ref: a53d5a4c442a84cacbd8225fac72db3789b3e10c
parent: c8df1656bd94928059204242e778bd5b8b9dc7aa
author: Scott LaVarnway <slavarnway@google.com>
date: Thu Dec 15 09:23:36 EST 2011
Moved dequant idct into common These functions are now used by the encoder. This is WIP with the goal of creating a common idct/add for the encoder and decoder. A boost of 1.8% was seen for the HD rt test clip used. [Tero] Added needed changes to ARM side. Change-Id: Ibbb8000be09034203d7adffc457d3c3f8b06a5bf
--- a/vp8/common/arm/arm_systemdependent.c
+++ b/vp8/common/arm/arm_systemdependent.c
@@ -63,6 +63,12 @@
rtcd->recon.copy8x8 = vp8_copy_mem8x8_v6;
rtcd->recon.copy8x4 = vp8_copy_mem8x4_v6;
rtcd->recon.intra4x4_predict = vp8_intra4x4_predict_armv6;
+
+ rtcd->dequant.block = vp8_dequantize_b_v6;
+ rtcd->dequant.idct_add = vp8_dequant_idct_add_v6;
+ rtcd->dequant.idct_add_y_block = vp8_dequant_idct_add_y_block_v6;
+ rtcd->dequant.idct_add_uv_block = vp8_dequant_idct_add_uv_block_v6;
+
}
#endif
@@ -97,6 +103,12 @@
vp8_build_intra_predictors_mby_neon;
rtcd->recon.build_intra_predictors_mby_s =
vp8_build_intra_predictors_mby_s_neon;
+
+ rtcd->dequant.block = vp8_dequantize_b_neon;
+ rtcd->dequant.idct_add = vp8_dequant_idct_add_neon;
+ rtcd->dequant.idct_add_y_block = vp8_dequant_idct_add_y_block_neon;
+ rtcd->dequant.idct_add_uv_block = vp8_dequant_idct_add_uv_block_neon;
+
}
#endif
--- /dev/null
+++ b/vp8/common/arm/armv6/dequant_idct_v6.asm
@@ -1,0 +1,190 @@
+;
+; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+; Use of this source code is governed by a BSD-style license and patent
+; grant that can be found in the LICENSE file in the root of the source
+; tree. All contributing project authors may be found in the AUTHORS
+; file in the root of the source tree.
+;
+
+ EXPORT |vp8_dequant_idct_add_v6|
+
+ AREA |.text|, CODE, READONLY
+;void vp8_dequant_idct_v6(short *input, short *dq,
+; unsigned char *dest, int stride)
+; r0 = q
+; r1 = dq
+; r2 = dst
+; r3 = stride
+
+|vp8_dequant_idct_add_v6| PROC
+ stmdb sp!, {r4-r11, lr}+
+ ldr r4, [r0] ;input
+ ldr r5, [r1], #4 ;dq
+
+ sub sp, sp, #4
+ str r3, [sp]
+
+ mov r12, #4
+
+vp8_dequant_add_loop
+ smulbb r6, r4, r5
+ smultt r7, r4, r5
+
+ ldr r4, [r0, #4] ;input
+ ldr r5, [r1], #4 ;dq
+
+ strh r6, [r0], #2
+ strh r7, [r0], #2
+
+ smulbb r6, r4, r5
+ smultt r7, r4, r5
+
+ subs r12, r12, #1
+
+ ldrne r4, [r0, #4]
+ ldrne r5, [r1], #4
+
+ strh r6, [r0], #2
+ strh r7, [r0], #2
+
+ bne vp8_dequant_add_loop
+
+ sub r0, r0, #32
+ mov r1, r0
+
+; short_idct4x4llm_v6_dual
+ ldr r3, cospi8sqrt2minus1
+ ldr r4, sinpi8sqrt2
+ ldr r6, [r0, #8]
+ mov r5, #2
+vp8_dequant_idct_loop1_v6
+ ldr r12, [r0, #24]
+ ldr r14, [r0, #16]
+ smulwt r9, r3, r6
+ smulwb r7, r3, r6
+ smulwt r10, r4, r6
+ smulwb r8, r4, r6
+ pkhbt r7, r7, r9, lsl #16
+ smulwt r11, r3, r12
+ pkhbt r8, r8, r10, lsl #16
+ uadd16 r6, r6, r7
+ smulwt r7, r4, r12
+ smulwb r9, r3, r12
+ smulwb r10, r4, r12
+ subs r5, r5, #1
+ pkhbt r9, r9, r11, lsl #16
+ ldr r11, [r0], #4
+ pkhbt r10, r10, r7, lsl #16
+ uadd16 r7, r12, r9
+ usub16 r7, r8, r7
+ uadd16 r6, r6, r10
+ uadd16 r10, r11, r14
+ usub16 r8, r11, r14
+ uadd16 r9, r10, r6
+ usub16 r10, r10, r6
+ uadd16 r6, r8, r7
+ usub16 r7, r8, r7
+ str r6, [r1, #8]
+ ldrne r6, [r0, #8]
+ str r7, [r1, #16]
+ str r10, [r1, #24]
+ str r9, [r1], #4
+ bne vp8_dequant_idct_loop1_v6
+
+ mov r5, #2
+ sub r0, r1, #8
+vp8_dequant_idct_loop2_v6
+ ldr r6, [r0], #4
+ ldr r7, [r0], #4
+ ldr r8, [r0], #4
+ ldr r9, [r0], #4
+ smulwt r1, r3, r6
+ smulwt r12, r4, r6
+ smulwt lr, r3, r8
+ smulwt r10, r4, r8
+ pkhbt r11, r8, r6, lsl #16
+ pkhbt r1, lr, r1, lsl #16
+ pkhbt r12, r10, r12, lsl #16
+ pkhtb r6, r6, r8, asr #16
+ uadd16 r6, r1, r6
+ pkhbt lr, r9, r7, lsl #16
+ uadd16 r10, r11, lr
+ usub16 lr, r11, lr
+ pkhtb r8, r7, r9, asr #16
+ subs r5, r5, #1
+ smulwt r1, r3, r8
+ smulwb r7, r3, r8
+ smulwt r11, r4, r8
+ smulwb r9, r4, r8
+ pkhbt r1, r7, r1, lsl #16
+ uadd16 r8, r1, r8
+ pkhbt r11, r9, r11, lsl #16
+ usub16 r1, r12, r8
+ uadd16 r8, r11, r6
+ ldr r9, c0x00040004
+ ldr r12, [sp] ; get stride from stack
+ uadd16 r6, r10, r8
+ usub16 r7, r10, r8
+ uadd16 r7, r7, r9
+ uadd16 r6, r6, r9
+ uadd16 r10, r14, r1
+ usub16 r1, r14, r1
+ uadd16 r10, r10, r9
+ uadd16 r1, r1, r9
+ ldr r11, [r2] ; load input from dst
+ mov r8, r7, asr #3
+ pkhtb r9, r8, r10, asr #19
+ mov r8, r1, asr #3
+ pkhtb r8, r8, r6, asr #19
+ uxtb16 lr, r11, ror #8
+ qadd16 r9, r9, lr
+ uxtb16 lr, r11
+ qadd16 r8, r8, lr
+ usat16 r9, #8, r9
+ usat16 r8, #8, r8
+ orr r9, r8, r9, lsl #8
+ ldr r11, [r2, r12] ; load input from dst
+ mov r7, r7, lsl #16
+ mov r1, r1, lsl #16
+ mov r10, r10, lsl #16
+ mov r6, r6, lsl #16
+ mov r7, r7, asr #3
+ pkhtb r7, r7, r10, asr #19
+ mov r1, r1, asr #3
+ pkhtb r1, r1, r6, asr #19
+ uxtb16 r8, r11, ror #8
+ qadd16 r7, r7, r8
+ uxtb16 r8, r11
+ qadd16 r1, r1, r8
+ usat16 r7, #8, r7
+ usat16 r1, #8, r1
+ orr r1, r1, r7, lsl #8
+ str r9, [r2], r12 ; store output to dst
+ str r1, [r2], r12 ; store output to dst
+ bne vp8_dequant_idct_loop2_v6
+
+; vpx_memset
+ sub r0, r0, #32
+ add sp, sp, #4
+
+ mov r12, #0
+ str r12, [r0]
+ str r12, [r0, #4]
+ str r12, [r0, #8]
+ str r12, [r0, #12]
+ str r12, [r0, #16]
+ str r12, [r0, #20]
+ str r12, [r0, #24]
+ str r12, [r0, #28]
+
+ ldmia sp!, {r4 - r11, pc}+ ENDP ; |vp8_dequant_idct_add_v6|
+
+; Constant Pool
+cospi8sqrt2minus1 DCD 0x00004E7B
+sinpi8sqrt2 DCD 0x00008A8C
+c0x00040004 DCD 0x00040004
+
+ END
--- /dev/null
+++ b/vp8/common/arm/armv6/dequantize_v6.asm
@@ -1,0 +1,69 @@
+;
+; 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.
+;
+
+
+ EXPORT |vp8_dequantize_b_loop_v6|
+
+ AREA |.text|, CODE, READONLY ; name this block of code
+;-------------------------------
+;void vp8_dequantize_b_loop_v6(short *Q, short *DQC, short *DQ);
+; r0 short *Q,
+; r1 short *DQC
+; r2 short *DQ
+|vp8_dequantize_b_loop_v6| PROC
+ stmdb sp!, {r4-r9, lr}+
+ ldr r3, [r0] ;load Q
+ ldr r4, [r1] ;load DQC
+ ldr r5, [r0, #4]
+ ldr r6, [r1, #4]
+
+ mov r12, #2 ;loop counter
+
+dequant_loop
+ smulbb r7, r3, r4 ;multiply
+ smultt r8, r3, r4
+ smulbb r9, r5, r6
+ smultt lr, r5, r6
+
+ ldr r3, [r0, #8]
+ ldr r4, [r1, #8]
+ ldr r5, [r0, #12]
+ ldr r6, [r1, #12]
+
+ strh r7, [r2], #2 ;store result
+ smulbb r7, r3, r4 ;multiply
+ strh r8, [r2], #2
+ smultt r8, r3, r4
+ strh r9, [r2], #2
+ smulbb r9, r5, r6
+ strh lr, [r2], #2
+ smultt lr, r5, r6
+
+ subs r12, r12, #1
+
+ add r0, r0, #16
+ add r1, r1, #16
+
+ ldrne r3, [r0]
+ strh r7, [r2], #2 ;store result
+ ldrne r4, [r1]
+ strh r8, [r2], #2
+ ldrne r5, [r0, #4]
+ strh r9, [r2], #2
+ ldrne r6, [r1, #4]
+ strh lr, [r2], #2
+
+ bne dequant_loop
+
+ ldmia sp!, {r4-r9, pc}+ ENDP ;|vp8_dequantize_b_loop_v6|
+
+ END
--- /dev/null
+++ b/vp8/common/arm/armv6/idct_blk_v6.c
@@ -1,0 +1,116 @@
+/*
+ * 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_config.h"
+#include "vp8/common/idct.h"
+#include "vp8/common/dequantize.h"
+
+
+void vp8_dequant_idct_add_y_block_v6(short *q, short *dq,
+ unsigned char *dst,
+ int stride, char *eobs)
+{+ int i;
+
+ for (i = 0; i < 4; i++)
+ {+ if (eobs[0] > 1)
+ vp8_dequant_idct_add_v6 (q, dq, dst, stride);
+ else if (eobs[0] == 1)
+ {+ vp8_dc_only_idct_add_v6 (q[0]*dq[0], dst, stride, dst, stride);
+ ((int *)q)[0] = 0;
+ }
+
+ if (eobs[1] > 1)
+ vp8_dequant_idct_add_v6 (q+16, dq, dst+4, stride);
+ else if (eobs[1] == 1)
+ {+ vp8_dc_only_idct_add_v6 (q[16]*dq[0], dst+4, stride, dst+4, stride);
+ ((int *)(q+16))[0] = 0;
+ }
+
+ if (eobs[2] > 1)
+ vp8_dequant_idct_add_v6 (q+32, dq, dst+8, stride);
+ else if (eobs[2] == 1)
+ {+ vp8_dc_only_idct_add_v6 (q[32]*dq[0], dst+8, stride, dst+8, stride);
+ ((int *)(q+32))[0] = 0;
+ }
+
+ if (eobs[3] > 1)
+ vp8_dequant_idct_add_v6 (q+48, dq, dst+12, stride);
+ else if (eobs[3] == 1)
+ {+ vp8_dc_only_idct_add_v6 (q[48]*dq[0], dst+12, stride,dst+12,stride);
+ ((int *)(q+48))[0] = 0;
+ }
+
+ q += 64;
+ dst += 4*stride;
+ eobs += 4;
+ }
+}
+
+void vp8_dequant_idct_add_uv_block_v6(short *q, short *dq,
+ unsigned char *dstu,
+ unsigned char *dstv,
+ int stride, char *eobs)
+{+ int i;
+
+ for (i = 0; i < 2; i++)
+ {+ if (eobs[0] > 1)
+ vp8_dequant_idct_add_v6 (q, dq, dstu, stride);
+ else if (eobs[0] == 1)
+ {+ vp8_dc_only_idct_add_v6 (q[0]*dq[0], dstu, stride, dstu, stride);
+ ((int *)q)[0] = 0;
+ }
+
+ if (eobs[1] > 1)
+ vp8_dequant_idct_add_v6 (q+16, dq, dstu+4, stride);
+ else if (eobs[1] == 1)
+ {+ vp8_dc_only_idct_add_v6 (q[16]*dq[0], dstu+4, stride,
+ dstu+4, stride);
+ ((int *)(q+16))[0] = 0;
+ }
+
+ q += 32;
+ dstu += 4*stride;
+ eobs += 2;
+ }
+
+ for (i = 0; i < 2; i++)
+ {+ if (eobs[0] > 1)
+ vp8_dequant_idct_add_v6 (q, dq, dstv, stride);
+ else if (eobs[0] == 1)
+ {+ vp8_dc_only_idct_add_v6 (q[0]*dq[0], dstv, stride, dstv, stride);
+ ((int *)q)[0] = 0;
+ }
+
+ if (eobs[1] > 1)
+ vp8_dequant_idct_add_v6 (q+16, dq, dstv+4, stride);
+ else if (eobs[1] == 1)
+ {+ vp8_dc_only_idct_add_v6 (q[16]*dq[0], dstv+4, stride,
+ dstv+4, stride);
+ ((int *)(q+16))[0] = 0;
+ }
+
+ q += 32;
+ dstv += 4*stride;
+ eobs += 2;
+ }
+}
--- /dev/null
+++ b/vp8/common/arm/dequantize_arm.c
@@ -1,0 +1,45 @@
+/*
+ * 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_config.h"
+#include "vp8/common/dequantize.h"
+#include "vp8/common/idct.h"
+
+#if HAVE_ARMV7
+extern void vp8_dequantize_b_loop_neon(short *Q, short *DQC, short *DQ);
+#endif
+
+#if HAVE_ARMV6
+extern void vp8_dequantize_b_loop_v6(short *Q, short *DQC, short *DQ);
+#endif
+
+#if HAVE_ARMV7
+
+void vp8_dequantize_b_neon(BLOCKD *d)
+{+ short *DQ = d->dqcoeff;
+ short *Q = d->qcoeff;
+ short *DQC = d->dequant;
+
+ vp8_dequantize_b_loop_neon(Q, DQC, DQ);
+}
+#endif
+
+#if HAVE_ARMV6
+void vp8_dequantize_b_v6(BLOCKD *d)
+{+ short *DQ = d->dqcoeff;
+ short *Q = d->qcoeff;
+ short *DQC = d->dequant;
+
+ vp8_dequantize_b_loop_v6(Q, DQC, DQ);
+}
+#endif
--- /dev/null
+++ b/vp8/common/arm/dequantize_arm.h
@@ -1,0 +1,59 @@
+/*
+ * 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 DEQUANTIZE_ARM_H
+#define DEQUANTIZE_ARM_H
+
+#if HAVE_ARMV6
+extern prototype_dequant_block(vp8_dequantize_b_v6);
+extern prototype_dequant_idct_add(vp8_dequant_idct_add_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
+
+#undef vp8_dequant_idct_add
+#define vp8_dequant_idct_add vp8_dequant_idct_add_v6
+
+#undef vp8_dequant_idct_add_y_block
+#define vp8_dequant_idct_add_y_block vp8_dequant_idct_add_y_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);
+extern prototype_dequant_idct_add(vp8_dequant_idct_add_neon);
+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
+
+#undef vp8_dequant_idct_add
+#define vp8_dequant_idct_add vp8_dequant_idct_add_neon
+
+#undef vp8_dequant_idct_add_y_block
+#define vp8_dequant_idct_add_y_block vp8_dequant_idct_add_y_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
--- /dev/null
+++ b/vp8/common/arm/neon/dequant_idct_neon.asm
@@ -1,0 +1,131 @@
+;
+; 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.
+;
+
+
+ EXPORT |vp8_dequant_idct_add_neon|
+ ARM
+ REQUIRE8
+ PRESERVE8
+
+ AREA ||.text||, CODE, READONLY, ALIGN=2
+;void vp8_dequant_idct_add_neon(short *input, short *dq,
+; unsigned char *dest, int stride)
+; r0 short *input,
+; r1 short *dq,
+; r2 unsigned char *dest
+; r3 int stride
+
+|vp8_dequant_idct_add_neon| PROC
+ vld1.16 {q3, q4}, [r0]+ vld1.16 {q5, q6}, [r1]+
+ add r1, r2, r3 ; r1 = dest + stride
+ lsl r3, #1 ; 2x stride
+
+ vld1.32 {d14[0]}, [r2], r3+ vld1.32 {d14[1]}, [r1], r3+ vld1.32 {d15[0]}, [r2]+ vld1.32 {d15[1]}, [r1]+
+ adr r12, cospi8sqrt2minus1 ; pointer to the first constant
+
+ vmul.i16 q1, q3, q5 ;input for short_idct4x4llm_neon
+ vmul.i16 q2, q4, q6
+
+;|short_idct4x4llm_neon| PROC
+ vld1.16 {d0}, [r12]+ vswp d3, d4 ;q2(vp[4] vp[12])
+
+ vqdmulh.s16 q3, q2, d0[2]
+ vqdmulh.s16 q4, q2, d0[0]
+
+ vqadd.s16 d12, d2, d3 ;a1
+ vqsub.s16 d13, d2, d3 ;b1
+
+ vshr.s16 q3, q3, #1
+ vshr.s16 q4, q4, #1
+
+ vqadd.s16 q3, q3, q2
+ vqadd.s16 q4, q4, q2
+
+ vqsub.s16 d10, d6, d9 ;c1
+ vqadd.s16 d11, d7, d8 ;d1
+
+ vqadd.s16 d2, d12, d11
+ vqadd.s16 d3, d13, d10
+ vqsub.s16 d4, d13, d10
+ vqsub.s16 d5, d12, d11
+
+ vtrn.32 d2, d4
+ vtrn.32 d3, d5
+ vtrn.16 d2, d3
+ vtrn.16 d4, d5
+
+; memset(input, 0, 32) -- 32bytes
+ vmov.i16 q14, #0
+
+ vswp d3, d4
+ vqdmulh.s16 q3, q2, d0[2]
+ vqdmulh.s16 q4, q2, d0[0]
+
+ vqadd.s16 d12, d2, d3 ;a1
+ vqsub.s16 d13, d2, d3 ;b1
+
+ vmov q15, q14
+
+ vshr.s16 q3, q3, #1
+ vshr.s16 q4, q4, #1
+
+ vqadd.s16 q3, q3, q2
+ vqadd.s16 q4, q4, q2
+
+ vqsub.s16 d10, d6, d9 ;c1
+ vqadd.s16 d11, d7, d8 ;d1
+
+ vqadd.s16 d2, d12, d11
+ vqadd.s16 d3, d13, d10
+ vqsub.s16 d4, d13, d10
+ vqsub.s16 d5, d12, d11
+
+ vst1.16 {q14, q15}, [r0]+
+ vrshr.s16 d2, d2, #3
+ vrshr.s16 d3, d3, #3
+ vrshr.s16 d4, d4, #3
+ vrshr.s16 d5, d5, #3
+
+ vtrn.32 d2, d4
+ vtrn.32 d3, d5
+ vtrn.16 d2, d3
+ vtrn.16 d4, d5
+
+ vaddw.u8 q1, q1, d14
+ vaddw.u8 q2, q2, d15
+
+ sub r2, r2, r3
+ sub r1, r1, r3
+
+ vqmovun.s16 d0, q1
+ vqmovun.s16 d1, q2
+
+ vst1.32 {d0[0]}, [r2], r3+ vst1.32 {d0[1]}, [r1], r3+ vst1.32 {d1[0]}, [r2]+ vst1.32 {d1[1]}, [r1]+
+ bx lr
+
+ ENDP ; |vp8_dequant_idct_add_neon|
+
+; Constant Pool
+cospi8sqrt2minus1 DCD 0x4e7b4e7b
+sinpi8sqrt2 DCD 0x8a8c8a8c
+
+ END
--- /dev/null
+++ b/vp8/common/arm/neon/dequantizeb_neon.asm
@@ -1,0 +1,34 @@
+;
+; 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.
