copy16x16 = vp8_copy_mem16x16_v6; - rtcd->recon.copy8x8 = vp8_copy_mem8x8_v6; - rtcd->recon.copy8x4 = vp8_copy_mem8x4_v6; - rtcd->recon.recon = vp8_recon_b_armv6; - rtcd->recon.recon2 = vp8_recon2b_armv6; - rtcd->recon.recon4 = vp8_recon4b_armv6; -#else -//pure c - rtcd->idct.idct1 = vp8_short_idct4x4llm_1_c; - rtcd->idct.idct16 = vp8_short_idct4x4llm_c; - rtcd->idct.idct1_scalar = vp8_dc_only_idct_c; - rtcd->idct.iwalsh1 = vp8_short_inv_walsh4x4_1_c; - rtcd->idct.iwalsh16 = vp8_short_inv_walsh4x4_c; - - rtcd->recon.copy16x16 = vp8_copy_mem16x16_c; - rtcd->recon.copy8x8 = vp8_copy_mem8x8_c; - rtcd->recon.copy8x4 = vp8_copy_mem8x4_c; - rtcd->recon.recon = vp8_recon_b_c; - rtcd->recon.recon2 = vp8_recon2b_c; - rtcd->recon.recon4 = vp8_recon4b_c; - - rtcd->subpix.sixtap16x16 = vp8_sixtap_predict16x16_c; - rtcd->subpix.sixtap8x8 = vp8_sixtap_predict8x8_c; - rtcd->subpix.sixtap8x4 = vp8_sixtap_predict8x4_c; - rtcd->subpix.sixtap4x4 = vp8_sixtap_predict_c; - rtcd->subpix.bilinear16x16 = vp8_bilinear_predict16x16_c; - rtcd->subpix.bilinear8x8 = vp8_bilinear_predict8x8_c; - rtcd->subpix.bilinear8x4 = vp8_bilinear_predict8x4_c; - rtcd->subpix.bilinear4x4 = vp8_bilinear_predict4x4_c; - - rtcd->loopfilter.normal_mb_v = vp8_loop_filter_mbv_c; - rtcd->loopfilter.normal_b_v = vp8_loop_filter_bv_c; - rtcd->loopfilter.normal_mb_h = vp8_loop_filter_mbh_c; - rtcd->loopfilter.normal_b_h = vp8_loop_filter_bh_c; - rtcd->loopfilter.simple_mb_v = vp8_loop_filter_mbvs_c; - rtcd->loopfilter.simple_b_v = vp8_loop_filter_bvs_c; - rtcd->loopfilter.simple_mb_h = vp8_loop_filter_mbhs_c; - rtcd->loopfilter.simple_b_h = vp8_loop_filter_bhs_c; -#endif - -#if CONFIG_POSTPROC || (CONFIG_VP8_ENCODER && CONFIG_PSNR) - rtcd->postproc.down = vp8_mbpost_proc_down_c; - rtcd->postproc.across = vp8_mbpost_proc_across_ip_c; - rtcd->postproc.downacross = vp8_post_proc_down_and_across_c; - rtcd->postproc.addnoise = vp8_plane_add_noise_c; -#endif -#endif - -#if HAVE_ARMV7 - vp8_build_intra_predictors_mby_ptr = vp8_build_intra_predictors_mby_neon; - vp8_build_intra_predictors_mby_s_ptr = vp8_build_intra_predictors_mby_s_neon; -#elif HAVE_ARMV6 - vp8_build_intra_predictors_mby_ptr = vp8_build_intra_predictors_mby; - vp8_build_intra_predictors_mby_s_ptr = vp8_build_intra_predictors_mby_s; -#else - vp8_build_intra_predictors_mby_ptr = vp8_build_intra_predictors_mby; - vp8_build_intra_predictors_mby_s_ptr = vp8_build_intra_predictors_mby_s; - -#endif - -} diff --git a/vp8/common/generic/systemdependent.c b/vp8/common/generic/systemdependent.c index c04e31ffe..0ef375e33 100644 --- a/vp8/common/generic/systemdependent.c +++ b/vp8/common/generic/systemdependent.c @@ -18,6 +18,7 @@ #include "onyxc_int.h" extern void vp8_arch_x86_common_init(VP8_COMMON *ctx); +extern void vp8_arch_arm_common_init(VP8_COMMON *ctx); void (*vp8_build_intra_predictors_mby_ptr)(MACROBLOCKD *x); extern void vp8_build_intra_predictors_mby(MACROBLOCKD *x); @@ -77,4 +78,8 @@ void vp8_machine_specific_config(VP8_COMMON *ctx) vp8_arch_x86_common_init(ctx); #endif +#if ARCH_ARM + vp8_arch_arm_common_init(ctx); +#endif + } diff --git a/vp8/common/onyxc_int.h b/vp8/common/onyxc_int.h index 4966002f5..d12143d4d 100644 --- a/vp8/common/onyxc_int.h +++ b/vp8/common/onyxc_int.h @@ -74,6 +74,7 @@ typedef struct VP8_COMMON_RTCD vp8_subpix_rtcd_vtable_t subpix; vp8_loopfilter_rtcd_vtable_t loopfilter; vp8_postproc_rtcd_vtable_t postproc; + int flags; #else int unused; #endif diff --git a/vp8/decoder/arm/arm_dsystemdependent.c b/vp8/decoder/arm/arm_dsystemdependent.c new file mode 100644 index 000000000..77cff47db --- /dev/null +++ b/vp8/decoder/arm/arm_dsystemdependent.c @@ -0,0 +1,66 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include "vpx_ports/config.h" +#include "vpx_ports/arm.h" +#include "blockd.h" +#include "pragmas.h" +#include "postproc.h" +#include "dboolhuff.h" +#include "dequantize.h" +#include "onyxd_int.h" + +void vp8_arch_arm_decode_init(VP8D_COMP *pbi) +{ +#if CONFIG_RUNTIME_CPU_DETECT + int flags = pbi->common.rtcd.flags; + int has_edsp = flags & HAS_EDSP; + int has_media = flags & HAS_MEDIA; + int has_neon = flags & HAS_NEON; + +#if HAVE_ARMV6 + if (has_media) + { + pbi->dequant.block = vp8_dequantize_b_v6; + pbi->dequant.idct_add = vp8_dequant_idct_add_v6; + pbi->dequant.dc_idct_add = vp8_dequant_dc_idct_add_v6; + pbi->dequant.dc_idct_add_y_block = vp8_dequant_dc_idct_add_y_block_v6; + pbi->dequant.idct_add_y_block = vp8_dequant_idct_add_y_block_v6; + pbi->dequant.idct_add_uv_block = vp8_dequant_idct_add_uv_block_v6; +#if 0 //For use with RTCD, when implemented + pbi->dboolhuff.start = vp8dx_start_decode_c; + pbi->dboolhuff.fill = vp8dx_bool_decoder_fill_c; + pbi->dboolhuff.debool = vp8dx_decode_bool_c; + pbi->dboolhuff.devalue = vp8dx_decode_value_c; +#endif + } +#endif + +#if HAVE_ARMV7 + if (has_neon) + { + pbi->dequant.block = vp8_dequantize_b_neon; + pbi->dequant.idct_add = vp8_dequant_idct_add_neon; + /*This is not used: NEON always dequants two blocks at once. + pbi->dequant.dc_idct_add = vp8_dequant_dc_idct_add_neon;*/ + pbi->dequant.dc_idct_add_y_block = vp8_dequant_dc_idct_add_y_block_neon; + pbi->dequant.idct_add_y_block = vp8_dequant_idct_add_y_block_neon; + pbi->dequant.idct_add_uv_block = vp8_dequant_idct_add_uv_block_neon; +#if 0 //For use with RTCD, when implemented + pbi->dboolhuff.start = vp8dx_start_decode_c; + pbi->dboolhuff.fill = vp8dx_bool_decoder_fill_c; + pbi->dboolhuff.debool = vp8dx_decode_bool_c; + pbi->dboolhuff.devalue = vp8dx_decode_value_c; +#endif + } +#endif +#endif +} diff --git a/vp8/decoder/arm/dequantize_arm.h b/vp8/decoder/arm/dequantize_arm.h index 40151e01a..b7d800d26 100644 --- a/vp8/decoder/arm/dequantize_arm.h +++ b/vp8/decoder/arm/dequantize_arm.h @@ -20,6 +20,7 @@ extern prototype_dequant_dc_idct_add_y_block(vp8_dequant_dc_idct_add_y_block_v6) extern prototype_dequant_idct_add_y_block(vp8_dequant_idct_add_y_block_v6); extern prototype_dequant_idct_add_uv_block(vp8_dequant_idct_add_uv_block_v6); +#if !CONFIG_RUNTIME_CPU_DETECT #undef vp8_dequant_block #define vp8_dequant_block vp8_dequantize_b_v6 @@ -38,6 +39,7 @@ extern prototype_dequant_idct_add_uv_block(vp8_dequant_idct_add_uv_block_v6); #undef vp8_dequant_idct_add_uv_block #define vp8_dequant_idct_add_uv_block vp8_dequant_idct_add_uv_block_v6 #endif +#endif #if HAVE_ARMV7 extern prototype_dequant_block(vp8_dequantize_b_neon); @@ -47,6 +49,7 @@ extern prototype_dequant_dc_idct_add_y_block(vp8_dequant_dc_idct_add_y_block_neo extern prototype_dequant_idct_add_y_block(vp8_dequant_idct_add_y_block_neon); extern prototype_dequant_idct_add_uv_block(vp8_dequant_idct_add_uv_block_neon); +#if !CONFIG_RUNTIME_CPU_DETECT #undef vp8_dequant_block #define vp8_dequant_block vp8_dequantize_b_neon @@ -65,5 +68,6 @@ extern prototype_dequant_idct_add_uv_block(vp8_dequant_idct_add_uv_block_neon); #undef vp8_dequant_idct_add_uv_block #define vp8_dequant_idct_add_uv_block vp8_dequant_idct_add_uv_block_neon #endif +#endif #endif diff --git a/vp8/decoder/arm/dsystemdependent.c b/vp8/decoder/arm/dsystemdependent.c deleted file mode 100644 index 9dcf7b657..000000000 --- a/vp8/decoder/arm/dsystemdependent.c +++ /dev/null @@ -1,39 +0,0 @@ -/* - * Copyright (c) 2010 The WebM project authors. All Rights Reserved. - * - * Use of this source code is governed by a BSD-style license - * that can be found in the LICENSE file in the root of the source - * tree. An additional intellectual property rights grant can be found - * in the file PATENTS. All contributing project authors may - * be found in the AUTHORS file in the root of the source tree. - */ - - -#include "vpx_ports/config.h" -#include "blockd.h" -#include "pragmas.h" -#include "postproc.h" -#include "dboolhuff.h" -#include "dequantize.h" -#include "onyxd_int.h" - -void vp8_dmachine_specific_config(VP8D_COMP *pbi) -{ -#if CONFIG_RUNTIME_CPU_DETECT - pbi->mb.rtcd = &pbi->common.rtcd; -#if HAVE_ARMV7 - pbi->dequant.block = vp8_dequantize_b_neon; - pbi->dboolhuff.start = vp8dx_start_decode_c; - pbi->dboolhuff.fill = vp8dx_bool_decoder_fill_c; - pbi->dboolhuff.debool = vp8dx_decode_bool_c; - pbi->dboolhuff.devalue = vp8dx_decode_value_c; - -#elif HAVE_ARMV6 - pbi->dequant.block = vp8_dequantize_b_v6; - pbi->dboolhuff.start = vp8dx_start_decode_c; - pbi->dboolhuff.fill = vp8dx_bool_decoder_fill_c; - pbi->dboolhuff.debool = vp8dx_decode_bool_c; - pbi->dboolhuff.devalue = vp8dx_decode_value_c; -#endif -#endif -} diff --git a/vp8/decoder/generic/dsystemdependent.c b/vp8/decoder/generic/dsystemdependent.c index 60f2af5b8..84de7af43 100644 --- a/vp8/decoder/generic/dsystemdependent.c +++ b/vp8/decoder/generic/dsystemdependent.c @@ -14,6 +14,7 @@ #include "onyxd_int.h" extern void vp8_arch_x86_decode_init(VP8D_COMP *pbi); +extern void vp8_arch_arm_decode_init(VP8D_COMP *pbi); void vp8_dmachine_specific_config(VP8D_COMP *pbi) { @@ -37,4 +38,8 @@ void vp8_dmachine_specific_config(VP8D_COMP *pbi) #if ARCH_X86 || ARCH_X86_64 vp8_arch_x86_decode_init(pbi); #endif + +#if ARCH_ARM + vp8_arch_arm_decode_init(pbi); +#endif } diff --git a/vp8/decoder/onyxd_if.c b/vp8/decoder/onyxd_if.c index 884c38da0..b5a6e3e85 100644 --- a/vp8/decoder/onyxd_if.c +++ b/vp8/decoder/onyxd_if.c @@ -30,6 +30,9 @@ #include "systemdependent.h" #include "vpx_ports/vpx_timer.h" #include "detokenize.h" +#if ARCH_ARM +#include "vpx_ports/arm.h" +#endif extern void vp8_init_loop_filter(VP8_COMMON *cm); extern void vp8cx_init_de_quantizer(VP8D_COMP *pbi); @@ -224,7 +227,6 @@ int vp8dx_set_reference(VP8D_PTR ptr, VP8_REFFRAME ref_frame_flag, YV12_BUFFER_C #if HAVE_ARMV7 extern void vp8_push_neon(INT64 *store); extern void vp8_pop_neon(INT64 *store); -static INT64 dx_store_reg[8]; #endif static int get_free_fb (VP8_COMMON *cm) @@ -312,6 +314,9 @@ static int swap_frame_buffers (VP8_COMMON *cm) int vp8dx_receive_compressed_data(VP8D_PTR ptr, unsigned long size, const unsigned char *source, INT64 time_stamp) { +#if HAVE_ARMV7 + INT64 dx_store_reg[8]; +#endif VP8D_COMP *pbi = (VP8D_COMP *) ptr; VP8_COMMON *cm = &pbi->common; int retcode = 0; @@ -327,10 +332,27 @@ int vp8dx_receive_compressed_data(VP8D_PTR ptr, unsigned long size, const unsign pbi->common.error.error_code = VPX_CODEC_OK; +#if HAVE_ARMV7 +#if CONFIG_RUNTIME_CPU_DETECT + if (cm->rtcd.flags & HAS_NEON) +#endif + { + vp8_push_neon(dx_store_reg); + } +#endif + cm->new_fb_idx = get_free_fb (cm); if (setjmp(pbi->common.error.jmp)) { +#if HAVE_ARMV7 +#if CONFIG_RUNTIME_CPU_DETECT + if (cm->rtcd.flags & HAS_NEON) +#endif + { + vp8_pop_neon(dx_store_reg); + } +#endif pbi->common.error.setjmp = 0; if (cm->fb_idx_ref_cnt[cm->new_fb_idx] > 0) cm->fb_idx_ref_cnt[cm->new_fb_idx]--; @@ -339,10 +361,6 @@ int vp8dx_receive_compressed_data(VP8D_PTR ptr, unsigned long size, const unsign pbi->common.error.setjmp = 1; -#if HAVE_ARMV7 - vp8_push_neon(dx_store_reg); -#endif - vpx_usec_timer_start(&timer); //cm->current_video_frame++; @@ -354,7 +372,12 @@ int vp8dx_receive_compressed_data(VP8D_PTR ptr, unsigned long size, const unsign if (retcode < 0) { #if HAVE_ARMV7 - vp8_pop_neon(dx_store_reg); +#if CONFIG_RUNTIME_CPU_DETECT + if (cm->rtcd.flags & HAS_NEON) +#endif + { + vp8_pop_neon(dx_store_reg); + } #endif pbi->common.error.error_code = VPX_CODEC_ERROR; pbi->common.error.setjmp = 0; @@ -367,6 +390,14 @@ int vp8dx_receive_compressed_data(VP8D_PTR ptr, unsigned long size, const unsign { if (swap_frame_buffers (cm)) { +#if HAVE_ARMV7 +#if CONFIG_RUNTIME_CPU_DETECT + if (cm->rtcd.flags & HAS_NEON) +#endif + { + vp8_pop_neon(dx_store_reg); + } +#endif pbi->common.error.error_code = VPX_CODEC_ERROR; pbi->common.error.setjmp = 0; return -1; @@ -375,6 +406,14 @@ int vp8dx_receive_compressed_data(VP8D_PTR ptr, unsigned long size, const unsign { if (swap_frame_buffers (cm)) { +#if HAVE_ARMV7 +#if CONFIG_RUNTIME_CPU_DETECT + if (cm->rtcd.flags & HAS_NEON) +#endif + { + vp8_pop_neon(dx_store_reg); + } +#endif pbi->common.error.error_code = VPX_CODEC_ERROR; pbi->common.error.setjmp = 0; return -1; @@ -455,7 +494,12 @@ int vp8dx_receive_compressed_data(VP8D_PTR ptr, unsigned long size, const unsign #endif #if HAVE_ARMV7 - vp8_pop_neon(dx_store_reg); +#if CONFIG_RUNTIME_CPU_DETECT + if (cm->rtcd.flags & HAS_NEON) +#endif + { + vp8_pop_neon(dx_store_reg); + } #endif pbi->common.error.setjmp = 0; return retcode; diff --git a/vp8/encoder/arm/arm_csystemdependent.c b/vp8/encoder/arm/arm_csystemdependent.c new file mode 100644 index 000000000..8736fcf1d --- /dev/null +++ b/vp8/encoder/arm/arm_csystemdependent.c @@ -0,0 +1,136 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include "vpx_ports/config.h" +#include "vpx_ports/arm.h" +#include "variance.h" +#include "onyx_int.h" + +extern void (*vp8_yv12_copy_partial_frame_ptr)(YV12_BUFFER_CONFIG *src_ybc, YV12_BUFFER_CONFIG *dst_ybc, int Fraction); +extern void vp8_yv12_copy_partial_frame(YV12_BUFFER_CONFIG *src_ybc, YV12_BUFFER_CONFIG *dst_ybc, int Fraction); +extern void vpxyv12_copy_partial_frame_neon(YV12_BUFFER_CONFIG *src_ybc, YV12_BUFFER_CONFIG *dst_ybc, int Fraction); + +void vp8_arch_arm_encoder_init(VP8_COMP *cpi) +{ +#if CONFIG_RUNTIME_CPU_DETECT + int flags = cpi->common.rtcd.flags; + int has_edsp = flags & HAS_EDSP; + int has_media = flags & HAS_MEDIA; + int has_neon = flags & HAS_NEON; + +#if HAVE_ARMV6 + if (has_media) + { + /*cpi->rtcd.variance.sad16x16 = vp8_sad16x16_c; + cpi->rtcd.variance.sad16x8 = vp8_sad16x8_c; + cpi->rtcd.variance.sad8x16 = vp8_sad8x16_c; + cpi->rtcd.variance.sad8x8 = vp8_sad8x8_c; + cpi->rtcd.variance.sad4x4 = vp8_sad4x4_c;*/ + + /*cpi->rtcd.variance.var4x4 = vp8_variance4x4_c; + cpi->rtcd.variance.var8x8 = vp8_variance8x8_c; + cpi->rtcd.variance.var8x16 = vp8_variance8x16_c; + cpi->rtcd.variance.var16x8 = vp8_variance16x8_c; + cpi->rtcd.variance.var16x16 = vp8_variance16x16_c;*/ + + /*cpi->rtcd.variance.subpixvar4x4 = vp8_sub_pixel_variance4x4_c; + cpi->rtcd.variance.subpixvar8x8 = vp8_sub_pixel_variance8x8_c; + cpi->rtcd.variance.subpixvar8x16 = vp8_sub_pixel_variance8x16_c; + cpi->rtcd.variance.subpixvar16x8 = vp8_sub_pixel_variance16x8_c; + cpi->rtcd.variance.subpixvar16x16 = vp8_sub_pixel_variance16x16_c;*/ + + /*cpi->rtcd.variance.mse16x16 = vp8_mse16x16_c; + cpi->rtcd.variance.getmbss = vp8_get_mb_ss_c;*/ + + /*cpi->rtcd.variance.get16x16prederror = vp8_get16x16pred_error_c; + cpi->rtcd.variance.get8x8var = vp8_get8x8var_c; + cpi->rtcd.variance.get16x16var = vp8_get16x16var_c;; + cpi->rtcd.variance.get4x4sse_cs = vp8_get4x4sse_cs_c;*/ + + /*cpi->rtcd.fdct.short4x4 = vp8_short_fdct4x4_c; + cpi->rtcd.fdct.short8x4 = vp8_short_fdct8x4_c; + cpi->rtcd.fdct.fast4x4 = vp8_fast_fdct4x4_c; + cpi->rtcd.fdct.fast8x4 = vp8_fast_fdct8x4_c;*/ + cpi->rtcd.fdct.walsh_short4x4 = vp8_short_walsh4x4_armv6; + + /*cpi->rtcd.encodemb.berr = vp8_block_error_c; + cpi->rtcd.encodemb.mberr = vp8_mbblock_error_c; + cpi->rtcd.encodemb.mbuverr = vp8_mbuverror_c; + cpi->rtcd.encodemb.subb = vp8_subtract_b_c; + cpi->rtcd.encodemb.submby = vp8_subtract_mby_c; + cpi->rtcd.encodemb.submbuv = vp8_subtract_mbuv_c;*/ + + /*cpi->rtcd.quantize.quantb = vp8_regular_quantize_b; + cpi->rtcd.quantize.fastquantb = vp8_fast_quantize_b_c;*/ + } +#endif + +#if HAVE_ARMV7 + if (has_neon) + { + cpi->rtcd.variance.sad16x16 = vp8_sad16x16_neon; + cpi->rtcd.variance.sad16x8 = vp8_sad16x8_neon; + cpi->rtcd.variance.sad8x16 = vp8_sad8x16_neon; + cpi->rtcd.variance.sad8x8 = vp8_sad8x8_neon; + cpi->rtcd.variance.sad4x4 = vp8_sad4x4_neon; + + /*cpi->rtcd.variance.var4x4 = vp8_variance4x4_c;*/ + cpi->rtcd.variance.var8x8 = vp8_variance8x8_neon; + cpi->rtcd.variance.var8x16 = vp8_variance8x16_neon; + cpi->rtcd.variance.var16x8 = vp8_variance16x8_neon; + cpi->rtcd.variance.var16x16 = vp8_variance16x16_neon; + + /*cpi->rtcd.variance.subpixvar4x4 = vp8_sub_pixel_variance4x4_c;*/ + cpi->rtcd.variance.subpixvar8x8 = vp8_sub_pixel_variance8x8_neon; + /*cpi->rtcd.variance.subpixvar8x16 = vp8_sub_pixel_variance8x16_c; + cpi->rtcd.variance.subpixvar16x8 = vp8_sub_pixel_variance16x8_c;*/ + cpi->rtcd.variance.subpixvar16x16 = vp8_sub_pixel_variance16x16_neon; + + cpi->rtcd.variance.mse16x16 = vp8_mse16x16_neon; + /*cpi->rtcd.variance.getmbss = vp8_get_mb_ss_c;*/ + + cpi->rtcd.variance.get16x16prederror = vp8_get16x16pred_error_neon; + /*cpi->rtcd.variance.get8x8var = vp8_get8x8var_c; + cpi->rtcd.variance.get16x16var = vp8_get16x16var_c;*/ + cpi->rtcd.variance.get4x4sse_cs = vp8_get4x4sse_cs_neon; + + cpi->rtcd.fdct.short4x4 = vp8_short_fdct4x4_neon; + cpi->rtcd.fdct.short8x4 = vp8_short_fdct8x4_neon; + cpi->rtcd.fdct.fast4x4 = vp8_fast_fdct4x4_neon; + cpi->rtcd.fdct.fast8x4 = vp8_fast_fdct8x4_neon; + cpi->rtcd.fdct.walsh_short4x4 = vp8_short_walsh4x4_neon; + + /*cpi->rtcd.encodemb.berr = vp8_block_error_c; + cpi->rtcd.encodemb.mberr = vp8_mbblock_error_c; + cpi->rtcd.encodemb.mbuverr = vp8_mbuverror_c;*/ + cpi->rtcd.encodemb.subb = vp8_subtract_b_neon; + cpi->rtcd.encodemb.submby = vp8_subtract_mby_neon; + cpi->rtcd.encodemb.submbuv = vp8_subtract_mbuv_neon; + + /*cpi->rtcd.quantize.quantb = vp8_regular_quantize_b; + cpi->rtcd.quantize.fastquantb = vp8_fast_quantize_b_c;*/ + /* The neon quantizer has not been updated to match the new exact + * quantizer introduced in commit e04e2935 + */ + /*cpi->rtcd.quantize.fastquantb = vp8_fast_quantize_b_neon;*/ + } +#endif + +#if HAVE_ARMV7 +#if CONFIG_RUNTIME_CPU_DETECT + if (has_neon) +#endif + { + vp8_yv12_copy_partial_frame_ptr = vpxyv12_copy_partial_frame_neon; + } +#endif +#endif +} diff --git a/vp8/encoder/arm/neon/boolhuff_armv7.asm b/vp8/encoder/arm/armv5te/boolhuff_armv5te.asm index 9c4823c51..e78dc3322 100644 --- a/vp8/encoder/arm/neon/boolhuff_armv7.asm +++ b/vp8/encoder/arm/armv5te/boolhuff_armv5te.asm @@ -205,17 +205,10 @@ token_count_lt_zero_se ldr r5, [r0, #vp8_writer_range] ldr r3, [r0, #vp8_writer_count] - ; reverse the stream of bits to be packed. Normally - ; the most significant bit is peeled off and compared - ; in the form of (v >> --n) & 1. ARM architecture has - ; the ability to set a flag based on the value of the - ; bit shifted off the bottom of the register. To make - ; that happen the bitstream is reversed. - rbit r11, r1 rsb r4, r10, #32 ; 32-n ; v is kept in r1 during the token pack loop - lsr r1, r11, r4 ; v >>= 32 - n + lsl r1, r1, r4 ; r1 = v << 32 - n encode_value_loop sub r7, r5, #1 ; range-1 @@ -223,7 +216,7 @@ encode_value_loop ; Decisions are made based on the bit value shifted ; off of v, so set a flag here based on this. ; This value is refered to as "bb" - lsrs r1, r1, #1 ; bit = v >> n + lsls r1, r1, #1 ; bit = v >> n mov r4, r7, lsl #7 ; ((range-1) * 128) mov r7, #1 diff --git a/vp8/encoder/arm/neon/vp8_packtokens_armv7.asm b/vp8/encoder/arm/armv5te/vp8_packtokens_armv5.asm index c19ac8250..3233d2a96 100644 --- a/vp8/encoder/arm/neon/vp8_packtokens_armv7.asm +++ b/vp8/encoder/arm/armv5te/vp8_packtokens_armv5.asm @@ -9,7 +9,7 @@ ; - EXPORT |vp8cx_pack_tokens_armv7| + EXPORT |vp8cx_pack_tokens_armv5| INCLUDE vpx_vp8_enc_asm_offsets.asm @@ -25,7 +25,7 @@ ; r3 vp8_coef_encodings ; s0 vp8_extra_bits ; s1 vp8_coef_tree -|vp8cx_pack_tokens_armv7| PROC +|vp8cx_pack_tokens_armv5| PROC push {r4-r11, lr} ; Add size of xcount * sizeof (TOKENEXTRA) to get stop @@ -57,18 +57,11 @@ while_p_lt_stop movne lr, #2 ; i = 2 subne r8, r8, #1 ; --n - ; reverse the stream of bits to be packed. Normally - ; the most significant bit is peeled off and compared - ; in the form of (v >> --n) & 1. ARM architecture has - ; the ability to set a flag based on the value of the - ; bit shifted off the bottom of the register. To make - ; that happen the bitstream is reversed. - rbit r12, r6 rsb r4, r8, #32 ; 32-n ldr r10, [sp, #52] ; vp8_coef_tree ; v is kept in r12 during the token pack loop - lsr r12, r12, r4 ; v >>= 32 - n + lsl r12, r6, r4 ; r12 = v << 32 - n ; loop start token_loop @@ -78,7 +71,7 @@ token_loop ; Decisions are made based on the bit value shifted ; off of v, so set a flag here based on this. ; This value is refered to as "bb" - lsrs r12, r12, #1 ; bb = v >> n + lsls r12, r12, #1 ; bb = v >> n mul r4, r4, r7 ; ((range-1) * pp[i>>1])) ; bb can only be 0 or 1. So only execute this statement @@ -172,16 +165,15 @@ token_count_lt_zero ldr r10, [r12, #vp8_extra_bit_struct_tree] str r10, [sp, #4] ; b->tree - rbit r12, r7 ; reverse v rsb r4, r8, #32 - lsr