shithub: libvpx

Download patch

ref: a6be7ba1aaf67989bf97bae04707fe9bf5c7d4d4
parent: efb56ec3ff6c8bfdeb1b29c97374017e1556aa16
author: James Zern <jzern@google.com>
date: Sat Oct 1 07:59:31 EDT 2016 

enable idct*_1_add_neon in high-bitdepth builds

these are compatible as they only load one element of the input so the
larger size of tran_low_t makes no difference in little endian builds.
note the asm is incompatible with big-endian, but there are other points of
failure there so currently it's considered unsupported.

BUG=webm:1294

Change-Id: Icd2665a0699bccae92d1bea43a95b0a83fb17028


--- a/test/partial_idct_test.cc
+++ b/test/partial_idct_test.cc
@@ -201,7 +201,19 @@
                       make_tuple(&vpx_fdct4x4_c, &vpx_idct4x4_16_add_c,
                                  &vpx_idct4x4_1_add_c, TX_4X4, 1)));
 
-#if HAVE_NEON && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
+#if HAVE_NEON && !CONFIG_EMULATE_HARDWARE
+#if CONFIG_VP9_HIGHBITDEPTH
+INSTANTIATE_TEST_CASE_P(
+    NEON, PartialIDctTest,
+    ::testing::Values(make_tuple(&vpx_fdct32x32_c, &vpx_idct32x32_1024_add_c,
+                                 &vpx_idct32x32_1_add_neon, TX_32X32, 1),
+                      make_tuple(&vpx_fdct16x16_c, &vpx_idct16x16_256_add_c,
+                                 &vpx_idct16x16_1_add_neon, TX_16X16, 1),
+                      make_tuple(&vpx_fdct8x8_c, &vpx_idct8x8_64_add_c,
+                                 &vpx_idct8x8_1_add_neon, TX_8X8, 1),
+                      make_tuple(&vpx_fdct4x4_c, &vpx_idct4x4_16_add_c,
+                                 &vpx_idct4x4_1_add_neon, TX_4X4, 1)));
+#else   // !CONFIG_VP9_HIGHBITDEPTH
 // 32x32_34_ 32x32_135_ are implemented using the 1024 version.
 INSTANTIATE_TEST_CASE_P(
     NEON, PartialIDctTest,
@@ -229,7 +241,8 @@
                                  &vpx_idct4x4_16_add_neon, TX_4X4, 16),
                       make_tuple(&vpx_fdct4x4_c, &vpx_idct4x4_16_add_c,
                                  &vpx_idct4x4_1_add_neon, TX_4X4, 1)));
-#endif  // HAVE_NEON && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+#endif  // HAVE_NEON && !CONFIG_EMULATE_HARDWARE
 
 #if HAVE_SSE2 && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
 // 32x32_135_ is implemented using the 1024 version.
--- a/vpx_dsp/vpx_dsp.mk
+++ b/vpx_dsp/vpx_dsp.mk
@@ -199,23 +199,15 @@
 