+;
+
+
+ EXPORT |vp8_dequantize_b_loop_neon|
+ ARM
+ REQUIRE8
+ PRESERVE8
+
+ AREA ||.text||, CODE, READONLY, ALIGN=2
+; r0 short *Q,
+; r1 short *DQC
+; r2 short *DQ
+|vp8_dequantize_b_loop_neon| PROC
+ vld1.16 {q0, q1}, [r0]+ vld1.16 {q2, q3}, [r1]+
+ vmul.i16 q4, q0, q2
+ vmul.i16 q5, q1, q3
+
+ vst1.16 {q4, q5}, [r2]+
+ bx lr
+
+ ENDP
+
+ END
--- /dev/null
+++ b/vp8/common/arm/neon/idct_blk_neon.c
@@ -1,0 +1,97 @@
+/*
+ * 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_config.h"
+#include "vp8/common/idct.h"
+#include "vp8/common/dequantize.h"
+
+/* place these declarations here because we don't want to maintain them
+ * outside of this scope
+ */
+void idct_dequant_full_2x_neon(short *q, short *dq,
+ unsigned char *dst, int stride);
+void idct_dequant_0_2x_neon(short *q, short dq,
+ unsigned char *dst, int stride);
+
+
+void vp8_dequant_idct_add_y_block_neon(short *q, short *dq,
+ unsigned char *dst,
+ int stride, char *eobs)
+{+ int i;
+
+ for (i = 0; i < 4; i++)
+ {+ if (((short *)(eobs))[0])
+ {+ if (((short *)eobs)[0] & 0xfefe)
+ idct_dequant_full_2x_neon (q, dq, dst, stride);
+ else
+ idct_dequant_0_2x_neon (q, dq[0], dst, stride);
+ }
+
+ if (((short *)(eobs))[1])
+ {+ if (((short *)eobs)[1] & 0xfefe)
+ idct_dequant_full_2x_neon (q+32, dq, dst+8, stride);
+ else
+ idct_dequant_0_2x_neon (q+32, dq[0], dst+8, stride);
+ }
+ q += 64;
+ dst += 4*stride;
+ eobs += 4;
+ }
+}
+
+void vp8_dequant_idct_add_uv_block_neon(short *q, short *dq,
+ unsigned char *dstu,
+ unsigned char *dstv,
+ int stride, char *eobs)
+{+ if (((short *)(eobs))[0])
+ {+ if (((short *)eobs)[0] & 0xfefe)
+ idct_dequant_full_2x_neon (q, dq, dstu, stride);
+ else
+ idct_dequant_0_2x_neon (q, dq[0], dstu, stride);
+ }
+
+ q += 32;
+ dstu += 4*stride;
+
+ if (((short *)(eobs))[1])
+ {+ if (((short *)eobs)[1] & 0xfefe)
+ idct_dequant_full_2x_neon (q, dq, dstu, stride);
+ else
+ idct_dequant_0_2x_neon (q, dq[0], dstu, stride);
+ }
+
+ q += 32;
+
+ if (((short *)(eobs))[2])
+ {+ if (((short *)eobs)[2] & 0xfefe)
+ idct_dequant_full_2x_neon (q, dq, dstv, stride);
+ else
+ idct_dequant_0_2x_neon (q, dq[0], dstv, stride);
+ }
+
+ q += 32;
+ dstv += 4*stride;
+
+ if (((short *)(eobs))[3])
+ {+ if (((short *)eobs)[3] & 0xfefe)
+ idct_dequant_full_2x_neon (q, dq, dstv, stride);
+ else
+ idct_dequant_0_2x_neon (q, dq[0], dstv, stride);
+ }
+}
--- /dev/null
+++ b/vp8/common/arm/neon/idct_dequant_0_2x_neon.asm
@@ -1,0 +1,79 @@
+;
+; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+; Use of this source code is governed by a BSD-style license and patent
+; grant that can be found in the LICENSE file in the root of the source
+; tree. All contributing project authors may be found in the AUTHORS
+; file in the root of the source tree.
+;
+
+
+ EXPORT |idct_dequant_0_2x_neon|
+ ARM
+ REQUIRE8
+ PRESERVE8
+
+ AREA ||.text||, CODE, READONLY, ALIGN=2
+;void idct_dequant_0_2x_neon(short *q, short dq,
+; unsigned char *dst, int stride);
+; r0 *q
+; r1 dq
+; r2 *dst
+; r3 stride
+|idct_dequant_0_2x_neon| PROC
+ push {r4, r5}+
+ add r12, r2, #4
+ vld1.32 {d2[0]}, [r2], r3+ vld1.32 {d8[0]}, [r12], r3+ vld1.32 {d2[1]}, [r2], r3+ vld1.32 {d8[1]}, [r12], r3+ vld1.32 {d4[0]}, [r2], r3+ vld1.32 {d10[0]}, [r12], r3+ vld1.32 {d4[1]}, [r2], r3+ vld1.32 {d10[1]}, [r12], r3+
+ ldrh r12, [r0] ; lo q
+ ldrh r4, [r0, #32] ; hi q
+ mov r5, #0
+ strh r5, [r0]
+ strh r5, [r0, #32]
+
+ sxth r12, r12 ; lo
+ mul r0, r12, r1
+ add r0, r0, #4
+ asr r0, r0, #3
+ vdup.16 q0, r0
+ sxth r4, r4 ; hi
+ mul r0, r4, r1
+ add r0, r0, #4
+ asr r0, r0, #3
+ vdup.16 q3, r0
+
+ vaddw.u8 q1, q0, d2 ; lo
+ vaddw.u8 q2, q0, d4
+ vaddw.u8 q4, q3, d8 ; hi
+ vaddw.u8 q5, q3, d10
+
+ sub r2, r2, r3, lsl #2 ; dst - 4*stride
+ add r0, r2, #4
+
+ vqmovun.s16 d2, q1 ; lo
+ vqmovun.s16 d4, q2
+ vqmovun.s16 d8, q4 ; hi
+ vqmovun.s16 d10, q5
+
+ vst1.32 {d2[0]}, [r2], r3 ; lo+ vst1.32 {d8[0]}, [r0], r3 ; hi+ vst1.32 {d2[1]}, [r2], r3+ vst1.32 {d8[1]}, [r0], r3+ vst1.32 {d4[0]}, [r2], r3+ vst1.32 {d10[0]}, [r0], r3+ vst1.32 {d4[1]}, [r2]+ vst1.32 {d10[1]}, [r0]+
+ pop {r4, r5}+ bx lr
+
+ ENDP ; |idct_dequant_0_2x_neon|
+ END
--- /dev/null
+++ b/vp8/common/arm/neon/idct_dequant_full_2x_neon.asm
@@ -1,0 +1,196 @@
+;
+; 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.
+;
+
+
+ EXPORT |idct_dequant_full_2x_neon|
+ ARM
+ REQUIRE8
+ PRESERVE8
+
+ AREA ||.text||, CODE, READONLY, ALIGN=2
+;void idct_dequant_full_2x_neon(short *q, short *dq,
+; unsigned char *dst, int stride);
+; r0 *q,
+; r1 *dq,
+; r2 *dst
+; r3 stride
+|idct_dequant_full_2x_neon| PROC
+ vld1.16 {q0, q1}, [r1] ; dq (same l/r)+ vld1.16 {q2, q3}, [r0] ; l q+ add r0, r0, #32
+ vld1.16 {q4, q5}, [r0] ; r q+ add r12, r2, #4
+
+ ; interleave the predictors
+ vld1.32 {d28[0]}, [r2], r3 ; l pre+ vld1.32 {d28[1]}, [r12], r3 ; r pre+ vld1.32 {d29[0]}, [r2], r3+ vld1.32 {d29[1]}, [r12], r3+ vld1.32 {d30[0]}, [r2], r3+ vld1.32 {d30[1]}, [r12], r3+ vld1.32 {d31[0]}, [r2], r3+ vld1.32 {d31[1]}, [r12]+
+ adr r1, cospi8sqrt2minus1 ; pointer to the first constant
+
+ ; dequant: q[i] = q[i] * dq[i]
+ vmul.i16 q2, q2, q0
+ vmul.i16 q3, q3, q1
+ vmul.i16 q4, q4, q0
+ vmul.i16 q5, q5, q1
+
+ vld1.16 {d0}, [r1]+
+ ; q2: l0r0 q3: l8r8
+ ; q4: l4r4 q5: l12r12
+ vswp d5, d8
+ vswp d7, d10
+
+ ; _CONSTANTS_ * 4,12 >> 16
+ ; q6: 4 * sinpi : c1/temp1
+ ; q7: 12 * sinpi : d1/temp2
+ ; q8: 4 * cospi
+ ; q9: 12 * cospi
+ vqdmulh.s16 q6, q4, d0[2] ; sinpi8sqrt2
+ vqdmulh.s16 q7, q5, d0[2]
+ vqdmulh.s16 q8, q4, d0[0] ; cospi8sqrt2minus1
+ vqdmulh.s16 q9, q5, d0[0]
+
+ vqadd.s16 q10, q2, q3 ; a1 = 0 + 8
+ vqsub.s16 q11, q2, q3 ; b1 = 0 - 8
+
+ ; vqdmulh only accepts signed values. this was a problem because
+ ; our constant had the high bit set, and was treated as a negative value.
+ ; vqdmulh also doubles the value before it shifts by 16. we need to
+ ; compensate for this. in the case of sinpi8sqrt2, the lowest bit is 0,
+ ; so we can shift the constant without losing precision. this avoids
+ ; shift again afterward, but also avoids the sign issue. win win!
+ ; for cospi8sqrt2minus1 the lowest bit is 1, so we lose precision if we
+ ; pre-shift it
+ vshr.s16 q8, q8, #1
+ vshr.s16 q9, q9, #1
+
+ ; q4: 4 + 4 * cospi : d1/temp1
+ ; q5: 12 + 12 * cospi : c1/temp2
+ vqadd.s16 q4, q4, q8
+ vqadd.s16 q5, q5, q9
+
+ ; c1 = temp1 - temp2
+ ; d1 = temp1 + temp2
+ vqsub.s16 q2, q6, q5
+ vqadd.s16 q3, q4, q7
+
+ ; [0]: a1+d1
+ ; [1]: b1+c1
+ ; [2]: b1-c1
+ ; [3]: a1-d1
+ vqadd.s16 q4, q10, q3
+ vqadd.s16 q5, q11, q2
+ vqsub.s16 q6, q11, q2
+ vqsub.s16 q7, q10, q3
+
+ ; rotate
+ vtrn.32 q4, q6
+ vtrn.32 q5, q7
+ vtrn.16 q4, q5
+ vtrn.16 q6, q7
+ ; idct loop 2
+ ; q4: l 0, 4, 8,12 r 0, 4, 8,12
+ ; q5: l 1, 5, 9,13 r 1, 5, 9,13
+ ; q6: l 2, 6,10,14 r 2, 6,10,14
+ ; q7: l 3, 7,11,15 r 3, 7,11,15
+
+ ; q8: 1 * sinpi : c1/temp1
+ ; q9: 3 * sinpi : d1/temp2
+ ; q10: 1 * cospi
+ ; q11: 3 * cospi
+ vqdmulh.s16 q8, q5, d0[2] ; sinpi8sqrt2
+ vqdmulh.s16 q9, q7, d0[2]
+ vqdmulh.s16 q10, q5, d0[0] ; cospi8sqrt2minus1
+ vqdmulh.s16 q11, q7, d0[0]
+
+ vqadd.s16 q2, q4, q6 ; a1 = 0 + 2
+ vqsub.s16 q3, q4, q6 ; b1 = 0 - 2
+
+ ; see note on shifting above
+ vshr.s16 q10, q10, #1
+ vshr.s16 q11, q11, #1
+
+ ; q10: 1 + 1 * cospi : d1/temp1
+ ; q11: 3 + 3 * cospi : c1/temp2
+ vqadd.s16 q10, q5, q10
+ vqadd.s16 q11, q7, q11
+
+ ; q8: c1 = temp1 - temp2
+ ; q9: d1 = temp1 + temp2
+ vqsub.s16 q8, q8, q11
+ vqadd.s16 q9, q10, q9
+
+ ; a1+d1
+ ; b1+c1
+ ; b1-c1
+ ; a1-d1
+ vqadd.s16 q4, q2, q9
+ vqadd.s16 q5, q3, q8
+ vqsub.s16 q6, q3, q8
+ vqsub.s16 q7, q2, q9
+
+ ; +4 >> 3 (rounding)
+ vrshr.s16 q4, q4, #3 ; lo
+ vrshr.s16 q5, q5, #3
+ vrshr.s16 q6, q6, #3 ; hi
+ vrshr.s16 q7, q7, #3
+
+ vtrn.32 q4, q6
+ vtrn.32 q5, q7
+ vtrn.16 q4, q5
+ vtrn.16 q6, q7
+
+ ; adding pre
+ ; input is still packed. pre was read interleaved
+ vaddw.u8 q4, q4, d28
+ vaddw.u8 q5, q5, d29
+ vaddw.u8 q6, q6, d30
+ vaddw.u8 q7, q7, d31
+
+ vmov.i16 q14, #0
+ vmov q15, q14
+ vst1.16 {q14, q15}, [r0] ; write over high input+ sub r0, r0, #32
+ vst1.16 {q14, q15}, [r0] ; write over low input+
+ sub r2, r2, r3, lsl #2 ; dst - 4*stride
+ add r1, r2, #4 ; hi
+
+ ;saturate and narrow
+ vqmovun.s16 d0, q4 ; lo
+ vqmovun.s16 d1, q5
+ vqmovun.s16 d2, q6 ; hi
+ vqmovun.s16 d3, q7
+
+ vst1.32 {d0[0]}, [r2], r3 ; lo+ vst1.32 {d0[1]}, [r1], r3 ; hi+ vst1.32 {d1[0]}, [r2], r3+ vst1.32 {d1[1]}, [r1], r3+ vst1.32 {d2[0]}, [r2], r3+ vst1.32 {d2[1]}, [r1], r3+ vst1.32 {d3[0]}, [r2]+ vst1.32 {d3[1]}, [r1]+
+ bx lr
+
+ ENDP ; |idct_dequant_full_2x_neon|
+
+; Constant Pool
+cospi8sqrt2minus1 DCD 0x4e7b
+; because the lowest bit in 0x8a8c is 0, we can pre-shift this
+sinpi8sqrt2 DCD 0x4546
+
+ END
--- /dev/null
+++ b/vp8/common/dequantize.c
@@ -1,0 +1,44 @@
+/*
+ * 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_config.h"
+#include "dequantize.h"
+#include "vp8/common/idct.h"
+#include "vpx_mem/vpx_mem.h"
+
+void vp8_dequantize_b_c(BLOCKD *d)
+{+ int i;
+ short *DQ = d->dqcoeff;
+ short *Q = d->qcoeff;
+ short *DQC = d->dequant;
+
+ for (i = 0; i < 16; i++)
+ {+ DQ[i] = Q[i] * DQC[i];
+ }
+}
+
+void vp8_dequant_idct_add_c(short *input, short *dq,
+ unsigned char *dest, int stride)
+{+ int i;
+
+ for (i = 0; i < 16; i++)
+ {+ input[i] = dq[i] * input[i];
+ }
+
+ vp8_short_idct4x4llm_c(input, dest, stride, dest, stride);
+
+ vpx_memset(input, 0, 32);
+
+}
--- /dev/null
+++ b/vp8/common/dequantize.h
@@ -1,0 +1,85 @@
+/*
+ * 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 DEQUANTIZE_H
+#define DEQUANTIZE_H
+#include "vp8/common/blockd.h"
+
+#define prototype_dequant_block(sym) \
+ void sym(BLOCKD *x)
+
+#define prototype_dequant_idct_add(sym) \
+ void sym(short *input, short *dq, \
+ unsigned char *output, \
+ int stride)
+
+#define prototype_dequant_idct_add_y_block(sym) \
+ void sym(short *q, short *dq, \
+ unsigned char *dst, \
+ int stride, char *eobs)
+
+#define prototype_dequant_idct_add_uv_block(sym) \
+ void sym(short *q, short *dq, \
+ unsigned char *dst_u, \
+ unsigned char *dst_v, int stride, char *eobs)
+
+#if ARCH_X86 || ARCH_X86_64
+#include "x86/dequantize_x86.h"
+#endif
+
+#if ARCH_ARM
+#include "arm/dequantize_arm.h"
+#endif
+
+#ifndef vp8_dequant_block
+#define vp8_dequant_block vp8_dequantize_b_c
+#endif
+extern prototype_dequant_block(vp8_dequant_block);
+
+#ifndef vp8_dequant_idct_add
+#define vp8_dequant_idct_add vp8_dequant_idct_add_c
+#endif
+extern prototype_dequant_idct_add(vp8_dequant_idct_add);
+
+#ifndef vp8_dequant_idct_add_y_block