r12, r12, r4 + lsl r12, r7, r4 mov lr, #0 ; i = 0 extra_bits_loop ldrb r4, [r9, lr, asr #1] ; pp[i>>1] sub r7, r5, #1 ; range-1 - lsrs r12, r12, #1 ; v >> n + lsls r12, r12, #1 ; v >> n mul r4, r4, r7 ; (range-1) * pp[i>>1] addcs lr, lr, #1 ; i + bb diff --git a/vp8/encoder/arm/neon/vp8_packtokens_mbrow_armv7.asm b/vp8/encoder/arm/armv5te/vp8_packtokens_mbrow_armv5.asm index 075645586..a9b552ae1 100644 --- a/vp8/encoder/arm/neon/vp8_packtokens_mbrow_armv7.asm +++ b/vp8/encoder/arm/armv5te/vp8_packtokens_mbrow_armv5.asm @@ -9,7 +9,7 @@ ; - EXPORT |vp8cx_pack_mb_row_tokens_armv7| + EXPORT |vp8cx_pack_mb_row_tokens_armv5| INCLUDE vpx_vp8_enc_asm_offsets.asm @@ -25,7 +25,7 @@ ; r3 vp8_extra_bits ; s0 vp8_coef_tree -|vp8cx_pack_mb_row_tokens_armv7| PROC +|vp8cx_pack_mb_row_tokens_armv5| PROC push {r4-r11, lr} sub sp, sp, #24 @@ -78,18 +78,11 @@ while_p_lt_stop movne lr, #2 ; i = 2 subne r8, r8, #1 ; --n - ; reverse the stream of bits to be packed. Normally - ; the most significant bit is peeled off and compared - ; in the form of (v >> --n) & 1. ARM architecture has - ; the ability to set a flag based on the value of the - ; bit shifted off the bottom of the register. To make - ; that happen the bitstream is reversed. - rbit r12, r6 rsb r4, r8, #32 ; 32-n ldr r10, [sp, #60] ; vp8_coef_tree ; v is kept in r12 during the token pack loop - lsr r12, r12, r4 ; v >>= 32 - n + lsl r12, r6, r4 ; r12 = v << 32 - n ; loop start token_loop @@ -99,7 +92,7 @@ token_loop ; Decisions are made based on the bit value shifted ; off of v, so set a flag here based on this. ; This value is refered to as "bb" - lsrs r12, r12, #1 ; bb = v >> n + lsls r12, r12, #1 ; bb = v >> n mul r4, r4, r7 ; ((range-1) * pp[i>>1])) ; bb can only be 0 or 1. So only execute this statement @@ -193,16 +186,15 @@ token_count_lt_zero ldr r10, [r12, #vp8_extra_bit_struct_tree] str r10, [sp, #4] ; b->tree - rbit r12, r7 ; reverse v rsb r4, r8, #32 - lsr r12, r12, r4 + lsl r12, r7, r4 mov lr, #0 ; i = 0 extra_bits_loop ldrb r4, [r9, lr, asr #1] ; pp[i>>1] sub r7, r5, #1 ; range-1 - lsrs r12, r12, #1 ; v >> n + lsls r12, r12, #1 ; v >> n mul r4, r4, r7 ; (range-1) * pp[i>>1] addcs lr, lr, #1 ; i + bb diff --git a/vp8/encoder/arm/neon/vp8_packtokens_partitions_armv7.asm b/vp8/encoder/arm/armv5te/vp8_packtokens_partitions_armv5.asm index 10a3d9851..0835164e5 100644 --- a/vp8/encoder/arm/neon/vp8_packtokens_partitions_armv7.asm +++ b/vp8/encoder/arm/armv5te/vp8_packtokens_partitions_armv5.asm @@ -9,7 +9,7 @@ ; - EXPORT |vp8cx_pack_tokens_into_partitions_armv7| + EXPORT |vp8cx_pack_tokens_into_partitions_armv5| INCLUDE vpx_vp8_enc_asm_offsets.asm @@ -27,7 +27,7 @@ ; s1 vp8_extra_bits, ; s2 const vp8_tree_index *, -|vp8cx_pack_tokens_into_partitions_armv7| PROC +|vp8cx_pack_tokens_into_partitions_armv5| PROC push {r4-r11, lr} sub sp, sp, #44 @@ -106,18 +106,11 @@ while_p_lt_stop movne lr, #2 ; i = 2 subne r8, r8, #1 ; --n - ; reverse the stream of bits to be packed. Normally - ; the most significant bit is peeled off and compared - ; in the form of (v >> --n) & 1. ARM architecture has - ; the ability to set a flag based on the value of the - ; bit shifted off the bottom of the register. To make - ; that happen the bitstream is reversed. - rbit r12, r6 rsb r4, r8, #32 ; 32-n ldr r10, [sp, #88] ; vp8_coef_tree ; v is kept in r12 during the token pack loop - lsr r12, r12, r4 ; v >>= 32 - n + lsl r12, r6, r4 ; r12 = v << 32 - n ; loop start token_loop @@ -127,7 +120,7 @@ token_loop ; Decisions are made based on the bit value shifted ; off of v, so set a flag here based on this. ; This value is refered to as "bb" - lsrs r12, r12, #1 ; bb = v >> n + lsls r12, r12, #1 ; bb = v >> n mul r4, r4, r7 ; ((range-1) * pp[i>>1])) ; bb can only be 0 or 1. So only execute this statement @@ -221,16 +214,15 @@ token_count_lt_zero ldr r10, [r12, #vp8_extra_bit_struct_tree] str r10, [sp, #4] ; b->tree - rbit r12, r7 ; reverse v rsb r4, r8, #32 - lsr r12, r12, r4 + lsl r12, r7, r4 mov lr, #0 ; i = 0 extra_bits_loop ldrb r4, [r9, lr, asr #1] ; pp[i>>1] sub r7, r5, #1 ; range-1 - lsrs r12, r12, #1 ; v >> n + lsls r12, r12, #1 ; v >> n mul r4, r4, r7 ; (range-1) * pp[i>>1] addcs lr, lr, #1 ; i + bb diff --git a/vp8/encoder/arm/csystemdependent.c b/vp8/encoder/arm/csystemdependent.c deleted file mode 100644 index 8d70d635a..000000000 --- a/vp8/encoder/arm/csystemdependent.c +++ /dev/null @@ -1,164 +0,0 @@ -/* - * Copyright (c) 2010 The WebM project authors. All Rights Reserved. - * - * Use of this source code is governed by a BSD-style license - * that can be found in the LICENSE file in the root of the source - * tree. An additional intellectual property rights grant can be found - * in the file PATENTS. All contributing project authors may - * be found in the AUTHORS file in the root of the source tree. - */ - - -#include "vpx_ports/config.h" -#include "variance.h" -#include "onyx_int.h" - -void (*vp8_yv12_copy_partial_frame_ptr)(YV12_BUFFER_CONFIG *src_ybc, YV12_BUFFER_CONFIG *dst_ybc, int Fraction); -extern void vp8_yv12_copy_partial_frame(YV12_BUFFER_CONFIG *src_ybc, YV12_BUFFER_CONFIG *dst_ybc, int Fraction); -extern void vpxyv12_copy_partial_frame_neon(YV12_BUFFER_CONFIG *src_ybc, YV12_BUFFER_CONFIG *dst_ybc, int Fraction); - -void vp8_cmachine_specific_config(VP8_COMP *cpi) -{ -#if CONFIG_RUNTIME_CPU_DETECT - cpi->rtcd.common = &cpi->common.rtcd; - -#if HAVE_ARMV7 - cpi->rtcd.variance.sad16x16 = vp8_sad16x16_neon; - cpi->rtcd.variance.sad16x8 = vp8_sad16x8_neon; - cpi->rtcd.variance.sad8x16 = vp8_sad8x16_neon; - cpi->rtcd.variance.sad8x8 = vp8_sad8x8_neon; - cpi->rtcd.variance.sad4x4 = vp8_sad4x4_neon; - - cpi->rtcd.variance.var4x4 = vp8_variance4x4_c; - cpi->rtcd.variance.var8x8 = vp8_variance8x8_neon; - cpi->rtcd.variance.var8x16 = vp8_variance8x16_neon; - cpi->rtcd.variance.var16x8 = vp8_variance16x8_neon; - cpi->rtcd.variance.var16x16 = vp8_variance16x16_neon; - - cpi->rtcd.variance.subpixvar4x4 = vp8_sub_pixel_variance4x4_c; - cpi->rtcd.variance.subpixvar8x8 = vp8_sub_pixel_variance8x8_neon; - cpi->rtcd.variance.subpixvar8x16 = vp8_sub_pixel_variance8x16_c; - cpi->rtcd.variance.subpixvar16x8 = vp8_sub_pixel_variance16x8_c; - cpi->rtcd.variance.subpixvar16x16 = vp8_sub_pixel_variance16x16_neon; - - cpi->rtcd.variance.mse16x16 = vp8_mse16x16_neon; - cpi->rtcd.variance.getmbss = vp8_get_mb_ss_c; - - cpi->rtcd.variance.get16x16prederror = vp8_get16x16pred_error_neon; - cpi->rtcd.variance.get8x8var = vp8_get8x8var_c; - cpi->rtcd.variance.get16x16var = vp8_get16x16var_c;; - cpi->rtcd.variance.get4x4sse_cs = vp8_get4x4sse_cs_neon; - - cpi->rtcd.fdct.short4x4 = vp8_short_fdct4x4_neon; - cpi->rtcd.fdct.short8x4 = vp8_short_fdct8x4_neon; - cpi->rtcd.fdct.fast4x4 = vp8_fast_fdct4x4_neon; - cpi->rtcd.fdct.fast8x4 = vp8_fast_fdct8x4_neon; - cpi->rtcd.fdct.walsh_short4x4 = vp8_short_walsh4x4_neon; - - cpi->rtcd.encodemb.berr = vp8_block_error_c; - cpi->rtcd.encodemb.mberr = vp8_mbblock_error_c; - cpi->rtcd.encodemb.mbuverr = vp8_mbuverror_c; - cpi->rtcd.encodemb.subb = vp8_subtract_b_neon; - cpi->rtcd.encodemb.submby = vp8_subtract_mby_neon; - cpi->rtcd.encodemb.submbuv = vp8_subtract_mbuv_neon; - - cpi->rtcd.quantize.quantb = vp8_regular_quantize_b; - cpi->rtcd.quantize.fastquantb = vp8_fast_quantize_b_c; - /* The neon quantizer has not been updated to match the new exact - * quantizer introduced in commit e04e2935 - */ - /*cpi->rtcd.quantize.fastquantb = vp8_fast_quantize_b_neon;*/ -#elif HAVE_ARMV6 - cpi->rtcd.variance.sad16x16 = vp8_sad16x16_c; - cpi->rtcd.variance.sad16x8 = vp8_sad16x8_c; - cpi->rtcd.variance.sad8x16 = vp8_sad8x16_c; - cpi->rtcd.variance.sad8x8 = vp8_sad8x8_c; - cpi->rtcd.variance.sad4x4 = vp8_sad4x4_c; - - cpi->rtcd.variance.var4x4 = vp8_variance4x4_c; - cpi->rtcd.variance.var8x8 = vp8_variance8x8_c; - cpi->rtcd.variance.var8x16 = vp8_variance8x16_c; - cpi->rtcd.variance.var16x8 = vp8_variance16x8_c; - cpi->rtcd.variance.var16x16 = vp8_variance16x16_c; - - cpi->rtcd.variance.subpixvar4x4 = vp8_sub_pixel_variance4x4_c; - cpi->rtcd.variance.subpixvar8x8 = vp8_sub_pixel_variance8x8_c; - cpi->rtcd.variance.subpixvar8x16 = vp8_sub_pixel_variance8x16_c; - cpi->rtcd.variance.subpixvar16x8 = vp8_sub_pixel_variance16x8_c; - cpi->rtcd.variance.subpixvar16x16 = vp8_sub_pixel_variance16x16_c; - - cpi->rtcd.variance.mse16x16 = vp8_mse16x16_c; - cpi->rtcd.variance.getmbss = vp8_get_mb_ss_c; - - cpi->rtcd.variance.get16x16prederror = vp8_get16x16pred_error_c; - cpi->rtcd.variance.get8x8var = vp8_get8x8var_c; - cpi->rtcd.variance.get16x16var = vp8_get16x16var_c;; - cpi->rtcd.variance.get4x4sse_cs = vp8_get4x4sse_cs_c; - - cpi->rtcd.fdct.short4x4 = vp8_short_fdct4x4_c; - cpi->rtcd.fdct.short8x4 = vp8_short_fdct8x4_c; - cpi->rtcd.fdct.fast4x4 = vp8_fast_fdct4x4_c; - cpi->rtcd.fdct.fast8x4 = vp8_fast_fdct8x4_c; - cpi->rtcd.fdct.walsh_short4x4 = vp8_short_walsh4x4_armv6; - - cpi->rtcd.encodemb.berr = vp8_block_error_c; - cpi->rtcd.encodemb.mberr = vp8_mbblock_error_c; - cpi->rtcd.encodemb.mbuverr = vp8_mbuverror_c; - cpi->rtcd.encodemb.subb = vp8_subtract_b_c; - cpi->rtcd.encodemb.submby = vp8_subtract_mby_c; - cpi->rtcd.encodemb.submbuv = vp8_subtract_mbuv_c; - - cpi->rtcd.quantize.quantb = vp8_regular_quantize_b; - cpi->rtcd.quantize.fastquantb = vp8_fast_quantize_b_c; -#else - //pure c - cpi->rtcd.variance.sad16x16 = vp8_sad16x16_c; - cpi->rtcd.variance.sad16x8 = vp8_sad16x8_c; - cpi->rtcd.variance.sad8x16 = vp8_sad8x16_c; - cpi->rtcd.variance.sad8x8 = vp8_sad8x8_c; - cpi->rtcd.variance.sad4x4 = vp8_sad4x4_c; - - cpi->rtcd.variance.var4x4 = vp8_variance4x4_c; - cpi->rtcd.variance.var8x8 = vp8_variance8x8_c; - cpi->rtcd.variance.var8x16 = vp8_variance8x16_c; - cpi->rtcd.variance.var16x8 = vp8_variance16x8_c; - cpi->rtcd.variance.var16x16 = vp8_variance16x16_c; - - cpi->rtcd.variance.subpixvar4x4 = vp8_sub_pixel_variance4x4_c; - cpi->rtcd.variance.subpixvar8x8 = vp8_sub_pixel_variance8x8_c; - cpi->rtcd.variance.subpixvar8x16 = vp8_sub_pixel_variance8x16_c; - cpi->rtcd.variance.subpixvar16x8 = vp8_sub_pixel_variance16x8_c; - cpi->rtcd.variance.subpixvar16x16 = vp8_sub_pixel_variance16x16_c; - - cpi->rtcd.variance.mse16x16 = vp8_mse16x16_c; - cpi->rtcd.variance.getmbss = vp8_get_mb_ss_c; - - cpi->rtcd.variance.get16x16prederror = vp8_get16x16pred_error_c; - cpi->rtcd.variance.get8x8var = vp8_get8x8var_c; - cpi->rtcd.variance.get16x16var = vp8_get16x16var_c;; - cpi->rtcd.variance.get4x4sse_cs = vp8_get4x4sse_cs_c; - - cpi->rtcd.fdct.short4x4 = vp8_short_fdct4x4_c; - cpi->rtcd.fdct.short8x4 = vp8_short_fdct8x4_c; - cpi->rtcd.fdct.fast4x4 = vp8_fast_fdct4x4_c; - cpi->rtcd.fdct.fast8x4 = vp8_fast_fdct8x4_c; - cpi->rtcd.fdct.walsh_short4x4 = vp8_short_walsh4x4_c; - - cpi->rtcd.encodemb.berr = vp8_block_error_c; - cpi->rtcd.encodemb.mberr = vp8_mbblock_error_c; - cpi->rtcd.encodemb.mbuverr = vp8_mbuverror_c; - cpi->rtcd.encodemb.subb = vp8_subtract_b_c; - cpi->rtcd.encodemb.submby = vp8_subtract_mby_c; - cpi->rtcd.encodemb.submbuv = vp8_subtract_mbuv_c; - - cpi->rtcd.quantize.quantb = vp8_regular_quantize_b; - cpi->rtcd.quantize.fastquantb = vp8_fast_quantize_b_c; -#endif -#endif - -#if HAVE_ARMV7 - vp8_yv12_copy_partial_frame_ptr = vpxyv12_copy_partial_frame_neon; -#else - vp8_yv12_copy_partial_frame_ptr = vp8_yv12_copy_partial_frame; -#endif -} diff --git a/vp8/encoder/arm/dct_arm.h b/vp8/encoder/arm/dct_arm.h index 774599bf0..41fa5d192 100644 --- a/vp8/encoder/arm/dct_arm.h +++ b/vp8/encoder/arm/dct_arm.h @@ -15,9 +15,11 @@ #if HAVE_ARMV6 extern prototype_fdct(vp8_short_walsh4x4_armv6); +#if !CONFIG_RUNTIME_CPU_DETECT #undef vp8_fdct_walsh_short4x4 #define vp8_fdct_walsh_short4x4 vp8_short_walsh4x4_armv6 #endif +#endif #if HAVE_ARMV7 extern prototype_fdct(vp8_short_fdct4x4_neon); @@ -26,6 +28,7 @@ extern prototype_fdct(vp8_fast_fdct4x4_neon); extern prototype_fdct(vp8_fast_fdct8x4_neon); extern prototype_fdct(vp8_short_walsh4x4_neon); +#if !CONFIG_RUNTIME_CPU_DETECT #undef vp8_fdct_short4x4 #define vp8_fdct_short4x4 vp8_short_fdct4x4_neon @@ -40,6 +43,7 @@ extern prototype_fdct(vp8_short_walsh4x4_neon); #undef vp8_fdct_walsh_short4x4 #define vp8_fdct_walsh_short4x4 vp8_short_walsh4x4_neon +#endif #endif diff --git a/vp8/encoder/arm/encodemb_arm.h b/vp8/encoder/arm/encodemb_arm.h index eb699433f..8fe453735 100644 --- a/vp8/encoder/arm/encodemb_arm.h +++ b/vp8/encoder/arm/encodemb_arm.h @@ -30,6 +30,7 @@ extern prototype_submbuv(vp8_subtract_mbuv_neon); //#undef vp8_encodemb_mbuverr //#define vp8_encodemb_mbuverr vp8_mbuverror_c +#if !CONFIG_RUNTIME_CPU_DETECT #undef vp8_encodemb_subb #define vp8_encodemb_subb vp8_subtract_b_neon @@ -38,6 +39,7 @@ extern prototype_submbuv(vp8_subtract_mbuv_neon); #undef vp8_encodemb_submbuv #define vp8_encodemb_submbuv vp8_subtract_mbuv_neon +#endif #endif diff --git a/vp8/encoder/arm/variance_arm.h b/vp8/encoder/arm/variance_arm.h index 859e43f51..fb9dd5a5b 100644 --- a/vp8/encoder/arm/variance_arm.h +++ b/vp8/encoder/arm/variance_arm.h @@ -38,6 +38,7 @@ extern prototype_sad(vp8_get16x16pred_error_neon); //extern prototype_variance2(vp8_get16x16var_c); extern prototype_sad(vp8_get4x4sse_cs_neon); +#if !CONFIG_RUNTIME_CPU_DETECT #undef vp8_variance_sad4x4 #define vp8_variance_sad4x4 vp8_sad4x4_neon @@ -100,6 +101,7 @@ extern prototype_sad(vp8_get4x4sse_cs_neon); #undef vp8_variance_get4x4sse_cs #define vp8_variance_get4x4sse_cs vp8_get4x4sse_cs_neon +#endif #endif diff --git a/vp8/encoder/bitstream.h b/vp8/encoder/bitstream.h index 559631338..f5d148ea4 100644 --- a/vp8/encoder/bitstream.h +++ b/vp8/encoder/bitstream.h @@ -12,25 +12,25 @@ #ifndef __INC_BITSTREAM_H #define __INC_BITSTREAM_H -#if HAVE_ARMV7 -void vp8cx_pack_tokens_armv7(vp8_writer *w, const TOKENEXTRA *p, int xcount, +#if HAVE_ARMV5TE +void vp8cx_pack_tokens_armv5(vp8_writer *w, const TOKENEXTRA *p, int xcount, vp8_token *, vp8_extra_bit_struct *, const vp8_tree_index *); -void vp8cx_pack_tokens_into_partitions_armv7(VP8_COMP *, unsigned char *, int , int *, +void vp8cx_pack_tokens_into_partitions_armv5(VP8_COMP *, unsigned char *, int , int *, vp8_token *, vp8_extra_bit_struct *, const vp8_tree_index *); -void vp8cx_pack_mb_row_tokens_armv7(VP8_COMP *cpi, vp8_writer *w, +void vp8cx_pack_mb_row_tokens_armv5(VP8_COMP *cpi, vp8_writer *w, vp8_token *, vp8_extra_bit_struct *, const vp8_tree_index *); # define pack_tokens(a,b,c) \ - vp8cx_pack_tokens_armv7(a,b,c,vp8_coef_encodings,vp8_extra_bits,vp8_coef_tree) + vp8cx_pack_tokens_armv5(a,b,c,vp8_coef_encodings,vp8_extra_bits,vp8_coef_tree) # define pack_tokens_into_partitions(a,b,c,d) \ - vp8cx_pack_tokens_into_partitions_armv7(a,b,c,d,vp8_coef_encodings,vp8_extra_bits,vp8_coef_tree) + vp8cx_pack_tokens_into_partitions_armv5(a,b,c,d,vp8_coef_encodings,vp8_extra_bits,vp8_coef_tree) # define pack_mb_row_tokens(a,b) \ - vp8cx_pack_mb_row_tokens_armv7(a,b,vp8_coef_encodings,vp8_extra_bits,vp8_coef_tree) + vp8cx_pack_mb_row_tokens_armv5(a,b,vp8_coef_encodings,vp8_extra_bits,vp8_coef_tree) #else # define pack_tokens(a,b,c) pack_tokens_c(a,b,c) # define pack_tokens_into_partitions(a,b,c,d) pack_tokens_into_partitions_c(a,b,c,d) diff --git a/vp8/encoder/generic/csystemdependent.c b/vp8/encoder/generic/csystemdependent.c index 1acb73d9c..520b08f51 100644 --- a/vp8/encoder/generic/csystemdependent.c +++ b/vp8/encoder/generic/csystemdependent.c @@ -15,6 +15,7 @@ void vp8_arch_x86_encoder_init(VP8_COMP *cpi); +void vp8_arch_arm_encoder_init(VP8_COMP *cpi); void (*vp8_fast_quantize_b)(BLOCK *b, BLOCKD *d); @@ -94,4 +95,8 @@ void vp8_cmachine_specific_config(VP8_COMP *cpi) vp8_arch_x86_encoder_init(cpi); #endif +#if ARCH_ARM + vp8_arch_arm_encoder_init(cpi); +#endif + } diff --git a/vp8/encoder/onyx_if.c b/vp8/encoder/onyx_if.c index 53d68be52..7e1583dd9 100644 --- a/vp8/encoder/onyx_if.c +++ b/vp8/encoder/onyx_if.c @@ -31,6 +31,9 @@ #include "vpx_ports/vpx_timer.h" #include "vpxerrors.h" #include "temporal_filter.h" +#if ARCH_ARM +#include "vpx_ports/arm.h" +#endif #include <math.h> #include <stdio.h> @@ -2106,8 +2109,8 @@ VP8_PTR vp8_create_compressor(VP8_CONFIG *oxcf) CHECK_MEM_ERROR(cpi->rdtok, vpx_calloc(256 * 3 / 2, sizeof(TOKENEXTRA))); CHECK_MEM_ERROR(cpi->mb.ss, vpx_calloc(sizeof(search_site), (MAX_MVSEARCH_STEPS * 8) + 1)); - vp8_cmachine_specific_config(cpi); vp8_create_common(&cpi->common); + vp8_cmachine_specific_config(cpi); vp8_init_config((VP8_PTR)cpi, oxcf); @@ -2852,9 +2855,20 @@ static void scale_and_extend_source(YV12_BUFFER_CONFIG *sd, VP8_COMP *cpi) { //vp8_yv12_copy_frame_ptr(sd, &cpi->scaled_source); #if HAVE_ARMV7 - vp8_yv12_copy_src_frame_func_neon(sd, &cpi->scaled_source); -#else - vp8_yv12_copy_frame_ptr(sd, &cpi->scaled_source); +#if CONFIG_RUNTIME_CPU_DETECT + if (cm->rtcd.flags & HAS_NEON) +#endif + { + vp8_yv12_copy_src_frame_func_neon(sd, &cpi->scaled_source); + } +#if CONFIG_RUNTIME_CPU_DETECT + else +#endif +#endif +#if !HAVE_ARMV7 || CONFIG_RUNTIME_CPU_DETECT + { + vp8_yv12_copy_frame_ptr(sd, &cpi->scaled_source); + } #endif cpi->Source = &cpi->scaled_source; @@ -4624,10 +4638,10 @@ static void Pass2Encode(VP8_COMP *cpi, unsigned long *size, unsigned char *dest, #if HAVE_ARMV7 extern void vp8_push_neon(INT64 *store); extern void vp8_pop_neon(INT64 *store); -static INT64 store_reg[8]; #endif int vp8_receive_raw_frame(VP8_PTR ptr, unsigned int frame_flags, YV12_BUFFER_CONFIG *sd, INT64 time_stamp, INT64 end_time) { + INT64 store_reg[8]; VP8_COMP *cpi = (VP8_COMP *) ptr; VP8_COMMON *cm = &cpi->common; struct vpx_usec_timer timer; @@ -4636,7 +4650,12 @@ int vp8_receive_raw_frame(VP8_PTR ptr, unsigned int frame_flags, YV12_BUFFER_CON return -1; #if HAVE_ARMV7 - vp8_push_neon(store_reg); +#if CONFIG_RUNTIME_CPU_DETECT + if (cm->rtcd.flags & HAS_NEON) +#endif + { + vp8_push_neon(store_reg); + } #endif vpx_usec_timer_start(&timer); @@ -4645,7 +4664,12 @@ int vp8_receive_raw_frame(VP8_PTR ptr, unsigned int frame_flags, YV12_BUFFER_CON if (cpi->source_buffer_count != 0 && cpi->source_buffer_count >= cpi->oxcf.lag_in_frames) { #if HAVE_ARMV7 - vp8_pop_neon(store_reg); +#if CONFIG_RUNTIME_CPU_DETECT + if (cm->rtcd.flags & HAS_NEON) +#endif + { + vp8_pop_neon(store_reg); + } #endif return -1; } @@ -4686,9 +4710,20 @@ int vp8_receive_raw_frame(VP8_PTR ptr, unsigned int frame_flags, YV12_BUFFER_CON s->source_time_stamp = time_stamp; s->source_frame_flags = frame_flags; #if HAVE_ARMV7 - vp8_yv12_copy_src_frame_func_neon(sd, &s->source_buffer); -#else - vp8_yv12_copy_frame_ptr(sd, &s->source_buffer); +#if CONFIG_RUNTIME_CPU_DETECT + if (cm->rtcd.flags & HAS_NEON) +#endif + { + vp8_yv12_copy_src_frame_func_neon(sd, &s->source_buffer); + } +#if CONFIG_RUNTIME_CPU_DETECT + else +#endif +#endif +#if !HAVE_ARMV7 || CONFIG_RUNTIME_CPU_DETECT + { + vp8_yv12_copy_frame_ptr(sd, &s->source_buffer); + } #endif cpi->source_buffer_count = 1; } @@ -4697,14 +4732,19 @@ int vp8_receive_raw_frame(VP8_PTR ptr, unsigned int frame_flags, YV12_BUFFER_CON cpi->time_receive_data += vpx_usec_timer_elapsed(&timer); #if HAVE_ARMV7 - vp8_pop_neon(store_reg); +#if CONFIG_RUNTIME_CPU_DETECT + if (cm->rtcd.flags & HAS_NEON) +#endif + { + vp8_pop_neon(store_reg); + } #endif return 0; } int vp8_get_compressed_data(VP8_PTR ptr, unsigned int *frame_flags, unsigned long *size, unsigned char *dest, INT64 *time_stamp, INT64 *time_end, int flush) { - + INT64 store_reg[8]; VP8_COMP *cpi = (VP8_COMP *) ptr; VP8_COMMON *cm = &cpi->common; struct vpx_usec_timer tsctimer; @@ -4715,7 +4755,12 @@ int vp8_get_compressed_data(VP8_PTR ptr, unsigned int *frame_flags, unsigned lon return -1; #if HAVE_ARMV7 - vp8_push_neon(store_reg); +#if CONFIG_RUNTIME_CPU_DETECT + if (cm->rtcd.flags & HAS_NEON) +#endif + { + vp8_push_neon(store_reg); + } #endif vpx_usec_timer_start(&cmptimer); @@ -4867,7 +4912,12 @@ int vp8_get_compressed_data(VP8_PTR ptr, unsigned int *frame_flags, unsigned lon #endif #if HAVE_ARMV7 - vp8_pop_neon(store_reg); +#if CONFIG_RUNTIME_CPU_DETECT + if (cm->rtcd.flags & HAS_NEON) +#endif + { + vp8_pop_neon(store_reg); + } #endif return -1; } @@ -4910,7 +4960,12 @@ int vp8_get_compressed_data(VP8_PTR ptr, unsigned int *frame_flags, unsigned lon if (!cpi) { #if HAVE_ARMV7 - vp8_pop_neon(store_reg); +#if CONFIG_RUNTIME_CPU_DETECT + if (cm->rtcd.flags & HAS_NEON) +#endif + { + vp8_pop_neon(store_reg); + } #endif return 0; } @@ -5099,7 +5154,12 @@ int