 ifneq ($(CONFIG_VP9_HIGHBITDEPTH),yes)
 ifeq ($(HAVE_NEON_ASM),yes)
-DSP_SRCS-yes  += arm/idct4x4_1_add_neon$(ASM)
 DSP_SRCS-yes  += arm/idct4x4_add_neon$(ASM)
-DSP_SRCS-yes  += arm/idct8x8_1_add_neon$(ASM)
 DSP_SRCS-yes  += arm/idct8x8_add_neon$(ASM)
-DSP_SRCS-yes  += arm/idct16x16_1_add_neon$(ASM)
 DSP_SRCS-yes  += arm/idct16x16_add_neon$(ASM)
-DSP_SRCS-yes  += arm/idct32x32_1_add_neon$(ASM)
 DSP_SRCS-yes  += arm/idct32x32_add_neon$(ASM)
 else
 ifeq ($(HAVE_NEON),yes)
-DSP_SRCS-yes  += arm/idct4x4_1_add_neon.c
 DSP_SRCS-yes  += arm/idct4x4_add_neon.c
-DSP_SRCS-yes  += arm/idct8x8_1_add_neon.c
 DSP_SRCS-yes  += arm/idct8x8_add_neon.c
-DSP_SRCS-yes  += arm/idct16x16_1_add_neon.c
 DSP_SRCS-yes  += arm/idct16x16_add_neon.c
-DSP_SRCS-yes  += arm/idct32x32_1_add_neon.c
 DSP_SRCS-yes  += arm/idct32x32_add_neon.c
 endif  # HAVE_NEON
 endif  # HAVE_NEON_ASM
@@ -233,7 +225,20 @@
 DSP_SRCS-$(HAVE_DSPR2) += mips/itrans16_dspr2.c
 DSP_SRCS-$(HAVE_DSPR2) += mips/itrans32_dspr2.c
 DSP_SRCS-$(HAVE_DSPR2) += mips/itrans32_cols_dspr2.c
-endif  # CONFIG_VP9_HIGHBITDEPTH
+endif  # !CONFIG_VP9_HIGHBITDEPTH
+
+ifeq ($(HAVE_NEON_ASM),yes)
+DSP_SRCS-yes += arm/idct4x4_1_add_neon$(ASM)
+DSP_SRCS-yes += arm/idct8x8_1_add_neon$(ASM)
+DSP_SRCS-yes += arm/idct16x16_1_add_neon$(ASM)
+DSP_SRCS-yes += arm/idct32x32_1_add_neon$(ASM)
+else
+DSP_SRCS-$(HAVE_NEON) += arm/idct4x4_1_add_neon.c
+DSP_SRCS-$(HAVE_NEON) += arm/idct8x8_1_add_neon.c
+DSP_SRCS-$(HAVE_NEON) += arm/idct16x16_1_add_neon.c
+DSP_SRCS-$(HAVE_NEON) += arm/idct32x32_1_add_neon.c
+endif  # HAVE_NEON_ASM
+
 endif  # CONFIG_VP9
 
 # quantization
--- a/vpx_dsp/vpx_dsp_rtcd_defs.pl
+++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl
@@ -647,7 +647,7 @@
     specialize qw/vpx_idct4x4_16_add sse2/;
 
     add_proto qw/void vpx_idct4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
-    specialize qw/vpx_idct4x4_1_add sse2/;
+    specialize qw/vpx_idct4x4_1_add neon sse2/;
 
     add_proto qw/void vpx_idct8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
     specialize qw/vpx_idct8x8_64_add sse2/, "$ssse3_x86_64";
@@ -656,7 +656,7 @@
     specialize qw/vpx_idct8x8_12_add sse2/, "$ssse3_x86_64";
 
     add_proto qw/void vpx_idct8x8_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
-    specialize qw/vpx_idct8x8_1_add sse2/;
+    specialize qw/vpx_idct8x8_1_add neon sse2/;
 
     add_proto qw/void vpx_idct16x16_256_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
     specialize qw/vpx_idct16x16_256_add sse2/;
@@ -665,7 +665,7 @@
     specialize qw/vpx_idct16x16_10_add sse2/;
 
     add_proto qw/void vpx_idct16x16_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
-    specialize qw/vpx_idct16x16_1_add sse2/;
+    specialize qw/vpx_idct16x16_1_add neon sse2/;
 
     add_proto qw/void vpx_idct32x32_1024_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
     specialize qw/vpx_idct32x32_1024_add sse2/, "$ssse3_x86_64";
@@ -679,7 +679,7 @@
     specialize qw/vpx_idct32x32_34_add sse2/, "$ssse3_x86_64";
 
     add_proto qw/void vpx_idct32x32_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
-    specialize qw/vpx_idct32x32_1_add sse2/;
+    specialize qw/vpx_idct32x32_1_add neon sse2/;
 
     add_proto qw/void vpx_highbd_idct4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
     specialize qw/vpx_highbd_idct4x4_16_add sse2/;