+#define vp8_dequant_idct_add_y_block vp8_dequant_idct_add_y_block_c
+#endif
+extern prototype_dequant_idct_add_y_block(vp8_dequant_idct_add_y_block);
+
+#ifndef vp8_dequant_idct_add_uv_block
+#define vp8_dequant_idct_add_uv_block vp8_dequant_idct_add_uv_block_c
+#endif
+extern prototype_dequant_idct_add_uv_block(vp8_dequant_idct_add_uv_block);
+
+
+typedef prototype_dequant_block((*vp8_dequant_block_fn_t));
+
+typedef prototype_dequant_idct_add((*vp8_dequant_idct_add_fn_t));
+
+typedef prototype_dequant_idct_add_y_block((*vp8_dequant_idct_add_y_block_fn_t));
+
+typedef prototype_dequant_idct_add_uv_block((*vp8_dequant_idct_add_uv_block_fn_t));
+
+typedef struct
+{+ vp8_dequant_block_fn_t block;
+ vp8_dequant_idct_add_fn_t idct_add;
+ vp8_dequant_idct_add_y_block_fn_t idct_add_y_block;
+ vp8_dequant_idct_add_uv_block_fn_t idct_add_uv_block;
+} vp8_dequant_rtcd_vtable_t;
+
+#if CONFIG_RUNTIME_CPU_DETECT
+#define DEQUANT_INVOKE(ctx,fn) (ctx)->fn
+#else
+#define DEQUANT_INVOKE(ctx,fn) vp8_dequant_##fn
+#endif
+
+#endif
--- a/vp8/common/generic/systemdependent.c
+++ b/vp8/common/generic/systemdependent.c
@@ -70,6 +70,14 @@
#if CONFIG_RUNTIME_CPU_DETECT
VP8_COMMON_RTCD *rtcd = &ctx->rtcd;
+
+ rtcd->dequant.block = vp8_dequantize_b_c;
+ rtcd->dequant.idct_add = vp8_dequant_idct_add_c;
+ rtcd->dequant.idct_add_y_block = vp8_dequant_idct_add_y_block_c;
+ rtcd->dequant.idct_add_uv_block =
+ vp8_dequant_idct_add_uv_block_c;
+
+
rtcd->idct.idct16 = vp8_short_idct4x4llm_c;
rtcd->idct.idct1_scalar_add = vp8_dc_only_idct_add_c;
rtcd->idct.iwalsh1 = vp8_short_inv_walsh4x4_1_c;
--- /dev/null
+++ b/vp8/common/idct_blk.c
@@ -1,0 +1,90 @@
+/*
+ * 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_config.h"
+#include "vp8/common/idct.h"
+#include "dequantize.h"
+
+void vp8_dequant_idct_add_c(short *input, short *dq,
+ unsigned char *dest, int stride);
+void vp8_dc_only_idct_add_c(short input_dc, unsigned char * pred,
+ int pred_stride, unsigned char *dst_ptr,
+ int dst_stride);
+
+void vp8_dequant_idct_add_y_block_c
+ (short *q, short *dq,
+ unsigned char *dst, int stride, char *eobs)
+{+ int i, j;
+
+ for (i = 0; i < 4; i++)
+ {+ for (j = 0; j < 4; j++)
+ {+ if (*eobs++ > 1)
+ vp8_dequant_idct_add_c (q, dq, dst, stride);
+ else
+ {+ vp8_dc_only_idct_add_c (q[0]*dq[0], dst, stride, dst, stride);
+ ((int *)q)[0] = 0;
+ }
+
+ q += 16;
+ dst += 4;
+ }
+
+ dst += 4*stride - 16;
+ }
+}
+
+void vp8_dequant_idct_add_uv_block_c
+ (short *q, short *dq,
+ unsigned char *dstu, unsigned char *dstv, int stride, char *eobs)
+{+ int i, j;
+
+ for (i = 0; i < 2; i++)
+ {+ for (j = 0; j < 2; j++)
+ {+ if (*eobs++ > 1)
+ vp8_dequant_idct_add_c (q, dq, dstu, stride);
+ else
+ {+ vp8_dc_only_idct_add_c (q[0]*dq[0], dstu, stride, dstu, stride);
+ ((int *)q)[0] = 0;
+ }
+
+ q += 16;
+ dstu += 4;
+ }
+
+ dstu += 4*stride - 8;
+ }
+
+ for (i = 0; i < 2; i++)
+ {+ for (j = 0; j < 2; j++)
+ {+ if (*eobs++ > 1)
+ vp8_dequant_idct_add_c (q, dq, dstv, stride);
+ else
+ {+ vp8_dc_only_idct_add_c (q[0]*dq[0], dstv, stride, dstv, stride);
+ ((int *)q)[0] = 0;
+ }
+
+ q += 16;
+ dstv += 4;
+ }
+
+ dstv += 4*stride - 8;
+ }
+}
--- a/vp8/common/invtrans.c
+++ /dev/null
@@ -1,56 +1,0 @@
-/*
- * 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 "invtrans.h"
-
-
-void vp8_inverse_transform_b(const vp8_idct_rtcd_vtable_t *rtcd, BLOCKD *b,
- int pitch)
-{- if (*b->eob > 1)
- {- IDCT_INVOKE(rtcd, idct16)(b->dqcoeff, b->predictor, pitch,
- *(b->base_dst) + b->dst, b->dst_stride);
- }
- else
- {- IDCT_INVOKE(rtcd, idct1_scalar_add)(b->dqcoeff[0], b->predictor, pitch,
- *(b->base_dst) + b->dst, b->dst_stride);
- }
-
-}
-
-void vp8_inverse_transform_mby(const vp8_idct_rtcd_vtable_t *rtcd, MACROBLOCKD *x)
-{- int i;
-
- if(x->mode_info_context->mbmi.mode != SPLITMV)
- {- /* do 2nd order transform on the dc block */
- IDCT_INVOKE(rtcd, iwalsh16)(x->block[24].dqcoeff, x->dqcoeff);
- }
-
- for (i = 0; i < 16; i++)
- {- vp8_inverse_transform_b(rtcd, &x->block[i], 16);
- }
-
-}
-void vp8_inverse_transform_mbuv(const vp8_idct_rtcd_vtable_t *rtcd, MACROBLOCKD *x)
-{- int i;
-
- for (i = 16; i < 24; i++)
- {- vp8_inverse_transform_b(rtcd, &x->block[i], 8);
- }
-
-}
--- a/vp8/common/invtrans.h
+++ b/vp8/common/invtrans.h
@@ -15,9 +15,51 @@
#include "vpx_config.h"
#include "idct.h"
#include "blockd.h"
-extern void vp8_inverse_transform_b(const vp8_idct_rtcd_vtable_t *rtcd, BLOCKD *b, int pitch);
-extern void vp8_inverse_transform_mb(const vp8_idct_rtcd_vtable_t *rtcd, MACROBLOCKD *x);
-extern void vp8_inverse_transform_mby(const vp8_idct_rtcd_vtable_t *rtcd, MACROBLOCKD *x);
-extern void vp8_inverse_transform_mbuv(const vp8_idct_rtcd_vtable_t *rtcd, MACROBLOCKD *x);
+#include "onyxc_int.h"
+static void eob_adjust(char *eobs, short *diff)
+{+ /* eob adjust.... the idct can only skip if both the dc and eob are zero */
+ int js;
+ for(js = 0; js < 16; js++)
+ {+ if((eobs[js] == 0) && (diff[0] != 0))
+ eobs[js]++;
+ diff+=16;
+ }
+}
+
+static void vp8_inverse_transform_mby(MACROBLOCKD *xd,
+ const VP8_COMMON_RTCD *rtcd)
+{+ short *DQC = xd->block[0].dequant;
+ /* save the dc dequant constant in case it is overridden */
+ short dc_dequant_temp = DQC[0];
+
+ if (xd->mode_info_context->mbmi.mode != SPLITMV)
+ {+ /* do 2nd order transform on the dc block */
+ if (xd->eobs[24] > 1)
+ {+ IDCT_INVOKE(&rtcd->idct, iwalsh16)
+ (&xd->block[24].dqcoeff[0], xd->qcoeff);
+ }
+ else
+ {+ IDCT_INVOKE(&rtcd->idct, iwalsh1)
+ (&xd->block[24].dqcoeff[0], xd->qcoeff);
+ }
+ eob_adjust(xd->eobs, xd->qcoeff);
+
+ /* override the dc dequant constant */
+ DQC[0] = 1;
+ }
+ DEQUANT_INVOKE (&rtcd->dequant, idct_add_y_block)
+ (xd->qcoeff, xd->block[0].dequant,
+ xd->dst.y_buffer,
+ xd->dst.y_stride, xd->eobs);
+
+ /* restore the dc dequant constant */
+ DQC[0] = dc_dequant_temp;
+}
#endif
--- a/vp8/common/onyxc_int.h
+++ b/vp8/common/onyxc_int.h
@@ -22,6 +22,7 @@
#if CONFIG_POSTPROC
#include "postproc.h"
#endif
+#include "dequantize.h"
/*#ifdef PACKET_TESTING*/
#include "header.h"
@@ -73,6 +74,7 @@
typedef struct VP8_COMMON_RTCD
{#if CONFIG_RUNTIME_CPU_DETECT
+ vp8_dequant_rtcd_vtable_t dequant;
vp8_idct_rtcd_vtable_t idct;
vp8_recon_rtcd_vtable_t recon;
vp8_subpix_rtcd_vtable_t subpix;
--- a/vp8/common/reconinter.c
+++ b/vp8/common/reconinter.c
@@ -334,11 +334,12 @@
/*encoder only*/
-void vp8_build_inter16x16_predictors_mby(MACROBLOCKD *x)
+void vp8_build_inter16x16_predictors_mby(MACROBLOCKD *x,
+ unsigned char *dst_y,
+ int dst_ystride)
{unsigned char *ptr_base;
unsigned char *ptr;
- unsigned char *pred_ptr = x->predictor;
int mv_row = x->mode_info_context->mbmi.mv.as_mv.row;
int mv_col = x->mode_info_context->mbmi.mv.as_mv.col;
int pre_stride = x->block[0].pre_stride;
@@ -348,11 +349,13 @@
if ((mv_row | mv_col) & 7)
{- x->subpixel_predict16x16(ptr, pre_stride, mv_col & 7, mv_row & 7, pred_ptr, 16);
+ x->subpixel_predict16x16(ptr, pre_stride, mv_col & 7, mv_row & 7,
+ dst_y, dst_ystride);
}
else
{- RECON_INVOKE(&x->rtcd->recon, copy16x16)(ptr, pre_stride, pred_ptr, 16);
+ RECON_INVOKE(&x->rtcd->recon, copy16x16)(ptr, pre_stride, dst_y,
+ dst_ystride);
}
}
@@ -594,71 +597,5 @@
{build_4x4uvmvs(xd);
build_inter4x4_predictors_mb(xd);
- }
-}
-/* encoder only*/
-static void build_inter4x4_predictors_mb_e(MACROBLOCKD *x)
-{- int i;
-
- if (x->mode_info_context->mbmi.partitioning < 3)
- {- x->block[ 0].bmi = x->mode_info_context->bmi[ 0];
- x->block[ 2].bmi = x->mode_info_context->bmi[ 2];
- x->block[ 8].bmi = x->mode_info_context->bmi[ 8];
- x->block[10].bmi = x->mode_info_context->bmi[10];
-
- build_inter_predictors4b(x, &x->block[ 0], x->block[ 0].predictor, 16);
- build_inter_predictors4b(x, &x->block[ 2], x->block[ 2].predictor, 16);
- build_inter_predictors4b(x, &x->block[ 8], x->block[ 8].predictor, 16);
- build_inter_predictors4b(x, &x->block[10], x->block[10].predictor, 16);
- }
- else
- {- for (i = 0; i < 16; i += 2)
- {- BLOCKD *d0 = &x->block[i];
- BLOCKD *d1 = &x->block[i+1];
-
- x->block[i+0].bmi = x->mode_info_context->bmi[i+0];
- x->block[i+1].bmi = x->mode_info_context->bmi[i+1];
-
- if (d0->bmi.mv.as_int == d1->bmi.mv.as_int)
- build_inter_predictors2b(x, d0, d0->predictor, 16);
- else
- {- build_inter_predictors_b(d0, d0->predictor, 16, x->subpixel_predict);
- build_inter_predictors_b(d1, d1->predictor, 16, x->subpixel_predict);
- }
-
- }
-
- }
-
- for (i = 16; i < 24; i += 2)
- {- BLOCKD *d0 = &x->block[i];
- BLOCKD *d1 = &x->block[i+1];
-
- if (d0->bmi.mv.as_int == d1->bmi.mv.as_int)
- build_inter_predictors2b(x, d0, d0->predictor, 8);
- else
- {- build_inter_predictors_b(d0, d0->predictor, 8, x->subpixel_predict);
- build_inter_predictors_b(d1, d1->predictor, 8, x->subpixel_predict);
- }
- }
-}
-void vp8_build_inter_predictors_mb_e(MACROBLOCKD *xd)
-{- if (xd->mode_info_context->mbmi.mode != SPLITMV)
- {- vp8_build_inter16x16_predictors_mb(xd, xd->predictor, &xd->predictor[256],
- &xd->predictor[320], 16, 8);
- }
- else
- {- build_4x4uvmvs(xd);
- build_inter4x4_predictors_mb_e(xd);
}
}
--- a/vp8/common/reconinter.h
+++ b/vp8/common/reconinter.h
@@ -21,11 +21,13 @@
int dst_uvstride);
-extern void vp8_build_inter16x16_predictors_mby(MACROBLOCKD *x);
-extern void vp8_build_inter_predictors_b(BLOCKD *d, int pitch, vp8_subpix_fn_t sppf);
+extern void vp8_build_inter16x16_predictors_mby(MACROBLOCKD *x,
+ unsigned char *dst_y,
+ int dst_ystride);
+extern void vp8_build_inter_predictors_b(BLOCKD *d, int pitch,
+ vp8_subpix_fn_t sppf);
extern void vp8_build_inter16x16_predictors_mbuv(MACROBLOCKD *x);
extern void vp8_build_inter4x4_predictors_mbuv(MACROBLOCKD *x);
-extern void vp8_build_inter_predictors_mb_e(MACROBLOCKD *xd);
#endif
--- /dev/null
+++ b/vp8/common/x86/dequantize_mmx.asm
@@ -1,0 +1,258 @@
+;
+; 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/x86_abi_support.asm"
+
+
+;void vp8_dequantize_b_impl_mmx(short *sq, short *dq, short *q)
+global sym(vp8_dequantize_b_impl_mmx)
+sym(vp8_dequantize_b_impl_mmx):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 3
+ push rsi
+ push rdi
+ ; end prolog
+
+ mov rsi, arg(0) ;sq
+ mov rdi, arg(1) ;dq
+ mov rax, arg(2) ;q
+
+ movq mm1, [rsi]
+ pmullw mm1, [rax+0] ; mm4 *= kernel 0 modifiers.
+ movq [rdi], mm1
+
+ movq mm1, [rsi+8]
+ pmullw mm1, [rax+8] ; mm4 *= kernel 0 modifiers.
+ movq [rdi+8], mm1
+
+ movq mm1, [rsi+16]
+ pmullw mm1, [rax+16] ; mm4 *= kernel 0 modifiers.
+ movq [rdi+16], mm1
+
+ movq mm1, [rsi+24]
+ pmullw mm1, [rax+24] ; mm4 *= kernel 0 modifiers.
