With the intrinsics it is no longer necessary to have a stub/helper
function.
Change-Id: I3695961c3c94f1bb750d3b7b29716e509ebba482
# For this we import the 'cpufeatures' module from the NDK sources.
# libvpx can also be configured without this runtime detection method.
# Configuring with --disable-runtime-cpu-detect will assume presence of NEON.
# For this we import the 'cpufeatures' module from the NDK sources.
# libvpx can also be configured without this runtime detection method.
# Configuring with --disable-runtime-cpu-detect will assume presence of NEON.
-# Configuring with --disable-runtime-cpu-detect --disable-neon will remove any
-# NEON dependency.
+# Configuring with --disable-runtime-cpu-detect --disable-neon \
+# --disable-neon-asm
+# will remove any NEON dependency.
# To change to building armeabi, run ./libvpx/configure again, but with
# --target=arm5te-android-gcc and modify the Application.mk file to
# To change to building armeabi, run ./libvpx/configure again, but with
# --target=arm5te-android-gcc and modify the Application.mk file to
#include "vpx_config.h"
#include "vp8/common/blockd.h"
#include "vpx_config.h"
#include "vp8/common/blockd.h"
-#if HAVE_NEON_ASM
-extern void vp8_dequantize_b_loop_neon(short *Q, short *DQC, short *DQ);
-#endif
-
#if HAVE_MEDIA
extern void vp8_dequantize_b_loop_v6(short *Q, short *DQC, short *DQ);
#if HAVE_MEDIA
extern void vp8_dequantize_b_loop_v6(short *Q, short *DQC, short *DQ);
-#endif
-
-#if HAVE_NEON_ASM
-
-void vp8_dequantize_b_neon(BLOCKD *d, short *DQC)
-{
- short *DQ = d->dqcoeff;
- short *Q = d->qcoeff;
-
- vp8_dequantize_b_loop_neon(Q, DQC, DQ);
-}
-#endif
void vp8_dequantize_b_v6(BLOCKD *d, short *DQC)
{
short *DQ = d->dqcoeff;
void vp8_dequantize_b_v6(BLOCKD *d, short *DQC)
{
short *DQ = d->dqcoeff;
-void vp8_dequantize_b_loop_neon(
- int16_t *Q,
- int16_t *DQC,
- int16_t *DQ) {
+#include "vp8/common/blockd.h"
+
+void vp8_dequantize_b_neon(BLOCKD *d, short *DQC) {
int16x8x2_t qQ, qDQC, qDQ;
int16x8x2_t qQ, qDQC, qDQ;
+ qQ = vld2q_s16(d->qcoeff);
qDQC = vld2q_s16(DQC);
qDQ.val[0] = vmulq_s16(qQ.val[0], qDQC.val[0]);
qDQ.val[1] = vmulq_s16(qQ.val[1], qDQC.val[1]);
qDQC = vld2q_s16(DQC);
qDQ.val[0] = vmulq_s16(qQ.val[0], qDQC.val[0]);
qDQ.val[1] = vmulq_s16(qQ.val[1], qDQC.val[1]);
- vst2q_s16(DQ, qDQ);
- return;
+ vst2q_s16(d->dqcoeff, qDQ);
# Dequant
#
add_proto qw/void vp8_dequantize_b/, "struct blockd*, short *dqc";
# Dequant
#
add_proto qw/void vp8_dequantize_b/, "struct blockd*, short *dqc";
-specialize qw/vp8_dequantize_b mmx media neon_asm/;
+specialize qw/vp8_dequantize_b mmx media neon/;
$vp8_dequantize_b_media=vp8_dequantize_b_v6;
$vp8_dequantize_b_media=vp8_dequantize_b_v6;
-$vp8_dequantize_b_neon_asm=vp8_dequantize_b_neon;
add_proto qw/void vp8_dequant_idct_add/, "short *input, short *dq, unsigned char *output, int stride";
specialize qw/vp8_dequant_idct_add mmx media neon dspr2/;
add_proto qw/void vp8_dequant_idct_add/, "short *input, short *dq, unsigned char *output, int stride";
specialize qw/vp8_dequant_idct_add mmx media neon dspr2/;