vp8_get_compressed_data(VP8_PTR ptr, unsigned int *frame_flags, unsigned lon #endif #if HAVE_ARMV7 - vp8_pop_neon(store_reg); +#if CONFIG_RUNTIME_CPU_DETECT + if (cm->rtcd.flags & HAS_NEON) +#endif + { + vp8_pop_neon(store_reg); + } #endif return 0; diff --git a/vp8/encoder/picklpf.c b/vp8/encoder/picklpf.c index 79e07dbc0..09e8b5412 100644 --- a/vp8/encoder/picklpf.c +++ b/vp8/encoder/picklpf.c @@ -16,6 +16,9 @@ #include "vpx_scale/yv12extend.h" #include "vpx_scale/vpxscale.h" #include "alloccommon.h" +#if ARCH_ARM +#include "vpx_ports/arm.h" +#endif extern void vp8_loop_filter_frame(VP8_COMMON *cm, MACROBLOCKD *mbd, int filt_val); extern void vp8_loop_filter_frame_yonly(VP8_COMMON *cm, MACROBLOCKD *mbd, int filt_val, int sharpness_lvl); @@ -306,9 +309,20 @@ void vp8cx_pick_filter_level(YV12_BUFFER_CONFIG *sd, VP8_COMP *cpi) // Make a copy of the unfiltered / processed recon buffer #if HAVE_ARMV7 - vp8_yv12_copy_frame_yonly_no_extend_frame_borders_neon(cm->frame_to_show, &cpi->last_frame_uf); -#else - vp8_yv12_copy_frame_ptr(cm->frame_to_show, &cpi->last_frame_uf); +#if CONFIG_RUNTIME_CPU_DETECT + if (cm->rtcd.flags & HAS_NEON) +#endif + { + vp8_yv12_copy_frame_yonly_no_extend_frame_borders_neon(cm->frame_to_show, &cpi->last_frame_uf); + } +#if CONFIG_RUNTIME_CPU_DETECT + else +#endif +#endif +#if !HAVE_ARMV7 || CONFIG_RUNTIME_CPU_DETECT + { + vp8_yv12_copy_frame_ptr(cm->frame_to_show, &cpi->last_frame_uf); + } #endif if (cm->frame_type == KEY_FRAME) @@ -343,9 +357,20 @@ void vp8cx_pick_filter_level(YV12_BUFFER_CONFIG *sd, VP8_COMP *cpi) // Re-instate the unfiltered frame #if HAVE_ARMV7 - vp8_yv12_copy_frame_yonly_no_extend_frame_borders_neon(&cpi->last_frame_uf, cm->frame_to_show); -#else - vp8_yv12_copy_frame_yonly_ptr(&cpi->last_frame_uf, cm->frame_to_show); +#if CONFIG_RUNTIME_CPU_DETECT + if (cm->rtcd.flags & HAS_NEON) +#endif + { + vp8_yv12_copy_frame_yonly_no_extend_frame_borders_neon(&cpi->last_frame_uf, cm->frame_to_show); + } +#if CONFIG_RUNTIME_CPU_DETECT + else +#endif +#endif +#if !HAVE_ARMV7 || CONFIG_RUNTIME_CPU_DETECT + { + vp8_yv12_copy_frame_yonly_ptr(&cpi->last_frame_uf, cm->frame_to_show); + } #endif while (filter_step > 0) @@ -372,9 +397,20 @@ void vp8cx_pick_filter_level(YV12_BUFFER_CONFIG *sd, VP8_COMP *cpi) // Re-instate the unfiltered frame #if HAVE_ARMV7 - vp8_yv12_copy_frame_yonly_no_extend_frame_borders_neon(&cpi->last_frame_uf, cm->frame_to_show); -#else - vp8_yv12_copy_frame_yonly_ptr(&cpi->last_frame_uf, cm->frame_to_show); +#if CONFIG_RUNTIME_CPU_DETECT + if (cm->rtcd.flags & HAS_NEON) +#endif + { + vp8_yv12_copy_frame_yonly_no_extend_frame_borders_neon(&cpi->last_frame_uf, cm->frame_to_show); + } +#if CONFIG_RUNTIME_CPU_DETECT + else +#endif +#endif +#if !HAVE_ARMV7 || CONFIG_RUNTIME_CPU_DETECT + { + vp8_yv12_copy_frame_yonly_ptr(&cpi->last_frame_uf, cm->frame_to_show); + } #endif // If value is close to the best so far then bias towards a lower loop filter value. @@ -401,9 +437,20 @@ void vp8cx_pick_filter_level(YV12_BUFFER_CONFIG *sd, VP8_COMP *cpi) // Re-instate the unfiltered frame #if HAVE_ARMV7 - vp8_yv12_copy_frame_yonly_no_extend_frame_borders_neon(&cpi->last_frame_uf, cm->frame_to_show); -#else - vp8_yv12_copy_frame_yonly_ptr(&cpi->last_frame_uf, cm->frame_to_show); +#if CONFIG_RUNTIME_CPU_DETECT + if (cm->rtcd.flags & HAS_NEON) +#endif + { + vp8_yv12_copy_frame_yonly_no_extend_frame_borders_neon(&cpi->last_frame_uf, cm->frame_to_show); + } +#if CONFIG_RUNTIME_CPU_DETECT + else +#endif +#endif +#if !HAVE_ARMV7 || CONFIG_RUNTIME_CPU_DETECT + { + vp8_yv12_copy_frame_yonly_ptr(&cpi->last_frame_uf, cm->frame_to_show); + } #endif // Was it better than the previous best? diff --git a/vp8/vp8_common.mk b/vp8/vp8_common.mk index ecca18a0a..3b5aaa548 100644 --- a/vp8/vp8_common.mk +++ b/vp8/vp8_common.mk @@ -112,6 +112,8 @@ VP8_COMMON_SRCS-$(HAVE_MMX) += common/x86/postproc_mmx.asm VP8_COMMON_SRCS-$(HAVE_SSE2) += common/x86/postproc_sse2.asm endif +VP8_COMMON_SRCS-$(ARCH_ARM) += common/arm/arm_systemdependent.c + # common (c) VP8_COMMON_SRCS-$(HAVE_ARMV6) += common/arm/bilinearfilter_arm.c VP8_COMMON_SRCS-$(HAVE_ARMV6) += common/arm/filter_arm.c @@ -119,15 +121,8 @@ VP8_COMMON_SRCS-$(HAVE_ARMV6) += common/arm/loopfilter_arm.c VP8_COMMON_SRCS-$(HAVE_ARMV6) += common/arm/recon_arm.c VP8_COMMON_SRCS-$(HAVE_ARMV6) += common/arm/reconintra4x4_arm.c VP8_COMMON_SRCS-$(HAVE_ARMV6) += common/arm/reconintra_arm.c -VP8_COMMON_SRCS-$(HAVE_ARMV6) += common/arm/systemdependent.c VP8_COMMON_SRCS-$(HAVE_ARMV6) += common/arm/vpx_asm_offsets.c -VP8_COMMON_SRCS_REMOVE-$(HAVE_ARMV6) += common/filter_c.c -VP8_COMMON_SRCS_REMOVE-$(HAVE_ARMV6) += common/idctllm.c -VP8_COMMON_SRCS_REMOVE-$(HAVE_ARMV6) += common/recon.c -VP8_COMMON_SRCS_REMOVE-$(HAVE_ARMV6) += common/reconintra4x4.c -VP8_COMMON_SRCS_REMOVE-$(HAVE_ARMV6) += common/generic/systemdependent.c - # common (armv6) VP8_COMMON_SRCS-$(HAVE_ARMV6) += common/arm/armv6/bilinearfilter_v6$(ASM) VP8_COMMON_SRCS-$(HAVE_ARMV6) += common/arm/armv6/copymem8x4_v6$(ASM) diff --git a/vp8/vp8cx_arm.mk b/vp8/vp8cx_arm.mk index 1424bd15a..d126faf32 100644 --- a/vp8/vp8cx_arm.mk +++ b/vp8/vp8cx_arm.mk @@ -13,17 +13,22 @@ #File list for arm # encoder -VP8_CX_SRCS-$(HAVE_ARMV6) += encoder/arm/csystemdependent.c +VP8_CX_SRCS-$(ARCH_ARM) += encoder/arm/arm_csystemdependent.c VP8_CX_SRCS-$(HAVE_ARMV7) += encoder/arm/encodemb_arm.c VP8_CX_SRCS-$(HAVE_ARMV7) += encoder/arm/quantize_arm.c VP8_CX_SRCS-$(HAVE_ARMV7) += encoder/arm/picklpf_arm.c -VP8_CX_SRCS-$(HAVE_ARMV7) += encoder/arm/boolhuff_arm.c +VP8_CX_SRCS-$(HAVE_ARMV5TE) += encoder/arm/boolhuff_arm.c VP8_CX_SRCS-$(HAVE_ARMV7) += encoder/arm/mcomp_arm.c -VP8_CX_SRCS_REMOVE-$(HAVE_ARMV6) += encoder/generic/csystemdependent.c -VP8_CX_SRCS_REMOVE-$(HAVE_ARMV7) += encoder/boolhuff.c -VP8_CX_SRCS_REMOVE-$(HAVE_ARMV7) += encoder/mcomp.c +VP8_CX_SRCS_REMOVE-$(HAVE_ARMV5TE) += encoder/boolhuff.c + +#File list for armv5te +# encoder +VP8_CX_SRCS-$(HAVE_ARMV5TE) += encoder/arm/armv5te/boolhuff_armv5te$(ASM) +VP8_CX_SRCS-$(HAVE_ARMV5TE) += encoder/arm/armv5te/vp8_packtokens_armv5$(ASM) +VP8_CX_SRCS-$(HAVE_ARMV5TE) += encoder/arm/armv5te/vp8_packtokens_mbrow_armv5$(ASM) +VP8_CX_SRCS-$(HAVE_ARMV5TE) += encoder/arm/armv5te/vp8_packtokens_partitions_armv5$(ASM) #File list for armv6 # encoder @@ -44,10 +49,6 @@ VP8_CX_SRCS-$(HAVE_ARMV7) += encoder/arm/neon/vp8_subpixelvariance8x8_neon$(ASM VP8_CX_SRCS-$(HAVE_ARMV7) += encoder/arm/neon/vp8_subpixelvariance16x16_neon$(ASM) VP8_CX_SRCS-$(HAVE_ARMV7) += encoder/arm/neon/vp8_subpixelvariance16x16s_neon$(ASM) VP8_CX_SRCS-$(HAVE_ARMV7) += encoder/arm/neon/vp8_memcpy_neon$(ASM) -VP8_CX_SRCS-$(HAVE_ARMV7) += encoder/arm/neon/vp8_packtokens_armv7$(ASM) -VP8_CX_SRCS-$(HAVE_ARMV7) += encoder/arm/neon/vp8_packtokens_mbrow_armv7$(ASM) -VP8_CX_SRCS-$(HAVE_ARMV7) += encoder/arm/neon/vp8_packtokens_partitions_armv7$(ASM) -VP8_CX_SRCS-$(HAVE_ARMV7) += encoder/arm/neon/boolhuff_armv7$(ASM) VP8_CX_SRCS-$(HAVE_ARMV7) += encoder/arm/neon/vp8_shortwalsh4x4_neon$(ASM) VP8_CX_SRCS-$(HAVE_ARMV7) += encoder/arm/vpx_vp8_enc_asm_offsets.c