+ movq [rdi+24], mm1
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+
+;void dequant_idct_add_mmx(
+;short *input, 0
+;short *dq, 1
+;unsigned char *dest, 2
+;int stride) 3
+global sym(vp8_dequant_idct_add_mmx)
+sym(vp8_dequant_idct_add_mmx):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 4
+ GET_GOT rbx
+ push rdi
+ ; end prolog
+
+ mov rax, arg(0) ;input
+ mov rdx, arg(1) ;dq
+
+
+ movq mm0, [rax ]
+ pmullw mm0, [rdx]
+
+ movq mm1, [rax +8]
+ pmullw mm1, [rdx +8]
+
+ movq mm2, [rax+16]
+ pmullw mm2, [rdx+16]
+
+ movq mm3, [rax+24]
+ pmullw mm3, [rdx+24]
+
+ mov rdx, arg(2) ;dest
+
+ pxor mm7, mm7
+
+
+ movq [rax], mm7
+ movq [rax+8], mm7
+
+ movq [rax+16],mm7
+ movq [rax+24],mm7
+
+
+ movsxd rdi, dword ptr arg(3) ;stride
+
+ psubw mm0, mm2 ; b1= 0-2
+ paddw mm2, mm2 ;
+
+ movq mm5, mm1
+ paddw mm2, mm0 ; a1 =0+2
+
+ pmulhw mm5, [GLOBAL(x_s1sqr2)];
+ paddw mm5, mm1 ; ip1 * sin(pi/8) * sqrt(2)
+
+ movq mm7, mm3 ;
+ pmulhw mm7, [GLOBAL(x_c1sqr2less1)];
+
+ paddw mm7, mm3 ; ip3 * cos(pi/8) * sqrt(2)
+ psubw mm7, mm5 ; c1
+
+ movq mm5, mm1
+ movq mm4, mm3
+
+ pmulhw mm5, [GLOBAL(x_c1sqr2less1)]
+ paddw mm5, mm1
+
+ pmulhw mm3, [GLOBAL(x_s1sqr2)]
+ paddw mm3, mm4
+
+ paddw mm3, mm5 ; d1
+ movq mm6, mm2 ; a1
+
+ movq mm4, mm0 ; b1
+ paddw mm2, mm3 ;0
+
+ paddw mm4, mm7 ;1
+ psubw mm0, mm7 ;2
+
+ psubw mm6, mm3 ;3
+
+ movq mm1, mm2 ; 03 02 01 00
+ movq mm3, mm4 ; 23 22 21 20
+
+ punpcklwd mm1, mm0 ; 11 01 10 00
+ punpckhwd mm2, mm0 ; 13 03 12 02
+
+ punpcklwd mm3, mm6 ; 31 21 30 20
+ punpckhwd mm4, mm6 ; 33 23 32 22
+
+ movq mm0, mm1 ; 11 01 10 00
+ movq mm5, mm2 ; 13 03 12 02
+
+ punpckldq mm0, mm3 ; 30 20 10 00
+ punpckhdq mm1, mm3 ; 31 21 11 01
+
+ punpckldq mm2, mm4 ; 32 22 12 02
+ punpckhdq mm5, mm4 ; 33 23 13 03
+
+ movq mm3, mm5 ; 33 23 13 03
+
+ psubw mm0, mm2 ; b1= 0-2
+ paddw mm2, mm2 ;
+
+ movq mm5, mm1
+ paddw mm2, mm0 ; a1 =0+2
+
+ pmulhw mm5, [GLOBAL(x_s1sqr2)];
+ paddw mm5, mm1 ; ip1 * sin(pi/8) * sqrt(2)
+
+ movq mm7, mm3 ;
+ pmulhw mm7, [GLOBAL(x_c1sqr2less1)];
+
+ paddw mm7, mm3 ; ip3 * cos(pi/8) * sqrt(2)
+ psubw mm7, mm5 ; c1
+
+ movq mm5, mm1
+ movq mm4, mm3
+
+ pmulhw mm5, [GLOBAL(x_c1sqr2less1)]
+ paddw mm5, mm1
+
+ pmulhw mm3, [GLOBAL(x_s1sqr2)]
+ paddw mm3, mm4
+
+ paddw mm3, mm5 ; d1
+ paddw mm0, [GLOBAL(fours)]
+
+ paddw mm2, [GLOBAL(fours)]
+ movq mm6, mm2 ; a1
+
+ movq mm4, mm0 ; b1
+ paddw mm2, mm3 ;0
+
+ paddw mm4, mm7 ;1
+ psubw mm0, mm7 ;2
+
+ psubw mm6, mm3 ;3
+ psraw mm2, 3
+
+ psraw mm0, 3
+ psraw mm4, 3
+
+ psraw mm6, 3
+
+ movq mm1, mm2 ; 03 02 01 00
+ movq mm3, mm4 ; 23 22 21 20
+
+ punpcklwd mm1, mm0 ; 11 01 10 00
+ punpckhwd mm2, mm0 ; 13 03 12 02
+
+ punpcklwd mm3, mm6 ; 31 21 30 20
+ punpckhwd mm4, mm6 ; 33 23 32 22
+
+ movq mm0, mm1 ; 11 01 10 00
+ movq mm5, mm2 ; 13 03 12 02
+
+ punpckldq mm0, mm3 ; 30 20 10 00
+ punpckhdq mm1, mm3 ; 31 21 11 01
+
+ punpckldq mm2, mm4 ; 32 22 12 02
+ punpckhdq mm5, mm4 ; 33 23 13 03
+
+ pxor mm7, mm7
+
+ movd mm4, [rdx]
+ punpcklbw mm4, mm7
+ paddsw mm0, mm4
+ packuswb mm0, mm7
+ movd [rdx], mm0
+
+ movd mm4, [rdx+rdi]
+ punpcklbw mm4, mm7
+ paddsw mm1, mm4
+ packuswb mm1, mm7
+ movd [rdx+rdi], mm1
+
+ movd mm4, [rdx+2*rdi]
+ punpcklbw mm4, mm7
+ paddsw mm2, mm4
+ packuswb mm2, mm7
+ movd [rdx+rdi*2], mm2
+
+ add rdx, rdi
+
+ movd mm4, [rdx+2*rdi]
+ punpcklbw mm4, mm7
+ paddsw mm5, mm4
+ packuswb mm5, mm7
+ movd [rdx+rdi*2], mm5
+
+ ; begin epilog
+ pop rdi
+ RESTORE_GOT
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+SECTION_RODATA
+align 16
+x_s1sqr2:
+ times 4 dw 0x8A8C
+align 16
+x_c1sqr2less1:
+ times 4 dw 0x4E7B
+align 16
+fours:
+ times 4 dw 0x0004
--- /dev/null
+++ b/vp8/common/x86/dequantize_x86.h
@@ -1,0 +1,58 @@
+/*
+ * 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 DEQUANTIZE_X86_H
+#define DEQUANTIZE_X86_H
+
+
+/* Note:
+ *
+ * This platform is commonly built for runtime CPU detection. If you modify
+ * any of the function mappings present in this file, be sure to also update
+ * them in the function pointer initialization code
+ */
+#if HAVE_MMX
+extern prototype_dequant_block(vp8_dequantize_b_mmx);
+extern prototype_dequant_idct_add(vp8_dequant_idct_add_mmx);
+extern prototype_dequant_idct_add_y_block(vp8_dequant_idct_add_y_block_mmx);
+extern prototype_dequant_idct_add_uv_block(vp8_dequant_idct_add_uv_block_mmx);
+
+#if !CONFIG_RUNTIME_CPU_DETECT
+#undef vp8_dequant_block
+#define vp8_dequant_block vp8_dequantize_b_mmx
+
+#undef vp8_dequant_idct_add
+#define vp8_dequant_idct_add vp8_dequant_idct_add_mmx
+
+#undef vp8_dequant_idct_add_y_block
+#define vp8_dequant_idct_add_y_block vp8_dequant_idct_add_y_block_mmx
+
+#undef vp8_dequant_idct_add_uv_block
+#define vp8_dequant_idct_add_uv_block vp8_dequant_idct_add_uv_block_mmx
+
+#endif
+#endif
+
+#if HAVE_SSE2
+extern prototype_dequant_idct_add_y_block(vp8_dequant_idct_add_y_block_sse2);
+extern prototype_dequant_idct_add_uv_block(vp8_dequant_idct_add_uv_block_sse2);
+
+#if !CONFIG_RUNTIME_CPU_DETECT
+#undef vp8_dequant_idct_add_y_block
+#define vp8_dequant_idct_add_y_block vp8_dequant_idct_add_y_block_sse2
+
+#undef vp8_dequant_idct_add_uv_block
+#define vp8_dequant_idct_add_uv_block vp8_dequant_idct_add_uv_block_sse2
+
+#endif
+#endif
+
+#endif
--- /dev/null
+++ b/vp8/common/x86/idct_blk_mmx.c
@@ -1,0 +1,127 @@
+/*
+ * 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_config.h"
+#include "vp8/common/idct.h"
+#include "vp8/common/dequantize.h"
+
+extern void vp8_dequantize_b_impl_mmx(short *sq, short *dq, short *q);
+
+void vp8_dequantize_b_mmx(BLOCKD *d)
+{+ short *sq = (short *) d->qcoeff;
+ short *dq = (short *) d->dqcoeff;
+ short *q = (short *) d->dequant;
+ vp8_dequantize_b_impl_mmx(sq, dq, q);
+}
+
+void vp8_dequant_idct_add_y_block_mmx
+ (short *q, short *dq,
+ unsigned char *dst, int stride, char *eobs)
+{+ int i;
+
+ for (i = 0; i < 4; i++)
+ {+ if (eobs[0] > 1)
+ vp8_dequant_idct_add_mmx (q, dq, dst, stride);
+ else if (eobs[0] == 1)
+ {+ vp8_dc_only_idct_add_mmx (q[0]*dq[0], dst, stride, dst, stride);
+ ((int *)q)[0] = 0;
+ }
+
+ if (eobs[1] > 1)
+ vp8_dequant_idct_add_mmx (q+16, dq, dst+4, stride);
+ else if (eobs[1] == 1)
+ {+ vp8_dc_only_idct_add_mmx (q[16]*dq[0], dst+4, stride,
+ dst+4, stride);
+ ((int *)(q+16))[0] = 0;
+ }
+
+ if (eobs[2] > 1)
+ vp8_dequant_idct_add_mmx (q+32, dq, dst+8, stride);
+ else if (eobs[2] == 1)
+ {+ vp8_dc_only_idct_add_mmx (q[32]*dq[0], dst+8, stride,
+ dst+8, stride);
+ ((int *)(q+32))[0] = 0;
+ }
+
+ if (eobs[3] > 1)
+ vp8_dequant_idct_add_mmx (q+48, dq, dst+12, stride);
+ else if (eobs[3] == 1)
+ {+ vp8_dc_only_idct_add_mmx (q[48]*dq[0], dst+12, stride,
+ dst+12, stride);
+ ((int *)(q+48))[0] = 0;
+ }
+
+ q += 64;
+ dst += 4*stride;
+ eobs += 4;
+ }
+}
+
+void vp8_dequant_idct_add_uv_block_mmx
+ (short *q, short *dq,
+ unsigned char *dstu, unsigned char *dstv, int stride, char *eobs)
+{+ int i;
+
+ for (i = 0; i < 2; i++)
+ {+ if (eobs[0] > 1)
+ vp8_dequant_idct_add_mmx (q, dq, dstu, stride);
+ else if (eobs[0] == 1)
+ {+ vp8_dc_only_idct_add_mmx (q[0]*dq[0], dstu, stride, dstu, stride);
+ ((int *)q)[0] = 0;
+ }
+
+ if (eobs[1] > 1)
+ vp8_dequant_idct_add_mmx (q+16, dq, dstu+4, stride);
+ else if (eobs[1] == 1)
+ {+ vp8_dc_only_idct_add_mmx (q[16]*dq[0], dstu+4, stride,
+ dstu+4, stride);
+ ((int *)(q+16))[0] = 0;
+ }
+
+ q += 32;
+ dstu += 4*stride;
+ eobs += 2;
+ }
+
+ for (i = 0; i < 2; i++)
+ {+ if (eobs[0] > 1)
+ vp8_dequant_idct_add_mmx (q, dq, dstv, stride);
+ else if (eobs[0] == 1)
+ {+ vp8_dc_only_idct_add_mmx (q[0]*dq[0], dstv, stride, dstv, stride);
+ ((int *)q)[0] = 0;
+ }
+
+ if (eobs[1] > 1)
+ vp8_dequant_idct_add_mmx (q+16, dq, dstv+4, stride);
+ else if (eobs[1] == 1)
+ {+ vp8_dc_only_idct_add_mmx (q[16]*dq[0], dstv+4, stride,
+ dstv+4, stride);
+ ((int *)(q+16))[0] = 0;
+ }
+
+ q += 32;
+ dstv += 4*stride;
+ eobs += 2;
+ }
+}
--- /dev/null
+++ b/vp8/common/x86/idct_blk_sse2.c
@@ -1,0 +1,90 @@
+/*
+ * 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_config.h"
+#include "vp8/common/idct.h"
+#include "vp8/common/dequantize.h"
+
+void vp8_idct_dequant_0_2x_sse2
+ (short *q, short *dq ,
+ unsigned char *dst, int dst_stride);
+void vp8_idct_dequant_full_2x_sse2
+ (short *q, short *dq ,
+ unsigned char *dst, int dst_stride);
+
+void vp8_dequant_idct_add_y_block_sse2
+ (short *q, short *dq,
+ unsigned char *dst, int stride, char *eobs)
+{+ int i;
+
+ for (i = 0; i < 4; i++)
+ {+ if (((short *)(eobs))[0])
+ {+ if (((short *)(eobs))[0] & 0xfefe)
+ vp8_idct_dequant_full_2x_sse2 (q, dq, dst, stride);
+ else
+ vp8_idct_dequant_0_2x_sse2 (q, dq, dst, stride);
+ }
+ if (((short *)(eobs))[1])
+ {+ if (((short *)(eobs))[1] & 0xfefe)
+ vp8_idct_dequant_full_2x_sse2 (q+32, dq, dst+8, stride);
+ else
+ vp8_idct_dequant_0_2x_sse2 (q+32, dq, dst+8, stride);
+ }
+ q += 64;
+ dst += stride*4;
+ eobs += 4;
+ }
+}
+
+void vp8_dequant_idct_add_uv_block_sse2
+ (short *q, short *dq,
+ unsigned char *dstu, unsigned char *dstv, int stride, char *eobs)
+{+ if (((short *)(eobs))[0])
+ {+ if (((short *)(eobs))[0] & 0xfefe)
+ vp8_idct_dequant_full_2x_sse2 (q, dq, dstu, stride);
+ else
+ vp8_idct_dequant_0_2x_sse2 (q, dq, dstu, stride);
+ }
+ q += 32;
+ dstu += stride*4;
+
+ if (((short *)(eobs))[1])
+ {+ if (((short *)(eobs))[1] & 0xfefe)
+ vp8_idct_dequant_full_2x_sse2 (q, dq, dstu, stride);
+ else
+ vp8_idct_dequant_0_2x_sse2 (q, dq, dstu, stride);
+ }
+ q += 32;
+
+ if (((short *)(eobs))[2])
+ {+ if (((short *)(eobs))[2] & 0xfefe)
+ vp8_idct_dequant_full_2x_sse2 (q, dq, dstv, stride);
+ else
+ vp8_idct_dequant_0_2x_sse2 (q, dq, dstv, stride);
+ }
+ q += 32;
+ dstv += stride*4;
+
+ if (((short *)(eobs))[3])
+ {+ if (((short *)(eobs))[3] & 0xfefe)
+ vp8_idct_dequant_full_2x_sse2 (q, dq, dstv, stride);
+ else
+ vp8_idct_dequant_0_2x_sse2 (q, dq, dstv, stride);
+ }
+}
--- a/vp8/common/x86/x86_systemdependent.c
+++ b/vp8/common/x86/x86_systemdependent.c
@@ -37,6 +37,11 @@
if (flags & HAS_MMX)
{+ rtcd->dequant.block = vp8_dequantize_b_mmx;
+ rtcd->dequant.idct_add = vp8_dequant_idct_add_mmx;
+ rtcd->dequant.idct_add_y_block = vp8_dequant_idct_add_y_block_mmx;
+ rtcd->dequant.idct_add_uv_block = vp8_dequant_idct_add_uv_block_mmx;
+
rtcd->idct.idct16 = vp8_short_idct4x4llm_mmx;
rtcd->idct.idct1_scalar_add = vp8_dc_only_idct_add_mmx;
rtcd->idct.iwalsh16 = vp8_short_inv_walsh4x4_mmx;
@@ -81,6 +86,9 @@
vp8_build_intra_predictors_mbuv_sse2;
rtcd->recon.build_intra_predictors_mbuv_s =
vp8_build_intra_predictors_mbuv_s_sse2;
+
+ rtcd->dequant.idct_add_y_block = vp8_dequant_idct_add_y_block_sse2;
+ rtcd->dequant.idct_add_uv_block = vp8_dequant_idct_add_uv_block_sse2;
rtcd->idct.iwalsh16 = vp8_short_inv_walsh4x4_sse2;
--- a/vp8/decoder/arm/arm_dsystemdependent.c
+++ b/vp8/decoder/arm/arm_dsystemdependent.c
@@ -11,9 +11,6 @@
#include "vpx_config.h"
#include "vpx_ports/arm.h"
-#include "vp8/common/blockd.h"
-#include "vp8/common/pragmas.h"
-#include "vp8/decoder/dequantize.h"
#include "vp8/decoder/onyxd_int.h"
void vp8_arch_arm_decode_init(VP8D_COMP *pbi)
@@ -30,10 +27,6 @@
#if HAVE_ARMV6
if (flags & HAS_MEDIA)
{- pbi->dequant.block = vp8_dequantize_b_v6;
- pbi->dequant.idct_add = vp8_dequant_idct_add_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;
}
#endif
@@ -40,10 +33,6 @@
#if HAVE_ARMV7
if (flags & HAS_NEON)
{- pbi->dequant.block = vp8_dequantize_b_neon;
- pbi->dequant.idct_add = vp8_dequant_idct_add_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;
}
#endif
#endif
--- a/vp8/decoder/arm/armv6/dequant_idct_v6.asm
+++ /dev/null
@@ -1,190 +1,0 @@
-;
-; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
-;
-; Use of this source code is governed by a BSD-style license and patent
-; grant that can be found in the LICENSE file in the root of the source
-; tree. All contributing project authors may be found in the AUTHORS
-; file in the root of the source tree.
-;
-
- EXPORT |vp8_dequant_idct_add_v6|
-
- AREA |.text|, CODE, READONLY
-;void vp8_dequant_idct_v6(short *input, short *dq,
-; unsigned char *dest, int stride)
-; r0 = q
-; r1 = dq
-; r2 = dst
-; r3 = stride
-
-|vp8_dequant_idct_add_v6| PROC
- stmdb sp!, {r4-r11, lr}-
- ldr r4, [r0] ;input
- ldr r5, [r1], #4 ;dq
-
- sub sp, sp, #4
- str r3, [sp]
-
- mov r12, #4
-
-vp8_dequant_add_loop
- smulbb r6, r4, r5
- smultt r7, r4, r5
-
- ldr r4, [r0, #4] ;input
- ldr r5, [r1], #4 ;dq
-
- strh r6, [r0], #2
- strh r7, [r0], #2
-
- smulbb r6, r4, r5
- smultt r7, r4, r5
-
- subs r12, r12, #1
-
- ldrne r4, [r0, #4]
- ldrne r5, [r1], #4
-
- strh r6, [r0], #2
- strh r7, [r0], #2
-
- bne vp8_dequant_add_loop
-
- sub r0, r0, #32
- mov r1, r0
-
-; short_idct4x4llm_v6_dual
- ldr r3, cospi8sqrt2minus1
- ldr r4, sinpi8sqrt2
- ldr r6, [r0, #8]
- mov r5, #2
-vp8_dequant_idct_loop1_v6
- ldr r12, [r0, #24]
- ldr r14, [r0, #16]
- smulwt r9, r3, r6
- smulwb r7, r3, r6
- smulwt r10, r4, r6
- smulwb r8, r4, r6
- pkhbt r7, r7, r9, lsl #16
- smulwt r11, r3, r12
- pkhbt r8, r8, r10, lsl #16
- uadd16 r6, r6, r7
- smulwt r7, r4, r12
- smulwb r9, r3, r12
- smulwb r10, r4, r12
- subs r5, r5, #1
- pkhbt r9, r9, r11, lsl #16
- ldr r11, [r0], #4
- pkhbt r10, r10, r7, lsl #16
- uadd16 r7, r12, r9
- usub16 r7, r8, r7
- uadd16 r6, r6, r10
- uadd16 r10, r11, r14
- usub16 r8, r11, r14
- uadd16 r9, r10, r6
- usub16 r10, r10, r6
- uadd16 r6, r8, r7
- usub16 r7, r8, r7
- str r6, [r1, #8]
- ldrne r6, [r0, #8]
- str r7, [r1, #16]
- str r10, [r1, #24]
- str r9, [r1], #4
- bne vp8_dequant_idct_loop1_v6
-
- mov r5, #2
- sub r0, r1, #8
-vp8_dequant_idct_loop2_v6
- ldr r6, [r0], #4
- ldr r7, [r0], #4
- ldr r8, [r0], #4
- ldr r9, [r0], #4
- smulwt r1, r3, r6
- smulwt r12, r4, r6
- smulwt lr, r3, r8
- smulwt r10, r4, r8
- pkhbt r11, r8, r6, lsl #16
- pkhbt r1, lr, r1, lsl #16
- pkhbt r12, r10, r12, lsl #16
- pkhtb r6, r6, r8, asr #16
- uadd16 r6, r1, r6
- pkhbt lr, r9, r7, lsl #16
- uadd16 r10, r11, lr
- usub16 lr, r11, lr
- pkhtb r8, r7, r9, asr #16
- subs r5, r5, #1
- smulwt r1, r3, r8
- smulwb r7, r3, r8
- smulwt r11, r4, r8
- smulwb r9, r4, r8
- pkhbt r1, r7, r1, lsl #16
- uadd16 r8, r1, r8
- pkhbt r11, r9, r11, lsl #16
- usub16 r1, r12, r8
- uadd16 r8, r11, r6
- ldr r9, c0x00040004
- ldr r12, [sp] ; get stride from stack
- uadd16 r6, r10, r8
- usub16 r7, r10, r8
- uadd16 r7, r7, r9
- uadd16 r6, r6, r9
- uadd16 r10, r14, r1
- usub16 r1, r14, r1
- uadd16 r10, r10, r9
- uadd16 r1, r1, r9
- ldr r11, [r2] ; load input from dst
- mov r8, r7, asr #3
- pkhtb r9, r8, r10, asr #19
- mov r8, r1, asr #3
- pkhtb r8, r8, r6, asr #19
- uxtb16 lr, r11, ror #8
- qadd16 r9, r9, lr
- uxtb16 lr, r11
- qadd16 r8, r8, lr
- usat16 r9, #8, r9
- usat16 r8, #8, r8
- orr r9, r8, r9, lsl #8
- ldr r11, [r2, r12] ; load input from dst
- mov r7, r7, lsl #16
- mov r1, r1, lsl #16
- mov r10, r10, lsl #16
- mov r6, r6, lsl #16
- mov r7, r7, asr #3
- pkhtb r7, r7, r10, asr #19
- mov r1, r1, asr #3
- pkhtb r1, r1, r6, asr #19
- uxtb16 r8, r11, ror #8
- qadd16 r7, r7, r8
- uxtb16 r8, r11
- qadd16 r1, r1, r8
- usat16 r7, #8, r7
- usat16 r1, #8, r1
- orr r1, r1, r7, lsl #8
- str r9, [r2], r12 ; store output to dst
- str r1, [r2], r12 ; store output to dst
- bne vp8_dequant_idct_loop2_v6
-
-; vpx_memset
- sub r0, r0, #32
- add sp, sp, #4
-
- mov r12, #0
- str r12, [r0]
- str r12, [r0, #4]
- str r12, [r0, #8]
- str r12, [r0, #12]
- str r12, [r0, #16]
- str r12, [r0, #20]
- str r12, [r0, #24]
- str r12, [r0, #28]
-
- ldmia sp!, {r4 - r11, pc}- ENDP ; |vp8_dequant_idct_add_v6|
-
-; Constant Pool
-cospi8sqrt2minus1 DCD 0x00004E7B
-sinpi8sqrt2 DCD 0x00008A8C
-c0x00040004 DCD 0x00040004
-
- END
--- a/vp8/decoder/arm/armv6/dequantize_v6.asm
+++ /dev/null
@@ -1,69 +1,0 @@
-;
-; 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.
-;
-
-
- EXPORT |vp8_dequantize_b_loop_v6|
-
- AREA |.text|, CODE, READONLY ; name this block of code
-;-------------------------------
-;void vp8_dequantize_b_loop_v6(short *Q, short *DQC, short *DQ);
-; r0 short *Q,
-; r1 short *DQC
-; r2 short *DQ
-|vp8_dequantize_b_loop_v6| PROC
- stmdb sp!, {r4-r9, lr}-
- ldr r3, [r0] ;load Q
- ldr r4, [r1] ;load DQC
- ldr r5, [r0, #4]
- ldr r6, [r1, #4]
-
- mov r12, #2 ;loop counter
-
-dequant_loop
- smulbb r7, r3, r4 ;multiply
- smultt r8, r3, r4
- smulbb r9, r5, r6
- smultt lr, r5, r6
-
- ldr r3, [r0, #8]
- ldr r4, [r1, #8]
- ldr r5, [r0, #12]
- ldr r6, [r1, #12]
-
- strh r7, [r2], #2 ;store result
- smulbb r7, r3, r4 ;multiply
- strh r8, [r2], #2
- smultt r8, r3, r4
- strh r9, [r2], #2
- smulbb r9, r5, r6
- strh lr, [r2], #2
- smultt lr, r5, r6
-
- subs r12, r12, #1
-
- add r0, r0, #16
- add r1, r1, #16
-
- ldrne r3, [r0]
- strh r7, [r2], #2 ;store result
- ldrne r4, [r1]
- strh r8, [r2], #2
- ldrne r5, [r0, #4]
- strh r9, [r2], #2
- ldrne r6, [r1, #4]
- strh lr, [r2], #2
-
- bne dequant_loop
-
- ldmia sp!, {r4-r9, pc}- ENDP ;|vp8_dequantize_b_loop_v6|
-
- END
--- a/vp8/decoder/arm/armv6/idct_blk_v6.c
+++ /dev/null
@@ -1,116 +1,0 @@
-/*
- * 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_config.h"
-#include "vp8/common/idct.h"
-#include "vp8/decoder/dequantize.h"
-
-
-void vp8_dequant_idct_add_y_block_v6(short *q, short *dq,
- unsigned char *dst,
- int stride, char *eobs)
-{- int i;
-
- for (i = 0; i < 4; i++)
- {- if (eobs[0] > 1)
- vp8_dequant_idct_add_v6 (q, dq, dst, stride);
- else if (eobs[0] == 1)
- {- vp8_dc_only_idct_add_v6 (q[0]*dq[0], dst, stride, dst, stride);
- ((int *)q)[0] = 0;
- }
-
- if (eobs[1] > 1)
- vp8_dequant_idct_add_v6 (q+16, dq, dst+4, stride);
- else if (eobs[1] == 1)
- {- vp8_dc_only_idct_add_v6 (q[16]*dq[0], dst+4, stride, dst+4, stride);
- ((int *)(q+16))[0] = 0;
- }
-
- if (eobs[2] > 1)
- vp8_dequant_idct_add_v6 (q+32, dq, dst+8, stride);
- else if (eobs[2] == 1)
- {- vp8_dc_only_idct_add_v6 (q[32]*dq[0], dst+8, stride, dst+8, stride);
- ((int *)(q+32))[0] = 0;
- }
-
- if (eobs[3] > 1)
- vp8_dequant_idct_add_v6 (q+48, dq, dst+12, stride);
- else if (eobs[3] == 1)
- {- vp8_dc_only_idct_add_v6 (q[48]*dq[0], dst+12, stride,dst+12,stride);
- ((int *)(q+48))[0] = 0;
- }
-
- q += 64;
- dst += 4*stride;
- eobs += 4;
- }
-}
-
-void vp8_dequant_idct_add_uv_block_v6(short *q, short *dq,
- unsigned char *dstu,
- unsigned char *dstv,
- int stride, char *eobs)
-{- int i;
-
- for (i = 0; i < 2; i++)
- {- if (eobs[0] > 1)
- vp8_dequant_idct_add_v6 (q, dq, dstu, stride);
- else if (eobs[0] == 1)
- {- vp8_dc_only_idct_add_v6 (q[0]*dq[0], dstu, stride, dstu, stride);
- ((int *)q)[0] = 0;
- }
-
- if (eobs[1] > 1)
- vp8_dequant_idct_add_v6 (q+16, dq, dstu+4, stride);
- else if (eobs[1] == 1)
- {- vp8_dc_only_idct_add_v6 (q[16]*dq[0], dstu+4, stride,
- dstu+4, stride);
- ((int *)(q+16))[0] = 0;
- }
-
- q += 32;
- dstu += 4*stride;
- eobs += 2;
- }
-
- for (i = 0; i < 2; i++)
- {- if (eobs[0] > 1)
- vp8_dequant_idct_add_v6 (q, dq, dstv, stride);
- else if (eobs[0] == 1)
- {- vp8_dc_only_idct_add_v6 (q[0]*dq[0], dstv, stride, dstv, stride);
- ((int *)q)[0] = 0;
- }
-
- if (eobs[1] > 1)
- vp8_dequant_idct_add_v6 (q+16, dq, dstv+4, stride);
- else if (eobs[1] == 1)
- {- vp8_dc_only_idct_add_v6 (q[16]*dq[0], dstv+4, stride,
- dstv+4, stride);
- ((int *)(q+16))[0] = 0;
- }
-
- q += 32;
- dstv += 4*stride;
- eobs += 2;
- }
-}
--- a/vp8/decoder/arm/dequantize_arm.c
+++ /dev/null
@@ -1,46 +1,0 @@
-/*
- * 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_config.h"
-#include "vp8/decoder/dequantize.h"
-#include "vp8/common/idct.h"
-#include "vpx_mem/vpx_mem.h"
-
-#if HAVE_ARMV7
-extern void vp8_dequantize_b_loop_neon(short *Q, short *DQC, short *DQ);
-#endif
-
-#if HAVE_ARMV6
-extern void vp8_dequantize_b_loop_v6(short *Q, short *DQC, short *DQ);
-#endif
-
-#if HAVE_ARMV7
-
-void vp8_dequantize_b_neon(BLOCKD *d)
-{- short *DQ = d->dqcoeff;
- short *Q = d->qcoeff;
- short *DQC = d->dequant;
-
- vp8_dequantize_b_loop_neon(Q, DQC, DQ);
-}
-#endif
-
-#if HAVE_ARMV6
-void vp8_dequantize_b_v6(BLOCKD *d)
-{- short *DQ = d->dqcoeff;
- short *Q = d->qcoeff;
- short *DQC = d->dequant;
-
- vp8_dequantize_b_loop_v6(Q, DQC, DQ);
-}
-#endif
--- a/vp8/decoder/arm/dequantize_arm.h
+++ /dev/null
@@ -1,59 +1,0 @@
-/*
- * 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.
- */
-
-
-#ifndef DEQUANTIZE_ARM_H
-#define DEQUANTIZE_ARM_H
-
-#if HAVE_ARMV6
-extern prototype_dequant_block(vp8_dequantize_b_v6);
-extern prototype_dequant_idct_add(vp8_dequant_idct_add_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
-
-#undef vp8_dequant_idct_add
-#define vp8_dequant_idct_add vp8_dequant_idct_add_v6
-
-#undef vp8_dequant_idct_add_y_block
-#define vp8_dequant_idct_add_y_block vp8_dequant_idct_add_y_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);
-extern prototype_dequant_idct_add(vp8_dequant_idct_add_neon);
-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
-
-#undef vp8_dequant_idct_add
-#define vp8_dequant_idct_add vp8_dequant_idct_add_neon
-
-#undef vp8_dequant_idct_add_y_block
-#define vp8_dequant_idct_add_y_block vp8_dequant_idct_add_y_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
--- a/vp8/decoder/arm/neon/dequant_idct_neon.asm
+++ /dev/null
@@ -1,131 +1,0 @@
-;
-; 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.
-;
-
-
- EXPORT |vp8_dequant_idct_add_neon|
- ARM
- REQUIRE8
- PRESERVE8
-
- AREA ||.text||, CODE, READONLY, ALIGN=2
-;void vp8_dequant_idct_add_neon(short *input, short *dq,
-; unsigned char *dest, int stride)
-; r0 short *input,
-; r1 short *dq,
-; r2 unsigned char *dest
-; r3 int stride
-
-|vp8_dequant_idct_add_neon| PROC
- vld1.16 {q3, q4}, [r0]- vld1.16 {q5, q6}, [r1]-
- add r1, r2, r3 ; r1 = dest + stride
- lsl r3, #1 ; 2x stride
-
- vld1.32 {d14[0]}, [r2], r3- vld1.32 {d14[1]}, [r1], r3- vld1.32 {d15[0]}, [r2]- vld1.32 {d15[1]}, [r1]-
- adr r12, cospi8sqrt2minus1 ; pointer to the first constant
-
- vmul.i16 q1, q3, q5 ;input for short_idct4x4llm_neon
- vmul.i16 q2, q4, q6
-
-;|short_idct4x4llm_neon| PROC
- vld1.16 {d0}, [r12]- vswp d3, d4 ;q2(vp[4] vp[12])
-
- vqdmulh.s16 q3, q2, d0[2]
- vqdmulh.s16 q4, q2, d0[0]
-
- vqadd.s16 d12, d2, d3 ;a1
- vqsub.s16 d13, d2, d3 ;b1
-
- vshr.s16 q3, q3, #1
- vshr.s16 q4, q4, #1
-
- vqadd.s16 q3, q3, q2
- vqadd.s16 q4, q4, q2
-
- vqsub.s16 d10, d6, d9 ;c1
- vqadd.s16 d11, d7, d8 ;d1
-
- vqadd.s16 d2, d12, d11
- vqadd.s16 d3, d13, d10
- vqsub.s16 d4, d13, d10
- vqsub.s16 d5, d12, d11
-
- vtrn.32 d2, d4
- vtrn.32 d3, d5
- vtrn.16 d2, d3
- vtrn.16 d4, d5
-
-; memset(input, 0, 32) -- 32bytes
- vmov.i16 q14, #0
-
- vswp d3, d4
- vqdmulh.s16 q3, q2, d0[2]
- vqdmulh.s16 q4, q2, d0[0]
-
- vqadd.s16 d12, d2, d3 ;a1
- vqsub.s16 d13, d2, d3 ;b1
-
- vmov q15, q14
-
- vshr.s16 q3, q3, #1
- vshr.s16 q4, q4, #1
-
- vqadd.s16 q3, q3, q2
- vqadd.s16 q4, q4, q2
-
- vqsub.s16 d10, d6, d9 ;c1
- vqadd.s16 d11, d7, d8 ;d1
-
- vqadd.s16 d2, d12, d11
- vqadd.s16 d3, d13, d10
- vqsub.s16 d4, d13, d10
- vqsub.s16 d5, d12, d11
-
- vst1.16 {q14, q15}, [r0]-
- vrshr.s16 d2, d2, #3
- vrshr.s16 d3, d3, #3
- vrshr.s16 d4, d4, #3
- vrshr.s16 d5, d5, #3
-
- vtrn.32 d2, d4
- vtrn.32 d3, d5
- vtrn.16 d2, d3
- vtrn.16 d4, d5
-
- vaddw.u8 q1, q1, d14
- vaddw.u8 q2, q2, d15
-
- sub r2, r2, r3
- sub r1, r1, r3
-
- vqmovun.s16 d0, q1
- vqmovun.s16 d1, q2
-
- vst1.32 {d0[0]}, [r2], r3- vst1.32 {d0[1]}, [r1], r3- vst1.32 {d1[0]}, [r2]- vst1.32 {d1[1]}, [r1]-
- bx lr
-
- ENDP ; |vp8_dequant_idct_add_neon|
-
-; Constant Pool
-cospi8sqrt2minus1 DCD 0x4e7b4e7b
-sinpi8sqrt2 DCD 0x8a8c8a8c
-
- END
--- a/vp8/decoder/arm/neon/dequantizeb_neon.asm
+++ /dev/null
@@ -1,34 +1,0 @@
-;
-; 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.
-;
-
-
- EXPORT |vp8_dequantize_b_loop_neon|
- ARM
- REQUIRE8
- PRESERVE8
-
- AREA ||.text||, CODE, READONLY, ALIGN=2
-; r0 short *Q,
-; r1 short *DQC
-; r2 short *DQ
-|vp8_dequantize_b_loop_neon| PROC
- vld1.16 {q0, q1}, [r0]- vld1.16 {q2, q3}, [r1]-
- vmul.i16 q4, q0, q2
- vmul.i16 q5, q1, q3
-
- vst1.16 {q4, q5}, [r2]-
- bx lr
-
- ENDP
-
- END
--- a/vp8/decoder/arm/neon/idct_blk_neon.c
+++ /dev/null
@@ -1,97 +1,0 @@
-/*
- * 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_config.h"
-#include "vp8/common/idct.h"
-#include "vp8/decoder/dequantize.h"
-
-/* place these declarations here because we don't want to maintain them
- * outside of this scope
- */
-void idct_dequant_full_2x_neon(short *q, short *dq,
- unsigned char *dst, int stride);
-void idct_dequant_0_2x_neon(short *q, short dq,
- unsigned char *dst, int stride);
-
-
-void vp8_dequant_idct_add_y_block_neon(short *q, short *dq,
- unsigned char *dst,
- int stride, char *eobs)
-{- int i;
-
- for (i = 0; i < 4; i++)
- {- if (((short *)(eobs))[0])
- {- if (((short *)eobs)[0] & 0xfefe)
- idct_dequant_full_2x_neon (q, dq, dst, stride);
- else
- idct_dequant_0_2x_neon (q, dq[0], dst, stride);
- }
-
- if (((short *)(eobs))[1])
- {- if (((short *)eobs)[1] & 0xfefe)
- idct_dequant_full_2x_neon (q+32, dq, dst+8, stride);
- else
- idct_dequant_0_2x_neon (q+32, dq[0], dst+8, stride);
- }
- q += 64;
- dst += 4*stride;
- eobs += 4;
- }
-}
-
-void vp8_dequant_idct_add_uv_block_neon(short *q, short *dq,
- unsigned char *dstu,
- unsigned char *dstv,
- int stride, char *eobs)
-{- if (((short *)(eobs))[0])
- {- if (((short *)eobs)[0] & 0xfefe)
- idct_dequant_full_2x_neon (q, dq, dstu, stride);
- else
- idct_dequant_0_2x_neon (q, dq[0], dstu, stride);
- }
-
- q += 32;
- dstu += 4*stride;
-
- if (((short *)(eobs))[1])
- {- if (((short *)eobs)[1] & 0xfefe)
- idct_dequant_full_2x_neon (q, dq, dstu, stride);
- else
- idct_dequant_0_2x_neon (q, dq[0], dstu, stride);
- }
-
- q += 32;
-
- if (((short *)(eobs))[2])
- {- if (((short *)eobs)[2] & 0xfefe)
- idct_dequant_full_2x_neon (q, dq, dstv, stride);
- else
- idct_dequant_0_2x_neon (q, dq[0], dstv, stride);
- }
-
- q += 32;
- dstv += 4*stride;
-
- if (((short *)(eobs))[3])
- {- if (((short *)eobs)[3] & 0xfefe)
- idct_dequant_full_2x_neon (q, dq, dstv, stride);
- else
- idct_dequant_0_2x_neon (q, dq[0], dstv, stride);
- }
-}
--- a/vp8/decoder/arm/neon/idct_dequant_0_2x_neon.asm
+++ /dev/null
@@ -1,79 +1,0 @@
-;
-; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
-;
-; Use of this source code is governed by a BSD-style license and patent
-; grant that can be found in the LICENSE file in the root of the source
-; tree. All contributing project authors may be found in the AUTHORS
-; file in the root of the source tree.
-;
-
-
- EXPORT |idct_dequant_0_2x_neon|
- ARM
- REQUIRE8
- PRESERVE8
-
- AREA ||.text||, CODE, READONLY, ALIGN=2
-;void idct_dequant_0_2x_neon(short *q, short dq,
-; unsigned char *dst, int stride);
-; r0 *q
-; r1 dq
-; r2 *dst
-; r3 stride
-|idct_dequant_0_2x_neon| PROC
- push {r4, r5}-
- add r12, r2, #4
- vld1.32 {d2[0]}, [r2], r3- vld1.32 {d8[0]}, [r12], r3- vld1.32 {d2[1]}, [r2], r3- vld1.32 {d8[1]}, [r12], r3- vld1.32 {d4[0]}, [r2], r3- vld1.32 {d10[0]}, [r12], r3- vld1.32 {d4[1]}, [r2], r3- vld1.32 {d10[1]}, [r12], r3-
- ldrh r12, [r0] ; lo q
- ldrh r4, [r0, #32] ; hi q
- mov r5, #0
- strh r5, [r0]
- strh r5, [r0, #32]
-
- sxth r12, r12 ; lo
- mul r0, r12, r1
- add r0, r0, #4
- asr r0, r0, #3
- vdup.16 q0, r0
- sxth r4, r4 ; hi
- mul r0, r4, r1
- add r0, r0, #4
- asr r0, r0, #3
- vdup.16 q3, r0
-
- vaddw.u8 q1, q0, d2 ; lo
- vaddw.u8 q2, q0, d4
- vaddw.u8 q4, q3, d8 ; hi
- vaddw.u8 q5, q3, d10
-
- sub r2, r2, r3, lsl #2 ; dst - 4*stride
- add r0, r2, #4
-
- vqmovun.s16 d2, q1 ; lo
- vqmovun.s16 d4, q2
- vqmovun.s16 d8, q4 ; hi
- vqmovun.s16 d10, q5
-
- vst1.32 {d2[0]}, [r2], r3 ; lo- vst1.32 {d8[0]}, [r0], r3 ; hi- vst1.32 {d2[1]}, [r2], r3- vst1.32 {d8[1]}, [r0], r3- vst1.32 {d4[0]}, [r2], r3- vst1.32 {d10[0]}, [r0], r3- vst1.32 {d4[1]}, [r2]- vst1.32 {d10[1]}, [r0]-
- pop {r4, r5}- bx lr
-
- ENDP ; |idct_dequant_0_2x_neon|
- END
--- a/vp8/decoder/arm/neon/idct_dequant_full_2x_neon.asm
+++ /dev/null
@@ -1,196 +1,0 @@
-;
-; 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.