diff --git a/vp8/vp8dx_arm.mk b/vp8/vp8dx_arm.mk index ae0610cda..0803a9cb0 100644 --- a/vp8/vp8dx_arm.mk +++ b/vp8/vp8dx_arm.mk @@ -11,11 +11,9 @@ #VP8_DX_SRCS list is modified according to different platforms. +VP8_DX_SRCS-$(ARCH_ARM) += decoder/arm/arm_dsystemdependent.c + VP8_DX_SRCS-$(HAVE_ARMV6) += decoder/arm/dequantize_arm.c -VP8_DX_SRCS-$(HAVE_ARMV6) += decoder/arm/dsystemdependent.c -VP8_DX_SRCS_REMOVE-$(HAVE_ARMV6) += decoder/generic/dsystemdependent.c -VP8_DX_SRCS_REMOVE-$(HAVE_ARMV6) += decoder/dequantize.c -VP8_DX_SRCS_REMOVE-$(HAVE_ARMV6) += decoder/idct_blk.c VP8_DX_SRCS-$(CONFIG_ARM_ASM_DETOK) += decoder/arm/detokenize$(ASM) #File list for armv6 diff --git a/vpx_ports/arm.h b/vpx_ports/arm.h new file mode 100644 index 000000000..81af1f11f --- /dev/null +++ b/vpx_ports/arm.h @@ -0,0 +1,27 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VPX_PORTS_ARM_H +#define VPX_PORTS_ARM_H +#include <stdlib.h> +#include "config.h" + +/*ARMv5TE "Enhanced DSP" instructions.*/ +#define HAS_EDSP 0x01 +/*ARMv6 "Parallel" or "Media" instructions.*/ +#define HAS_MEDIA 0x02 +/*ARMv7 optional NEON instructions.*/ +#define HAS_NEON 0x04 + +int arm_cpu_caps(void); + +#endif + diff --git a/vpx_ports/arm_cpudetect.c b/vpx_ports/arm_cpudetect.c new file mode 100644 index 000000000..4109924cf --- /dev/null +++ b/vpx_ports/arm_cpudetect.c @@ -0,0 +1,190 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <stdlib.h> +#include <string.h> +#include "arm.h" + +static int arm_cpu_env_flags(int *flags) +{ + char *env; + env = getenv("VPX_SIMD_CAPS"); + if (env && *env) + { + *flags = (int)strtol(env, NULL, 0); + return 0; + } + *flags = 0; + return -1; +} + +static int arm_cpu_env_mask(void) +{ + char *env; + env = getenv("VPX_SIMD_CAPS_MASK"); + return env && *env ? (int)strtol(env, NULL, 0) : ~0; +} + + +#if defined(_MSC_VER) +/*For GetExceptionCode() and EXCEPTION_ILLEGAL_INSTRUCTION.*/ +#define WIN32_LEAN_AND_MEAN +#define WIN32_EXTRA_LEAN +#include <windows.h> + +int arm_cpu_caps(void) +{ + int flags; + int mask; + if (!arm_cpu_env_flags(&flags)) + { + return flags; + } + mask = arm_cpu_env_mask(); + /* MSVC has no inline __asm support for ARM, but it does let you __emit + * instructions via their assembled hex code. + * All of these instructions should be essentially nops. + */ +#if defined(HAVE_ARMV5TE) + if (mask & HAS_EDSP) + { + __try + { + /*PLD [r13]*/ + __emit(0xF5DDF000); + flags |= HAS_EDSP; + } + __except(GetExceptionCode() == EXCEPTION_ILLEGAL_INSTRUCTION) + { + /*Ignore exception.*/ + } + } +#if defined(HAVE_ARMV6) + if (mask & HAS_MEDIA) + __try + { + /*SHADD8 r3,r3,r3*/ + __emit(0xE6333F93); + flags |= HAS_MEDIA; + } + __except(GetExceptionCode() == EXCEPTION_ILLEGAL_INSTRUCTION) + { + /*Ignore exception.*/ + } + } +#if defined(HAVE_ARMV7) + if (mask & HAS_NEON) + { + __try + { + /*VORR q0,q0,q0*/ + __emit(0xF2200150); + flags |= HAS_NEON; + } + __except(GetExceptionCode() == EXCEPTION_ILLEGAL_INSTRUCTION) + { + /*Ignore exception.*/ + } + } +#endif +#endif +#endif + return flags & mask; +} + +#elif defined(__linux__) +#include <stdio.h> + +int arm_cpu_caps(void) +{ + FILE *fin; + int flags; + int mask; + if (!arm_cpu_env_flags(&flags)) + { + return flags; + } + mask = arm_cpu_env_mask(); + /* Reading /proc/self/auxv would be easier, but that doesn't work reliably + * on Android. + * This also means that detection will fail in Scratchbox. + */ + fin = fopen("/proc/cpuinfo","r"); + if(fin != NULL) + { + /* 512 should be enough for anybody (it's even enough for all the flags + * that x86 has accumulated... so far). + */ + char buf[512]; + while (fgets(buf, 511, fin) != NULL) + { +#if defined(HAVE_ARMV5TE) || defined(HAVE_ARMV7) + if (memcmp(buf, "Features", 8) == 0) + { + char *p; +#if defined(HAVE_ARMV5TE) + p=strstr(buf, " edsp"); + if (p != NULL && (p[5] == ' ' || p[5] == '\n')) + { + flags |= HAS_EDSP; + } +#if defined(HAVE_ARMV7) + p = strstr(buf, " neon"); + if (p != NULL && (p[5] == ' ' || p[5] == '\n')) + { + flags |= HAS_NEON; + } +#endif +#endif + } +#endif +#if defined(HAVE_ARMV6) + if (memcmp(buf, "CPU architecture:",17) == 0){ + int version; + version = atoi(buf+17); + if (version >= 6) + { + flags |= HAS_MEDIA; + } + } +#endif + } + fclose(fin); + } + return flags & mask; +} + +#elif !CONFIG_RUNTIME_CPU_DETECT + +int arm_cpu_caps(void) +{ + int flags; + int mask; + if (!arm_cpu_env_flags(&flags)) + { + return flags; + } + mask = arm_cpu_env_mask(); +#if defined(HAVE_ARMV5TE) + flags |= HAS_EDSP; +#endif +#if defined(HAVE_ARMV6) + flags |= HAS_MEDIA; +#endif +#if defined(HAVE_ARMV7) + flags |= HAS_NEON; +#endif + return flags & mask; +} + +#else +#error "--enable-runtime-cpu-detect selected, but no CPU detection method " \ + "available for your platform. Reconfigure without --enable-runtime-cpu-detect." +#endif diff --git a/vpx_scale/arm/scalesystemdependant.c b/vpx_scale/arm/scalesystemdependant.c index 1e8bcb89d..fee76fff7 100644 --- a/vpx_scale/arm/scalesystemdependant.c +++ b/vpx_scale/arm/scalesystemdependant.c @@ -10,6 +10,7 @@ #include "vpx_ports/config.h" +#include "vpx_ports/arm.h" #include "vpx_scale/vpxscale.h" @@ -47,6 +48,9 @@ extern void vp8_yv12_copy_frame_neon(YV12_BUFFER_CONFIG *src_ybc, YV12_BUFFER_CO ****************************************************************************/ void vp8_scale_machine_specific_config() { +#if HAVE_ARMV7 && CONFIG_RUNTIME_CPU_DETECT + int flags; +#endif /* vp8_horizontal_line_1_2_scale = horizontal_line_1_2_scale_armv4; vp8_vertical_band_1_2_scale = vertical_band_1_2_scale_armv4; @@ -73,14 +77,20 @@ void vp8_scale_machine_specific_config() vp8_horizontal_line_5_4_scale = vp8cx_horizontal_line_5_4_scale_c; */ -#if HAVE_ARMV7 - vp8_yv12_extend_frame_borders_ptr = vp8_yv12_extend_frame_borders_neon; - vp8_yv12_copy_frame_yonly_ptr = vp8_yv12_copy_frame_yonly_neon; - vp8_yv12_copy_frame_ptr = vp8_yv12_copy_frame_neon; -#else +#if !HAVE_ARMV7 || CONFIG_RUNTIME_CPU_DETECT vp8_yv12_extend_frame_borders_ptr = vp8_yv12_extend_frame_borders; vp8_yv12_copy_frame_yonly_ptr = vp8_yv12_copy_frame_yonly; vp8_yv12_copy_frame_ptr = vp8_yv12_copy_frame; #endif - +#if HAVE_ARMV7 +#if CONFIG_RUNTIME_CPU_DETECT + flags = arm_cpu_caps(); + if (flags & HAS_NEON) +#endif + { + vp8_yv12_extend_frame_borders_ptr = vp8_yv12_extend_frame_borders_neon; + vp8_yv12_copy_frame_yonly_ptr = vp8_yv12_copy_frame_yonly_neon; + vp8_yv12_copy_frame_ptr = vp8_yv12_copy_frame_neon; + } +#endif } |