-;
-
-
- EXPORT |idct_dequant_full_2x_neon|
- ARM
- REQUIRE8
- PRESERVE8
-
- AREA ||.text||, CODE, READONLY, ALIGN=2
-;void idct_dequant_full_2x_neon(short *q, short *dq,
-; unsigned char *dst, int stride);
-; r0 *q,
-; r1 *dq,
-; r2 *dst
-; r3 stride
-|idct_dequant_full_2x_neon| PROC
- vld1.16 {q0, q1}, [r1] ; dq (same l/r)- vld1.16 {q2, q3}, [r0] ; l q- add r0, r0, #32
- vld1.16 {q4, q5}, [r0] ; r q- add r12, r2, #4
-
- ; interleave the predictors
- vld1.32 {d28[0]}, [r2], r3 ; l pre- vld1.32 {d28[1]}, [r12], r3 ; r pre- vld1.32 {d29[0]}, [r2], r3- vld1.32 {d29[1]}, [r12], r3- vld1.32 {d30[0]}, [r2], r3- vld1.32 {d30[1]}, [r12], r3- vld1.32 {d31[0]}, [r2], r3- vld1.32 {d31[1]}, [r12]-
- adr r1, cospi8sqrt2minus1 ; pointer to the first constant
-
- ; dequant: q[i] = q[i] * dq[i]
- vmul.i16 q2, q2, q0
- vmul.i16 q3, q3, q1
- vmul.i16 q4, q4, q0
- vmul.i16 q5, q5, q1
-
- vld1.16 {d0}, [r1]-
- ; q2: l0r0 q3: l8r8
- ; q4: l4r4 q5: l12r12
- vswp d5, d8
- vswp d7, d10
-
- ; _CONSTANTS_ * 4,12 >> 16
- ; q6: 4 * sinpi : c1/temp1
- ; q7: 12 * sinpi : d1/temp2
- ; q8: 4 * cospi
- ; q9: 12 * cospi
- vqdmulh.s16 q6, q4, d0[2] ; sinpi8sqrt2
- vqdmulh.s16 q7, q5, d0[2]
- vqdmulh.s16 q8, q4, d0[0] ; cospi8sqrt2minus1
- vqdmulh.s16 q9, q5, d0[0]
-
- vqadd.s16 q10, q2, q3 ; a1 = 0 + 8
- vqsub.s16 q11, q2, q3 ; b1 = 0 - 8
-
- ; vqdmulh only accepts signed values. this was a problem because
- ; our constant had the high bit set, and was treated as a negative value.
- ; vqdmulh also doubles the value before it shifts by 16. we need to
- ; compensate for this. in the case of sinpi8sqrt2, the lowest bit is 0,
- ; so we can shift the constant without losing precision. this avoids
- ; shift again afterward, but also avoids the sign issue. win win!
- ; for cospi8sqrt2minus1 the lowest bit is 1, so we lose precision if we
- ; pre-shift it
- vshr.s16 q8, q8, #1
- vshr.s16 q9, q9, #1
-
- ; q4: 4 + 4 * cospi : d1/temp1
- ; q5: 12 + 12 * cospi : c1/temp2
- vqadd.s16 q4, q4, q8
- vqadd.s16 q5, q5, q9
-
- ; c1 = temp1 - temp2
- ; d1 = temp1 + temp2
- vqsub.s16 q2, q6, q5
- vqadd.s16 q3, q4, q7
-
- ; [0]: a1+d1
- ; [1]: b1+c1
- ; [2]: b1-c1
- ; [3]: a1-d1
- vqadd.s16 q4, q10, q3
- vqadd.s16 q5, q11, q2
- vqsub.s16 q6, q11, q2
- vqsub.s16 q7, q10, q3
-
- ; rotate
- vtrn.32 q4, q6
- vtrn.32 q5, q7
- vtrn.16 q4, q5
- vtrn.16 q6, q7
- ; idct loop 2
- ; q4: l 0, 4, 8,12 r 0, 4, 8,12
- ; q5: l 1, 5, 9,13 r 1, 5, 9,13
- ; q6: l 2, 6,10,14 r 2, 6,10,14
- ; q7: l 3, 7,11,15 r 3, 7,11,15
-
- ; q8: 1 * sinpi : c1/temp1
- ; q9: 3 * sinpi : d1/temp2
- ; q10: 1 * cospi
- ; q11: 3 * cospi
- vqdmulh.s16 q8, q5, d0[2] ; sinpi8sqrt2
- vqdmulh.s16 q9, q7, d0[2]
- vqdmulh.s16 q10, q5, d0[0] ; cospi8sqrt2minus1
- vqdmulh.s16 q11, q7, d0[0]
-
- vqadd.s16 q2, q4, q6 ; a1 = 0 + 2
- vqsub.s16 q3, q4, q6 ; b1 = 0 - 2
-
- ; see note on shifting above
- vshr.s16 q10, q10, #1
- vshr.s16 q11, q11, #1
-
- ; q10: 1 + 1 * cospi : d1/temp1
- ; q11: 3 + 3 * cospi : c1/temp2
- vqadd.s16 q10, q5, q10
- vqadd.s16 q11, q7, q11
-
- ; q8: c1 = temp1 - temp2
- ; q9: d1 = temp1 + temp2
- vqsub.s16 q8, q8, q11
- vqadd.s16 q9, q10, q9
-
- ; a1+d1
- ; b1+c1
- ; b1-c1
- ; a1-d1
- vqadd.s16 q4, q2, q9
- vqadd.s16 q5, q3, q8
- vqsub.s16 q6, q3, q8
- vqsub.s16 q7, q2, q9
-
- ; +4 >> 3 (rounding)
- vrshr.s16 q4, q4, #3 ; lo
- vrshr.s16 q5, q5, #3
- vrshr.s16 q6, q6, #3 ; hi
- vrshr.s16 q7, q7, #3
-
- vtrn.32 q4, q6
- vtrn.32 q5, q7
- vtrn.16 q4, q5
- vtrn.16 q6, q7
-
- ; adding pre
- ; input is still packed. pre was read interleaved
- vaddw.u8 q4, q4, d28
- vaddw.u8 q5, q5, d29
- vaddw.u8 q6, q6, d30
- vaddw.u8 q7, q7, d31
-
- vmov.i16 q14, #0
- vmov q15, q14
- vst1.16 {q14, q15}, [r0] ; write over high input- sub r0, r0, #32
- vst1.16 {q14, q15}, [r0] ; write over low input-
- sub r2, r2, r3, lsl #2 ; dst - 4*stride
- add r1, r2, #4 ; hi
-
- ;saturate and narrow
- vqmovun.s16 d0, q4 ; lo
- vqmovun.s16 d1, q5
- vqmovun.s16 d2, q6 ; hi
- vqmovun.s16 d3, q7
-
- vst1.32 {d0[0]}, [r2], r3 ; lo- vst1.32 {d0[1]}, [r1], r3 ; hi- vst1.32 {d1[0]}, [r2], r3- vst1.32 {d1[1]}, [r1], r3- vst1.32 {d2[0]}, [r2], r3- vst1.32 {d2[1]}, [r1], r3- vst1.32 {d3[0]}, [r2]- vst1.32 {d3[1]}, [r1]-
- bx lr
-
- ENDP ; |idct_dequant_full_2x_neon|
-
-; Constant Pool
-cospi8sqrt2minus1 DCD 0x4e7b
-; because the lowest bit in 0x8a8c is 0, we can pre-shift this
-sinpi8sqrt2 DCD 0x4546
-
- END
--- a/vp8/decoder/decodframe.c
+++ b/vp8/decoder/decodframe.c
@@ -15,7 +15,7 @@
#include "vp8/common/reconintra4x4.h"
#include "vp8/common/recon.h"
#include "vp8/common/reconinter.h"
-#include "dequantize.h"
+#include "vp8/common/dequantize.h"
#include "detokenize.h"
#include "vp8/common/invtrans.h"
#include "vp8/common/alloccommon.h"
@@ -32,7 +32,7 @@
#endif
#include "vpx_mem/vpx_mem.h"
#include "vp8/common/idct.h"
-#include "dequantize.h"
+
#include "vp8/common/threading.h"
#include "decoderthreading.h"
#include "dboolhuff.h"
@@ -218,7 +218,7 @@
{if (xd->eobs[i] > 1)
{- DEQUANT_INVOKE(&pbi->dequant, idct_add)
+ DEQUANT_INVOKE(&pbi->common.rtcd.dequant, idct_add)
(b->qcoeff, b->dequant,
*(b->base_dst) + b->dst, b->dst_stride);
}
@@ -247,7 +247,7 @@
/* do 2nd order transform on the dc block */
if (xd->eobs[24] > 1)
{- DEQUANT_INVOKE(&pbi->dequant, block)(b);
+ DEQUANT_INVOKE(&pbi->common.rtcd.dequant, block)(b);
IDCT_INVOKE(RTCD_VTABLE(idct), iwalsh16)(&b->dqcoeff[0],
xd->qcoeff);
@@ -272,7 +272,7 @@
DQC[0] = 1;
}
- DEQUANT_INVOKE (&pbi->dequant, idct_add_y_block)
+ DEQUANT_INVOKE (&pbi->common.rtcd.dequant, idct_add_y_block)
(xd->qcoeff, xd->block[0].dequant,
xd->dst.y_buffer,
xd->dst.y_stride, xd->eobs);
@@ -281,7 +281,7 @@
DQC[0] = dc_dequant_temp;
}
- DEQUANT_INVOKE (&pbi->dequant, idct_add_uv_block)
+ DEQUANT_INVOKE (&pbi->common.rtcd.dequant, idct_add_uv_block)
(xd->qcoeff+16*16, xd->block[16].dequant,
xd->dst.u_buffer, xd->dst.v_buffer,
xd->dst.uv_stride, xd->eobs+16);
--- a/vp8/decoder/dequantize.c
+++ /dev/null
@@ -1,44 +1,0 @@
-/*
- * 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_config.h"
-#include "dequantize.h"
-#include "vp8/common/idct.h"
-#include "vpx_mem/vpx_mem.h"
-
-void vp8_dequantize_b_c(BLOCKD *d)
-{- int i;
- short *DQ = d->dqcoeff;
- short *Q = d->qcoeff;
- short *DQC = d->dequant;
-
- for (i = 0; i < 16; i++)
- {- DQ[i] = Q[i] * DQC[i];
- }
-}
-
-void vp8_dequant_idct_add_c(short *input, short *dq,
- unsigned char *dest, int stride)
-{- int i;
-
- for (i = 0; i < 16; i++)
- {- input[i] = dq[i] * input[i];
- }
-
- vp8_short_idct4x4llm_c(input, dest, stride, dest, stride);
-
- vpx_memset(input, 0, 32);
-
-}
--- a/vp8/decoder/dequantize.h
+++ /dev/null
@@ -1,85 +1,0 @@
-/*
- * 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.
- */
-
-
-#ifndef DEQUANTIZE_H
-#define DEQUANTIZE_H
-#include "vp8/common/blockd.h"
-
-#define prototype_dequant_block(sym) \
- void sym(BLOCKD *x)
-
-#define prototype_dequant_idct_add(sym) \
- void sym(short *input, short *dq, \
- unsigned char *output, \
- int stride)
-
-#define prototype_dequant_idct_add_y_block(sym) \
- void sym(short *q, short *dq, \
- unsigned char *dst, \
- int stride, char *eobs)
-
-#define prototype_dequant_idct_add_uv_block(sym) \
- void sym(short *q, short *dq, \
- unsigned char *dst_u, \
- unsigned char *dst_v, int stride, char *eobs)
-
-#if ARCH_X86 || ARCH_X86_64
-#include "x86/dequantize_x86.h"
-#endif
-
-#if ARCH_ARM
-#include "arm/dequantize_arm.h"
-#endif
-
-#ifndef vp8_dequant_block
-#define vp8_dequant_block vp8_dequantize_b_c
-#endif
-extern prototype_dequant_block(vp8_dequant_block);
-
-#ifndef vp8_dequant_idct_add
-#define vp8_dequant_idct_add vp8_dequant_idct_add_c
-#endif
-extern prototype_dequant_idct_add(vp8_dequant_idct_add);
-
-#ifndef vp8_dequant_idct_add_y_block
-#define vp8_dequant_idct_add_y_block vp8_dequant_idct_add_y_block_c
-#endif
-extern prototype_dequant_idct_add_y_block(vp8_dequant_idct_add_y_block);
-
-#ifndef vp8_dequant_idct_add_uv_block
-#define vp8_dequant_idct_add_uv_block vp8_dequant_idct_add_uv_block_c
-#endif
-extern prototype_dequant_idct_add_uv_block(vp8_dequant_idct_add_uv_block);
-
-
-typedef prototype_dequant_block((*vp8_dequant_block_fn_t));
-
-typedef prototype_dequant_idct_add((*vp8_dequant_idct_add_fn_t));
-
-typedef prototype_dequant_idct_add_y_block((*vp8_dequant_idct_add_y_block_fn_t));
-
-typedef prototype_dequant_idct_add_uv_block((*vp8_dequant_idct_add_uv_block_fn_t));
-
-typedef struct
-{- vp8_dequant_block_fn_t block;
- vp8_dequant_idct_add_fn_t idct_add;
- vp8_dequant_idct_add_y_block_fn_t idct_add_y_block;
- vp8_dequant_idct_add_uv_block_fn_t idct_add_uv_block;
-} vp8_dequant_rtcd_vtable_t;
-
-#if CONFIG_RUNTIME_CPU_DETECT
-#define DEQUANT_INVOKE(ctx,fn) (ctx)->fn
-#else
-#define DEQUANT_INVOKE(ctx,fn) vp8_dequant_##fn
-#endif
-
-#endif
--- a/vp8/decoder/generic/dsystemdependent.c
+++ b/vp8/decoder/generic/dsystemdependent.c
@@ -10,7 +10,7 @@
#include "vpx_config.h"
-#include "vp8/decoder/dequantize.h"
+#include "vp8/common/dequantize.h"
#include "vp8/decoder/onyxd_int.h"
extern void vp8_arch_x86_decode_init(VP8D_COMP *pbi);
@@ -20,11 +20,7 @@
{/* Pure C: */
#if CONFIG_RUNTIME_CPU_DETECT
- pbi->mb.rtcd = &pbi->common.rtcd;
- pbi->dequant.block = vp8_dequantize_b_c;
- pbi->dequant.idct_add = vp8_dequant_idct_add_c;
- pbi->dequant.idct_add_y_block = vp8_dequant_idct_add_y_block_c;
- pbi->dequant.idct_add_uv_block = vp8_dequant_idct_add_uv_block_c;
+ pbi->mb.rtcd = &pbi->common.rtcd;
#endif
#if ARCH_X86 || ARCH_X86_64
--- a/vp8/decoder/idct_blk.c
+++ /dev/null
@@ -1,90 +1,0 @@
-/*
- * 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_config.h"
-#include "vp8/common/idct.h"
-#include "dequantize.h"
-
-void vp8_dequant_idct_add_c(short *input, short *dq,
- unsigned char *dest, int stride);
-void vp8_dc_only_idct_add_c(short input_dc, unsigned char * pred,
- int pred_stride, unsigned char *dst_ptr,
- int dst_stride);
-
-void vp8_dequant_idct_add_y_block_c
- (short *q, short *dq,
- unsigned char *dst, int stride, char *eobs)
-{- int i, j;
-
- for (i = 0; i < 4; i++)
- {- for (j = 0; j < 4; j++)
- {- if (*eobs++ > 1)
- vp8_dequant_idct_add_c (q, dq, dst, stride);
- else
- {- vp8_dc_only_idct_add_c (q[0]*dq[0], dst, stride, dst, stride);
- ((int *)q)[0] = 0;
- }
-
- q += 16;
- dst += 4;
- }
-
- dst += 4*stride - 16;
- }
-}
-
-void vp8_dequant_idct_add_uv_block_c
- (short *q, short *dq,
- unsigned char *dstu, unsigned char *dstv, int stride, char *eobs)
-{- int i, j;
-
- for (i = 0; i < 2; i++)
- {- for (j = 0; j < 2; j++)
- {- if (*eobs++ > 1)
- vp8_dequant_idct_add_c (q, dq, dstu, stride);
- else
- {- vp8_dc_only_idct_add_c (q[0]*dq[0], dstu, stride, dstu, stride);
- ((int *)q)[0] = 0;
- }
-
- q += 16;
- dstu += 4;
- }
-
- dstu += 4*stride - 8;
- }
-
- for (i = 0; i < 2; i++)
- {- for (j = 0; j < 2; j++)
- {- if (*eobs++ > 1)
- vp8_dequant_idct_add_c (q, dq, dstv, stride);
- else
- {- vp8_dc_only_idct_add_c (q[0]*dq[0], dstv, stride, dstv, stride);
- ((int *)q)[0] = 0;
- }
-
- q += 16;
- dstv += 4;
- }
-
- dstv += 4*stride - 8;
- }
-}
--- a/vp8/decoder/onyxd_int.h
+++ b/vp8/decoder/onyxd_int.h
@@ -16,7 +16,8 @@
#include "treereader.h"
#include "vp8/common/onyxc_int.h"
#include "vp8/common/threading.h"
-#include "dequantize.h"
+
+
#if CONFIG_ERROR_CONCEALMENT
#include "ec_types.h"
#endif
@@ -92,11 +93,6 @@
int ready_for_new_data;
DATARATE dr[16];
-
-#if CONFIG_RUNTIME_CPU_DETECT
- vp8_dequant_rtcd_vtable_t dequant;
-#endif
-
vp8_prob prob_intra;
vp8_prob prob_last;
--- a/vp8/decoder/threading.c
+++ b/vp8/decoder/threading.c
@@ -189,7 +189,7 @@
{if (xd->eobs[i] > 1)
{- DEQUANT_INVOKE(&pbi->dequant, idct_add)
+ DEQUANT_INVOKE(&pbi->common.rtcd.dequant, idct_add)
(b->qcoeff, b->dequant,
*(b->base_dst) + b->dst, b->dst_stride);
}
@@ -217,7 +217,7 @@
/* do 2nd order transform on the dc block */
if (xd->eobs[24] > 1)
{- DEQUANT_INVOKE(&pbi->dequant, block)(b);
+ DEQUANT_INVOKE(&pbi->common.rtcd.dequant, block)(b);
IDCT_INVOKE(RTCD_VTABLE(idct), iwalsh16)(&b->dqcoeff[0],
xd->qcoeff);
@@ -248,13 +248,13 @@
DQC = local_dequant;
}
- DEQUANT_INVOKE (&pbi->dequant, idct_add_y_block)
+ DEQUANT_INVOKE (&pbi->common.rtcd.dequant, idct_add_y_block)
(xd->qcoeff, DQC,
xd->dst.y_buffer,
xd->dst.y_stride, xd->eobs);
}
- DEQUANT_INVOKE (&pbi->dequant, idct_add_uv_block)
+ DEQUANT_INVOKE (&pbi->common.rtcd.dequant, idct_add_uv_block)
(xd->qcoeff+16*16, xd->block[16].dequant,
xd->dst.u_buffer, xd->dst.v_buffer,
xd->dst.uv_stride, xd->eobs+16);
--- a/vp8/decoder/x86/dequantize_mmx.asm
+++ /dev/null
@@ -1,258 +1,0 @@
-;
-; 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/x86_abi_support.asm"
-
-
-;void vp8_dequantize_b_impl_mmx(short *sq, short *dq, short *q)
-global sym(vp8_dequantize_b_impl_mmx)
-sym(vp8_dequantize_b_impl_mmx):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 3
- push rsi
- push rdi
- ; end prolog
-
- mov rsi, arg(0) ;sq
- mov rdi, arg(1) ;dq
- mov rax, arg(2) ;q
-
- movq mm1, [rsi]
- pmullw mm1, [rax+0] ; mm4 *= kernel 0 modifiers.
- movq [rdi], mm1
-
- movq mm1, [rsi+8]
- pmullw mm1, [rax+8] ; mm4 *= kernel 0 modifiers.
- movq [rdi+8], mm1
-
- movq mm1, [rsi+16]
- pmullw mm1, [rax+16] ; mm4 *= kernel 0 modifiers.
- movq [rdi+16], mm1
-
- movq mm1, [rsi+24]
- pmullw mm1, [rax+24] ; mm4 *= kernel 0 modifiers.
- movq [rdi+24], mm1
-
- ; begin epilog
- pop rdi
- pop rsi
- UNSHADOW_ARGS
- pop rbp
- ret
-
-
-;void dequant_idct_add_mmx(
-;short *input, 0
-;short *dq, 1
-;unsigned char *dest, 2
-;int stride) 3
-global sym(vp8_dequant_idct_add_mmx)
-sym(vp8_dequant_idct_add_mmx):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 4
- GET_GOT rbx
- push rdi
- ; end prolog
-
- mov rax, arg(0) ;input
- mov rdx, arg(1) ;dq
-
-
- movq mm0, [rax ]
- pmullw mm0, [rdx]
-
- movq mm1, [rax +8]
- pmullw mm1, [rdx +8]
-
- movq mm2, [rax+16]
- pmullw mm2, [rdx+16]
-
- movq mm3, [rax+24]
- pmullw mm3, [rdx+24]
-
- mov rdx, arg(2) ;dest
-
- pxor mm7, mm7
-
-
- movq [rax], mm7
- movq [rax+8], mm7
-
- movq [rax+16],mm7
- movq [rax+24],mm7
-
-
- movsxd rdi, dword ptr arg(3) ;stride
-
- psubw mm0, mm2 ; b1= 0-2
- paddw mm2, mm2 ;
-
- movq mm5, mm1
- paddw mm2, mm0 ; a1 =0+2
-
- pmulhw mm5, [GLOBAL(x_s1sqr2)];
- paddw mm5, mm1 ; ip1 * sin(pi/8) * sqrt(2)
-
- movq mm7, mm3 ;
- pmulhw mm7, [GLOBAL(x_c1sqr2less1)];
-
- paddw mm7, mm3 ; ip3 * cos(pi/8) * sqrt(2)
- psubw mm7, mm5 ; c1
-
- movq mm5, mm1
- movq mm4, mm3
-
- pmulhw mm5, [GLOBAL(x_c1sqr2less1)]
- paddw mm5, mm1
-
- pmulhw mm3, [GLOBAL(x_s1sqr2)]
- paddw mm3, mm4
-
- paddw mm3, mm5 ; d1
- movq mm6, mm2 ; a1
-
- movq mm4, mm0 ; b1
- paddw mm2, mm3 ;0
-
- paddw mm4, mm7 ;1
- psubw mm0, mm7 ;2
-
- psubw mm6, mm3 ;3
-
- movq mm1, mm2 ; 03 02 01 00
- movq mm3, mm4 ; 23 22 21 20
-
- punpcklwd mm1, mm0 ; 11 01 10 00
- punpckhwd mm2, mm0 ; 13 03 12 02
-
- punpcklwd mm3, mm6 ; 31 21 30 20
- punpckhwd mm4, mm6 ; 33 23 32 22
-
- movq mm0, mm1 ; 11 01 10 00
- movq mm5, mm2 ; 13 03 12 02
-
- punpckldq mm0, mm3 ; 30 20 10 00
- punpckhdq mm1, mm3 ; 31 21 11 01
-
- punpckldq mm2, mm4 ; 32 22 12 02
- punpckhdq mm5, mm4 ; 33 23 13 03
-
- movq mm3, mm5 ; 33 23 13 03
-
- psubw mm0, mm2 ; b1= 0-2
- paddw mm2, mm2 ;
-
- movq mm5, mm1
- paddw mm2, mm0 ; a1 =0+2
-
- pmulhw mm5, [GLOBAL(x_s1sqr2)];
- paddw mm5, mm1 ; ip1 * sin(pi/8) * sqrt(2)
-
- movq mm7, mm3 ;
- pmulhw mm7, [GLOBAL(x_c1sqr2less1)];
-
- paddw mm7, mm3 ; ip3 * cos(pi/8) * sqrt(2)
- psubw mm7, mm5 ; c1
-
- movq mm5, mm1
- movq mm4, mm3
-
- pmulhw mm5, [GLOBAL(x_c1sqr2less1)]
- paddw mm5, mm1
-
- pmulhw mm3, [GLOBAL(x_s1sqr2)]
- paddw mm3, mm4
-
- paddw mm3, mm5 ; d1
- paddw mm0, [GLOBAL(fours)]
-
- paddw mm2, [GLOBAL(fours)]
- movq mm6, mm2 ; a1
-
- movq mm4, mm0 ; b1
- paddw mm2, mm3 ;0
-
- paddw mm4, mm7 ;1
- psubw mm0, mm7 ;2
-
- psubw mm6, mm3 ;3
- psraw mm2, 3
-
- psraw mm0, 3
- psraw mm4, 3
-
- psraw mm6, 3
-
- movq mm1, mm2 ; 03 02 01 00
- movq mm3, mm4 ; 23 22 21 20
-
- punpcklwd mm1, mm0 ; 11 01 10 00
- punpckhwd mm2, mm0 ; 13 03 12 02
-
- punpcklwd mm3, mm6 ; 31 21 30 20
- punpckhwd mm4, mm6 ; 33 23 32 22
-
- movq mm0, mm1 ; 11 01 10 00
- movq mm5, mm2 ; 13 03 12 02
-
- punpckldq mm0, mm3 ; 30 20 10 00
- punpckhdq mm1, mm3 ; 31 21 11 01
-
- punpckldq mm2, mm4 ; 32 22 12 02
- punpckhdq mm5, mm4 ; 33 23 13 03
-
- pxor mm7, mm7
-
- movd mm4, [rdx]
- punpcklbw mm4, mm7
- paddsw mm0, mm4
- packuswb mm0, mm7
- movd [rdx], mm0
-
- movd mm4, [rdx+rdi]
- punpcklbw mm4, mm7
- paddsw mm1, mm4
- packuswb mm1, mm7
- movd [rdx+rdi], mm1
-
- movd mm4, [rdx+2*rdi]
- punpcklbw mm4, mm7
- paddsw mm2, mm4
- packuswb mm2, mm7
- movd [rdx+rdi*2], mm2
-
- add rdx, rdi
-
- movd mm4, [rdx+2*rdi]
- punpcklbw mm4, mm7
- paddsw mm5, mm4
- packuswb mm5, mm7
- movd [rdx+rdi*2], mm5
-
- ; begin epilog
- pop rdi
- RESTORE_GOT
- UNSHADOW_ARGS
- pop rbp
- ret
-
-SECTION_RODATA
-align 16
-x_s1sqr2:
- times 4 dw 0x8A8C
-align 16
-x_c1sqr2less1:
- times 4 dw 0x4E7B
-align 16
-fours:
- times 4 dw 0x0004
--- a/vp8/decoder/x86/dequantize_x86.h
+++ /dev/null
@@ -1,58 +1,0 @@
-/*
- * 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.
- */
-
-
-#ifndef DEQUANTIZE_X86_H
-#define DEQUANTIZE_X86_H
-
-
-/* Note:
- *
- * This platform is commonly built for runtime CPU detection. If you modify
- * any of the function mappings present in this file, be sure to also update
- * them in the function pointer initialization code
- */
-#if HAVE_MMX
-extern prototype_dequant_block(vp8_dequantize_b_mmx);
-extern prototype_dequant_idct_add(vp8_dequant_idct_add_mmx);
-extern prototype_dequant_idct_add_y_block(vp8_dequant_idct_add_y_block_mmx);
-extern prototype_dequant_idct_add_uv_block(vp8_dequant_idct_add_uv_block_mmx);
-
-#if !CONFIG_RUNTIME_CPU_DETECT
-#undef vp8_dequant_block
-#define vp8_dequant_block vp8_dequantize_b_mmx
-
-#undef vp8_dequant_idct_add
-#define vp8_dequant_idct_add vp8_dequant_idct_add_mmx
-
-#undef vp8_dequant_idct_add_y_block
-#define vp8_dequant_idct_add_y_block vp8_dequant_idct_add_y_block_mmx
-
-#undef vp8_dequant_idct_add_uv_block
-#define vp8_dequant_idct_add_uv_block vp8_dequant_idct_add_uv_block_mmx
-
-#endif
-#endif
-
-#if HAVE_SSE2
-extern prototype_dequant_idct_add_y_block(vp8_dequant_idct_add_y_block_sse2);
-extern prototype_dequant_idct_add_uv_block(vp8_dequant_idct_add_uv_block_sse2);
-
-#if !CONFIG_RUNTIME_CPU_DETECT
-#undef vp8_dequant_idct_add_y_block
-#define vp8_dequant_idct_add_y_block vp8_dequant_idct_add_y_block_sse2
-
-#undef vp8_dequant_idct_add_uv_block
-#define vp8_dequant_idct_add_uv_block vp8_dequant_idct_add_uv_block_sse2
-
-#endif
-#endif
-
-#endif
--- a/vp8/decoder/x86/idct_blk_mmx.c
+++ /dev/null
@@ -1,117 +1,0 @@
-/*
- * 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_config.h"
-#include "vp8/common/idct.h"
-#include "vp8/decoder/dequantize.h"
-
-void vp8_dequant_idct_add_y_block_mmx
- (short *q, short *dq,
- unsigned char *dst, int stride, char *eobs)
-{- int i;
-
- for (i = 0; i < 4; i++)
- {- if (eobs[0] > 1)
- vp8_dequant_idct_add_mmx (q, dq, dst, stride);
- else if (eobs[0] == 1)
- {- vp8_dc_only_idct_add_mmx (q[0]*dq[0], dst, stride, dst, stride);
- ((int *)q)[0] = 0;
- }
-
- if (eobs[1] > 1)
- vp8_dequant_idct_add_mmx (q+16, dq, dst+4, stride);
- else if (eobs[1] == 1)
- {- vp8_dc_only_idct_add_mmx (q[16]*dq[0], dst+4, stride,
- dst+4, stride);
- ((int *)(q+16))[0] = 0;
- }
-
- if (eobs[2] > 1)
- vp8_dequant_idct_add_mmx (q+32, dq, dst+8, stride);
- else if (eobs[2] == 1)
- {- vp8_dc_only_idct_add_mmx (q[32]*dq[0], dst+8, stride,
- dst+8, stride);
- ((int *)(q+32))[0] = 0;
- }
-
- if (eobs[3] > 1)
- vp8_dequant_idct_add_mmx (q+48, dq, dst+12, stride);
- else if (eobs[3] == 1)
- {- vp8_dc_only_idct_add_mmx (q[48]*dq[0], dst+12, stride,
- dst+12, stride);
- ((int *)(q+48))[0] = 0;
- }
-
- q += 64;
- dst += 4*stride;
- eobs += 4;
- }
-}
-
-void vp8_dequant_idct_add_uv_block_mmx
- (short *q, short *dq,
- unsigned char *dstu, unsigned char *dstv, int stride, char *eobs)
-{- int i;
-
- for (i = 0; i < 2; i++)
- {- if (eobs[0] > 1)
- vp8_dequant_idct_add_mmx (q, dq, dstu, stride);
- else if (eobs[0] == 1)
- {- vp8_dc_only_idct_add_mmx (q[0]*dq[0], dstu, stride, dstu, stride);
- ((int *)q)[0] = 0;
- }
-
- if (eobs[1] > 1)
- vp8_dequant_idct_add_mmx (q+16, dq, dstu+4, stride);
- else if (eobs[1] == 1)
- {- vp8_dc_only_idct_add_mmx (q[16]*dq[0], dstu+4, stride,
- dstu+4, stride);
- ((int *)(q+16))[0] = 0;
- }
-
- q += 32;
- dstu += 4*stride;
- eobs += 2;
- }
-
- for (i = 0; i < 2; i++)
- {- if (eobs[0] > 1)
- vp8_dequant_idct_add_mmx (q, dq, dstv, stride);
- else if (eobs[0] == 1)
- {- vp8_dc_only_idct_add_mmx (q[0]*dq[0], dstv, stride, dstv, stride);
- ((int *)q)[0] = 0;
- }
-
- if (eobs[1] > 1)
- vp8_dequant_idct_add_mmx (q+16, dq, dstv+4, stride);
- else if (eobs[1] == 1)
- {- vp8_dc_only_idct_add_mmx (q[16]*dq[0], dstv+4, stride,
- dstv+4, stride);
- ((int *)(q+16))[0] = 0;
- }
-
- q += 32;
- dstv += 4*stride;
- eobs += 2;
- }
-}
--- a/vp8/decoder/x86/idct_blk_sse2.c
+++ /dev/null
@@ -1,90 +1,0 @@
-/*
- * 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_config.h"
-#include "vp8/common/idct.h"
-#include "vp8/decoder/dequantize.h"
-
-void vp8_idct_dequant_0_2x_sse2
- (short *q, short *dq ,
- unsigned char *dst, int dst_stride);
-void vp8_idct_dequant_full_2x_sse2
- (short *q, short *dq ,
- unsigned char *dst, int dst_stride);
-
-void vp8_dequant_idct_add_y_block_sse2
- (short *q, short *dq,
- unsigned char *dst, int stride, char *eobs)
-{- int i;
-
- for (i = 0; i < 4; i++)
- {- if (((short *)(eobs))[0])
- {- if (((short *)(eobs))[0] & 0xfefe)
- vp8_idct_dequant_full_2x_sse2 (q, dq, dst, stride);
- else
- vp8_idct_dequant_0_2x_sse2 (q, dq, dst, stride);
- }
- if (((short *)(eobs))[1])
- {- if (((short *)(eobs))[1] & 0xfefe)
- vp8_idct_dequant_full_2x_sse2 (q+32, dq, dst+8, stride);
- else
- vp8_idct_dequant_0_2x_sse2 (q+32, dq, dst+8, stride);
- }
- q += 64;
- dst += stride*4;
- eobs += 4;
- }
-}
-
-void vp8_dequant_idct_add_uv_block_sse2
- (short *q, short *dq,
- unsigned char *dstu, unsigned char *dstv, int stride, char *eobs)
-{- if (((short *)(eobs))[0])
- {- if (((short *)(eobs))[0] & 0xfefe)
- vp8_idct_dequant_full_2x_sse2 (q, dq, dstu, stride);
- else
- vp8_idct_dequant_0_2x_sse2 (q, dq, dstu, stride);
- }
- q += 32;
- dstu += stride*4;
-
- if (((short *)(eobs))[1])
- {- if (((short *)(eobs))[1] & 0xfefe)
- vp8_idct_dequant_full_2x_sse2 (q, dq, dstu, stride);
- else
- vp8_idct_dequant_0_2x_sse2 (q, dq, dstu, stride);
- }
- q += 32;
-
- if (((short *)(eobs))[2])
- {- if (((short *)(eobs))[2] & 0xfefe)
- vp8_idct_dequant_full_2x_sse2 (q, dq, dstv, stride);
- else
- vp8_idct_dequant_0_2x_sse2 (q, dq, dstv, stride);
- }
- q += 32;
- dstv += stride*4;
-
- if (((short *)(eobs))[3])
- {- if (((short *)(eobs))[3] & 0xfefe)
- vp8_idct_dequant_full_2x_sse2 (q, dq, dstv, stride);
- else
- vp8_idct_dequant_0_2x_sse2 (q, dq, dstv, stride);
- }
-}
--- a/vp8/decoder/x86/x86_dsystemdependent.c
+++ b/vp8/decoder/x86/x86_dsystemdependent.c
@@ -13,47 +13,7 @@
#include "vpx_ports/x86.h"
#include "vp8/decoder/onyxd_int.h"
-
-#if HAVE_MMX
-void vp8_dequantize_b_impl_mmx(short *sq, short *dq, short *q);
-
-void vp8_dequantize_b_mmx(BLOCKD *d)
-{- short *sq = (short *) d->qcoeff;
- short *dq = (short *) d->dqcoeff;
- short *q = (short *) d->dequant;
- vp8_dequantize_b_impl_mmx(sq, dq, q);
-}
-#endif
-
void vp8_arch_x86_decode_init(VP8D_COMP *pbi)
{-#if CONFIG_RUNTIME_CPU_DETECT
- int flags = x86_simd_caps();
- /* Note:
- *
- * This platform can be built without runtime CPU detection as well. If
- * you modify any of the function mappings present in this file, be sure
- * to also update them in static mapings (<arch>/filename_<arch>.h)
- */
- /* Override default functions with fastest ones for this CPU. */
-#if HAVE_MMX
- if (flags & HAS_MMX)
- {- pbi->dequant.block = vp8_dequantize_b_mmx;
- pbi->dequant.idct_add = vp8_dequant_idct_add_mmx;
- pbi->dequant.idct_add_y_block = vp8_dequant_idct_add_y_block_mmx;
- pbi->dequant.idct_add_uv_block = vp8_dequant_idct_add_uv_block_mmx;
- }
-#endif
-#if HAVE_SSE2
- if (flags & HAS_SSE2)
- {- pbi->dequant.idct_add_y_block = vp8_dequant_idct_add_y_block_sse2;
- pbi->dequant.idct_add_uv_block = vp8_dequant_idct_add_uv_block_sse2;
- }
-#endif
-
-#endif
}
--- a/vp8/encoder/encodeframe.c
+++ b/vp8/encoder/encodeframe.c
@@ -1091,8 +1091,10 @@
#endif
}
-int vp8cx_encode_intra_macro_block(VP8_COMP *cpi, MACROBLOCK *x, TOKENEXTRA **t, int mb_row, int mb_col)
+int vp8cx_encode_intra_macro_block(VP8_COMP *cpi, MACROBLOCK *x, TOKENEXTRA **t,
+ int mb_row, int mb_col)
{+ MACROBLOCKD *xd = &x->e_mbd;
int rate;
if (cpi->sf.RD && cpi->compressor_speed != 2)
@@ -1112,14 +1114,17 @@
vp8_encode_intra16x16mby(IF_RTCD(&cpi->rtcd), x);
vp8_encode_intra16x16mbuv(IF_RTCD(&cpi->rtcd), x);
+
sum_intra_stats(cpi, x);
vp8_tokenize_mb(cpi, &x->e_mbd, t);
- if (x->e_mbd.mode_info_context->mbmi.mode != B_PRED)
- vp8_inverse_transform_mby(IF_RTCD(&cpi->rtcd.common->idct), &x->e_mbd);
+ if (xd->mode_info_context->mbmi.mode != B_PRED)
+ vp8_inverse_transform_mby(xd, IF_RTCD(&cpi->common.rtcd));
- vp8_inverse_transform_mbuv(IF_RTCD(&cpi->rtcd.common->idct), &x->e_mbd);
-
+ DEQUANT_INVOKE (&cpi->common.rtcd.dequant, idct_add_uv_block)
+ (xd->qcoeff+16*16, xd->block[16].dequant,
+ xd->dst.u_buffer, xd->dst.v_buffer,
+ xd->dst.uv_stride, xd->eobs+16);
return rate;
}
#ifdef SPEEDSTATS
@@ -1312,12 +1317,14 @@
if (!x->skip)
{vp8_tokenize_mb(cpi, xd, t);
- if (x->e_mbd.mode_info_context->mbmi.mode != B_PRED)
- {- vp8_inverse_transform_mby(IF_RTCD(&cpi->rtcd.common->idct),
- &x->e_mbd);
- }
- vp8_inverse_transform_mbuv(IF_RTCD(&cpi->rtcd.common->idct), &x->e_mbd);
+
+ if (xd->mode_info_context->mbmi.mode != B_PRED)
+ vp8_inverse_transform_mby(xd, IF_RTCD(&cpi->common.rtcd));
+
+ DEQUANT_INVOKE (&cpi->common.rtcd.dequant, idct_add_uv_block)
+ (xd->qcoeff+16*16, xd->block[16].dequant,
+ xd->dst.u_buffer, xd->dst.v_buffer,
+ xd->dst.uv_stride, xd->eobs+16);
}
else
{--- a/vp8/encoder/encodeintra.c
+++ b/vp8/encoder/encodeintra.c
@@ -45,7 +45,7 @@
vp8_encode_intra16x16mby(rtcd, x);
- vp8_inverse_transform_mby(IF_RTCD(&rtcd->common->idct), &x->e_mbd);
+ vp8_inverse_transform_mby(&x->e_mbd, IF_RTCD(&cpi->common.rtcd));
}
else
{@@ -77,8 +77,17 @@
x->quantize_b(be, b);
- vp8_inverse_transform_b(IF_RTCD(&rtcd->common->idct), b, 16);
-
+ if (*b->eob > 1)
+ {+ IDCT_INVOKE(IF_RTCD(&rtcd->common->idct), idct16)(b->dqcoeff,
+ b->predictor, 16, *(b->base_dst) + b->dst, b->dst_stride);
+ }
+ else
+ {+ IDCT_INVOKE(IF_RTCD(&rtcd->common->idct), idct1_scalar_add)
+ (b->dqcoeff[0], b->predictor, 16, *(b->base_dst) + b->dst,
+ b->dst_stride);
+ }
}
void vp8_encode_intra4x4mby(const VP8_ENCODER_RTCD *rtcd, MACROBLOCK *mb)
@@ -96,11 +105,12 @@
void vp8_encode_intra16x16mby(const VP8_ENCODER_RTCD *rtcd, MACROBLOCK *x)
{BLOCK *b = &x->block[0];
+ MACROBLOCKD *xd = &x->e_mbd;
- RECON_INVOKE(&rtcd->common->recon, build_intra_predictors_mby)(&x->e_mbd);
+ RECON_INVOKE(&rtcd->common->recon, build_intra_predictors_mby_s)(&x->e_mbd);
- ENCODEMB_INVOKE(&rtcd->encodemb, submby)(x->src_diff, *(b->base_src),
- b->src_stride, x->e_mbd.predictor, 16);
+ ENCODEMB_INVOKE(&rtcd->encodemb, submby) (x->src_diff, *(b->base_src),
+ b->src_stride, xd->dst.y_buffer, xd->dst.y_stride);
vp8_transform_intra_mby(x);
@@ -108,16 +118,17 @@
if (x->optimize)
vp8_optimize_mby(x, rtcd);
-
}
void vp8_encode_intra16x16mbuv(const VP8_ENCODER_RTCD *rtcd, MACROBLOCK *x)
{- RECON_INVOKE(&rtcd->common->recon, build_intra_predictors_mbuv)(&x->e_mbd);
+ MACROBLOCKD *xd = &x->e_mbd;
+ RECON_INVOKE(&rtcd->common->recon, build_intra_predictors_mbuv_s)(&x->e_mbd);
+
ENCODEMB_INVOKE(&rtcd->encodemb, submbuv)(x->src_diff, x->src.u_buffer,
- x->src.v_buffer, x->src.uv_stride, &x->e_mbd.predictor[256],
- &x->e_mbd.predictor[320], 8);
+ x->src.v_buffer, x->src.uv_stride, xd->dst.u_buffer,
+ xd->dst.v_buffer, xd->dst.uv_stride);
vp8_transform_mbuv(x);
@@ -125,5 +136,4 @@
if (x->optimize)
vp8_optimize_mbuv(x, rtcd);
-
}
--- a/vp8/encoder/encodemb.c
+++ b/vp8/encoder/encodemb.c
@@ -105,10 +105,10 @@
BLOCK *b = &x->block[0];
ENCODEMB_INVOKE(&rtcd->encodemb, submby)(x->src_diff, *(b->base_src),
- b->src_stride, x->e_mbd.predictor, 16);
+ b->src_stride, x->e_mbd.dst.y_buffer, x->e_mbd.dst.y_stride);
ENCODEMB_INVOKE(&rtcd->encodemb, submbuv)(x->src_diff, x->src.u_buffer,
- x->src.v_buffer, x->src.uv_stride, &x->e_mbd.predictor[256],
- &x->e_mbd.predictor[320], 8);
+ x->src.v_buffer, x->src.uv_stride, x->e_mbd.dst.u_buffer,
+ x->e_mbd.dst.v_buffer, x->e_mbd.dst.uv_stride);
}
static void build_dcblock(MACROBLOCK *x)
@@ -625,7 +625,7 @@
void vp8_encode_inter16x16(const VP8_ENCODER_RTCD *rtcd, MACROBLOCK *x)
{- vp8_build_inter_predictors_mb_e(&x->e_mbd);
+ vp8_build_inter_predictors_mb(&x->e_mbd);
vp8_subtract_mb(rtcd, x);
@@ -635,7 +635,6 @@
if (x->optimize)
optimize_mb(x, rtcd);
-
}
/* this funciton is used by first pass only */
@@ -643,15 +642,15 @@
{BLOCK *b = &x->block[0];
- vp8_build_inter16x16_predictors_mby(&x->e_mbd);
+ vp8_build_inter16x16_predictors_mby(&x->e_mbd, x->e_mbd.dst.y_buffer,
+ x->e_mbd.dst.y_stride);
ENCODEMB_INVOKE(&rtcd->encodemb, submby)(x->src_diff, *(b->base_src),
- b->src_stride, x->e_mbd.predictor, 16);
+ b->src_stride, x->e_mbd.dst.y_buffer, x->e_mbd.dst.y_stride);
transform_mby(x);
vp8_quantize_mby(x);
- vp8_inverse_transform_mby(IF_RTCD(&rtcd->common->idct), &x->e_mbd);
-
+ vp8_inverse_transform_mby(&x->e_mbd, IF_RTCD(rtcd->common));
}
--- a/vp8/encoder/encodemb.h
+++ b/vp8/encoder/encodemb.h
@@ -12,6 +12,7 @@
#ifndef __INC_ENCODEMB_H
#define __INC_ENCODEMB_H
+
#include "vpx_config.h"
#include "block.h"
--- a/vp8/encoder/rdopt.c
+++ b/vp8/encoder/rdopt.c
@@ -2166,7 +2166,7 @@
continue;
vp8_set_mbmode_and_mvs(x, this_mode, &mode_mv[this_mode]);
- vp8_build_inter16x16_predictors_mby(&x->e_mbd);
+ vp8_build_inter16x16_predictors_mby(&x->e_mbd, x->e_mbd.predictor, 16);
if (cpi->active_map_enabled && x->active_ptr[0] == 0) {x->skip = 1;
--- a/vp8/vp8_common.mk
+++ b/vp8/vp8_common.mk
@@ -20,6 +20,8 @@
VP8_COMMON_SRCS-yes += common/coefupdateprobs.h
VP8_COMMON_SRCS-yes += common/debugmodes.c
VP8_COMMON_SRCS-yes += common/default_coef_probs.h
+VP8_COMMON_SRCS-yes += common/dequantize.c
+VP8_COMMON_SRCS-yes += common/dequantize.h
VP8_COMMON_SRCS-yes += common/entropy.c
VP8_COMMON_SRCS-yes += common/entropymode.c
VP8_COMMON_SRCS-yes += common/entropymv.c
@@ -28,6 +30,7 @@
VP8_COMMON_SRCS-yes += common/filter.h
VP8_COMMON_SRCS-yes += common/findnearmv.c
VP8_COMMON_SRCS-yes += common/generic/systemdependent.c
+VP8_COMMON_SRCS-yes += common/idct_blk.c
VP8_COMMON_SRCS-yes += common/idctllm.c
VP8_COMMON_SRCS-yes += common/alloccommon.h
VP8_COMMON_SRCS-yes += common/blockd.h
@@ -57,7 +60,6 @@
VP8_COMMON_SRCS-yes += common/systemdependent.h
VP8_COMMON_SRCS-yes += common/threading.h
VP8_COMMON_SRCS-yes += common/treecoder.h
-VP8_COMMON_SRCS-yes += common/invtrans.c
VP8_COMMON_SRCS-yes += common/loopfilter.c
VP8_COMMON_SRCS-yes += common/loopfilter_filters.c
VP8_COMMON_SRCS-yes += common/mbpitch.c
@@ -69,9 +71,13 @@
VP8_COMMON_SRCS-yes += common/reconintra4x4.c
VP8_COMMON_SRCS-yes += common/setupintrarecon.c
VP8_COMMON_SRCS-yes += common/swapyv12buffer.c
+
+
+
VP8_COMMON_SRCS-$(CONFIG_POSTPROC_VISUALIZER) += common/textblit.c
VP8_COMMON_SRCS-yes += common/treecoder.c
+VP8_COMMON_SRCS-$(ARCH_X86)$(ARCH_X86_64) += common/x86/dequantize_x86.h
VP8_COMMON_SRCS-$(ARCH_X86)$(ARCH_X86_64) += common/x86/filter_x86.c
VP8_COMMON_SRCS-$(ARCH_X86)$(ARCH_X86_64) += common/x86/filter_x86.h
VP8_COMMON_SRCS-$(ARCH_X86)$(ARCH_X86_64) += common/x86/idct_x86.h
@@ -84,11 +90,14 @@
VP8_COMMON_SRCS-$(ARCH_X86)$(ARCH_X86_64) += common/x86/loopfilter_x86.c
VP8_COMMON_SRCS-$(CONFIG_POSTPROC) += common/postproc.h
VP8_COMMON_SRCS-$(CONFIG_POSTPROC) += common/postproc.c
+VP8_COMMON_SRCS-$(HAVE_MMX) += common/x86/dequantize_mmx.asm
+VP8_COMMON_SRCS-$(HAVE_MMX) += common/x86/idct_blk_mmx.c
VP8_COMMON_SRCS-$(HAVE_MMX) += common/x86/idctllm_mmx.asm
VP8_COMMON_SRCS-$(HAVE_MMX) += common/x86/iwalsh_mmx.asm
VP8_COMMON_SRCS-$(HAVE_MMX) += common/x86/recon_mmx.asm
VP8_COMMON_SRCS-$(HAVE_MMX) += common/x86/subpixel_mmx.asm
VP8_COMMON_SRCS-$(HAVE_MMX) += common/x86/loopfilter_mmx.asm
+VP8_COMMON_SRCS-$(HAVE_SSE2) += common/x86/idct_blk_sse2.c
VP8_COMMON_SRCS-$(HAVE_SSE2) += common/x86/idctllm_sse2.asm
VP8_COMMON_SRCS-$(HAVE_SSE2) += common/x86/recon_sse2.asm
VP8_COMMON_SRCS-$(HAVE_SSE2) += common/x86/recon_wrapper_sse2.c
@@ -115,6 +124,8 @@
VP8_COMMON_SRCS-$(ARCH_ARM) += common/arm/recon_arm.h
VP8_COMMON_SRCS-$(ARCH_ARM) += common/arm/reconintra_arm.c
VP8_COMMON_SRCS-$(ARCH_ARM) += common/arm/subpixel_arm.h
+VP8_COMMON_SRCS-$(ARCH_ARM) += common/arm/dequantize_arm.c
+VP8_COMMON_SRCS-$(ARCH_ARM) += common/arm/dequantize_arm.h
# common (armv6)
VP8_COMMON_SRCS-$(HAVE_ARMV6) += common/arm/armv6/bilinearfilter_v6$(ASM)
@@ -129,6 +140,9 @@
VP8_COMMON_SRCS-$(HAVE_ARMV6) += common/arm/armv6/simpleloopfilter_v6$(ASM)
VP8_COMMON_SRCS-$(HAVE_ARMV6) += common/arm/armv6/sixtappredict8x4_v6$(ASM)
VP8_COMMON_SRCS-$(HAVE_ARMV6) += common/arm/armv6/intra4x4_predict_v6$(ASM)
+VP8_COMMON_SRCS-$(HAVE_ARMV6) += common/arm/armv6/dequant_idct_v6$(ASM)
+VP8_COMMON_SRCS-$(HAVE_ARMV6) += common/arm/armv6/dequantize_v6$(ASM)
+VP8_COMMON_SRCS-$(HAVE_ARMV6) += common/arm/armv6/idct_blk_v6.c
# common (neon)
VP8_COMMON_SRCS-$(HAVE_ARMV7) += common/arm/neon/bilinearpredict4x4_neon$(ASM)
@@ -151,3 +165,8 @@
VP8_COMMON_SRCS-$(HAVE_ARMV7) += common/arm/neon/sixtappredict16x16_neon$(ASM)
VP8_COMMON_SRCS-$(HAVE_ARMV7) += common/arm/neon/buildintrapredictorsmby_neon$(ASM)
VP8_COMMON_SRCS-$(HAVE_ARMV7) += common/arm/neon/save_neon_reg$(ASM)
+VP8_COMMON_SRCS-$(HAVE_ARMV7) += common/arm/neon/dequant_idct_neon$(ASM)
+VP8_COMMON_SRCS-$(HAVE_ARMV7) += common/arm/neon/idct_dequant_full_2x_neon$(ASM)
+VP8_COMMON_SRCS-$(HAVE_ARMV7) += common/arm/neon/idct_dequant_0_2x_neon$(ASM)
+VP8_COMMON_SRCS-$(HAVE_ARMV7) += common/arm/neon/dequantizeb_neon$(ASM)
+VP8_COMMON_SRCS-$(HAVE_ARMV7) += common/arm/neon/idct_blk_neon.c
--- a/vp8/vp8dx.mk
+++ b/vp8/vp8dx.mk
@@ -52,7 +52,6 @@
VP8_DX_SRCS-yes += decoder/dboolhuff.c
VP8_DX_SRCS-yes += decoder/decodemv.c
VP8_DX_SRCS-yes += decoder/decodframe.c
-VP8_DX_SRCS-yes += decoder/dequantize.c
VP8_DX_SRCS-yes += decoder/detokenize.c
VP8_DX_SRCS-$(CONFIG_ERROR_CONCEALMENT) += decoder/ec_types.h
VP8_DX_SRCS-$(CONFIG_ERROR_CONCEALMENT) += decoder/error_concealment.h
@@ -61,20 +60,14 @@
VP8_DX_SRCS-yes += decoder/dboolhuff.h
VP8_DX_SRCS-yes += decoder/decodemv.h
VP8_DX_SRCS-yes += decoder/decoderthreading.h
-VP8_DX_SRCS-yes += decoder/dequantize.h
VP8_DX_SRCS-yes += decoder/detokenize.h
VP8_DX_SRCS-yes += decoder/onyxd_int.h
VP8_DX_SRCS-yes += decoder/treereader.h
VP8_DX_SRCS-yes += decoder/onyxd_if.c
VP8_DX_SRCS-$(CONFIG_MULTITHREAD) += decoder/threading.c
-VP8_DX_SRCS-yes += decoder/idct_blk.c
VP8_DX_SRCS-$(CONFIG_MULTITHREAD) += decoder/reconintra_mt.h
VP8_DX_SRCS-$(CONFIG_MULTITHREAD) += decoder/reconintra_mt.c
VP8_DX_SRCS-yes := $(filter-out $(VP8_DX_SRCS_REMOVE-yes),$(VP8_DX_SRCS-yes))
-VP8_DX_SRCS-$(ARCH_X86)$(ARCH_X86_64) += decoder/x86/dequantize_x86.h
VP8_DX_SRCS-$(ARCH_X86)$(ARCH_X86_64) += decoder/x86/x86_dsystemdependent.c
-VP8_DX_SRCS-$(HAVE_MMX) += decoder/x86/dequantize_mmx.asm
-VP8_DX_SRCS-$(HAVE_MMX) += decoder/x86/idct_blk_mmx.c
-VP8_DX_SRCS-$(HAVE_SSE2) += decoder/x86/idct_blk_sse2.c
--- a/vp8/vp8dx_arm.mk
+++ b/vp8/vp8dx_arm.mk
@@ -12,17 +12,3 @@
#VP8_DX_SRCS list is modified according to different platforms.
VP8_DX_SRCS-$(ARCH_ARM) += decoder/arm/arm_dsystemdependent.c
-VP8_DX_SRCS-$(ARCH_ARM) += decoder/arm/dequantize_arm.c
-VP8_DX_SRCS-$(ARCH_ARM) += decoder/arm/dequantize_arm.h
-
-#File list for armv6
-VP8_DX_SRCS-$(HAVE_ARMV6) += decoder/arm/armv6/dequant_idct_v6$(ASM)
-VP8_DX_SRCS-$(HAVE_ARMV6) += decoder/arm/armv6/dequantize_v6$(ASM)
-VP8_DX_SRCS-$(HAVE_ARMV6) += decoder/arm/armv6/idct_blk_v6.c
-
-#File list for neon
-VP8_DX_SRCS-$(HAVE_ARMV7) += decoder/arm/neon/dequant_idct_neon$(ASM)
-VP8_DX_SRCS-$(HAVE_ARMV7) += decoder/arm/neon/idct_dequant_full_2x_neon$(ASM)
-VP8_DX_SRCS-$(HAVE_ARMV7) += decoder/arm/neon/idct_dequant_0_2x_neon$(ASM)
-VP8_DX_SRCS-$(HAVE_ARMV7) += decoder/arm/neon/dequantizeb_neon$(ASM)
-VP8_DX_SRCS-$(HAVE_ARMV7) += decoder/arm/neon/idct_blk_neon.c
--
⑨