ref: b7a6f1d20e958ce4dbce4555637a5948390797b1
parent: f8e3d23b99ac0ce695a0515456d1d293e526a247
parent: 734b1b2041a209ace033481c0fc919164177a3fc
author: Fritz Koenig <frkoenig@google.com>
date: Mon Aug 22 08:32:12 EDT 2011
Merge "Revert "Reclasify optimized ssim calculations as SSE2.""
--- a/vp8/encoder/generic/csystemdependent.c
+++ b/vp8/encoder/generic/csystemdependent.c
@@ -94,14 +94,15 @@
#if !(CONFIG_REALTIME_ONLY)
cpi->rtcd.temporal.apply = vp8_temporal_filter_apply_c;
#endif
-#if CONFIG_INTERNAL_STATS
- cpi->rtcd.variance.ssimpf_8x8 = vp8_ssim_parms_8x8_c;
- cpi->rtcd.variance.ssimpf_16x16 = vp8_ssim_parms_16x16_c;
#endif
-#endif
// Pure C:
vp8_yv12_copy_partial_frame_ptr = vp8_yv12_copy_partial_frame;
+
+#if CONFIG_INTERNAL_STATS
+ cpi->rtcd.variance.ssimpf_8x8 = ssim_parms_8x8_c;
+ cpi->rtcd.variance.ssimpf = ssim_parms_c;
+#endif
#if ARCH_X86 || ARCH_X86_64
vp8_arch_x86_encoder_init(cpi);
--- a/vp8/encoder/ssim.c
+++ b/vp8/encoder/ssim.c
@@ -9,9 +9,18 @@
*/
+#include "vpx_scale/yv12config.h"
+#include "math.h"
#include "onyx_int.h"
-void vp8_ssim_parms_16x16_c
+#if CONFIG_RUNTIME_CPU_DETECT
+#define IF_RTCD(x) (x)
+#else
+#define IF_RTCD(x) NULL
+#endif
+
+
+void ssim_parms_c
(
unsigned char *s,
int sp,
@@ -37,7 +46,7 @@
}
}
}
-void vp8_ssim_parms_8x8_c
+void ssim_parms_8x8_c
(
unsigned char *s,
int sp,
@@ -98,7 +107,7 @@
const vp8_variance_rtcd_vtable_t *rtcd)
{
unsigned long sum_s=0,sum_r=0,sum_sq_s=0,sum_sq_r=0,sum_sxr=0;
- SSIMPF_INVOKE(rtcd,16x16)(s, sp, r, rp, &sum_s, &sum_r, &sum_sq_s, &sum_sq_r, &sum_sxr);
+ rtcd->ssimpf(s, sp, r, rp, &sum_s, &sum_r, &sum_sq_s, &sum_sq_r, &sum_sxr);
return similarity(sum_s, sum_r, sum_sq_s, sum_sq_r, sum_sxr, 256);
}
static double ssim_8x8(unsigned char *s,int sp, unsigned char *r,int rp,
@@ -105,7 +114,7 @@
const vp8_variance_rtcd_vtable_t *rtcd)
{
unsigned long sum_s=0,sum_r=0,sum_sq_s=0,sum_sq_r=0,sum_sxr=0;
- SSIMPF_INVOKE(rtcd,8x8)(s, sp, r, rp, &sum_s, &sum_r, &sum_sq_s, &sum_sq_r, &sum_sxr);
+ rtcd->ssimpf_8x8(s, sp, r, rp, &sum_s, &sum_r, &sum_sq_s, &sum_sq_r, &sum_sxr);
return similarity(sum_s, sum_r, sum_sq_s, sum_sq_r, sum_sxr, 64);
}
@@ -125,7 +134,7 @@
c1 = cc1*16;
c2 = cc2*16;
- SSIMPF_INVOKE(rtcd,16x16)(s, sp, r, rp, &sum_s, &sum_r, &sum_sq_s, &sum_sq_r, &sum_sxr);
+ rtcd->ssimpf(s, sp, r, rp, &sum_s, &sum_r, &sum_sq_s, &sum_sq_r, &sum_sxr);
ssim_n1 = (2*sum_s*sum_r+ c1);
ssim_n2 =((int64_t) 2*256*sum_sxr-(int64_t) 2*sum_s*sum_r+c2);
--- a/vp8/encoder/variance.h
+++ b/vp8/encoder/variance.h
@@ -320,16 +320,16 @@
#endif
extern prototype_get16x16prederror(vp8_variance_get4x4sse_cs);
+#ifndef vp8_ssimpf
+#define vp8_ssimpf ssim_parms_c
+#endif
+extern prototype_ssimpf(vp8_ssimpf)
+
#ifndef vp8_ssimpf_8x8
-#define vp8_ssimpf_8x8 vp8_ssim_parms_8x8_c
+#define vp8_ssimpf_8x8 ssim_parms_8x8_c
#endif
extern prototype_ssimpf(vp8_ssimpf_8x8)
-#ifndef vp8_ssimpf_16x16
-#define vp8_ssimpf_16x16 vp8_ssim_parms_16x16_c
-#endif
-extern prototype_ssimpf(vp8_ssimpf_16x16)
-
typedef prototype_sad(*vp8_sad_fn_t);
typedef prototype_sad_multi_same_address(*vp8_sad_multi_fn_t);
typedef prototype_sad_multi_same_address_1(*vp8_sad_multi1_fn_t);
@@ -394,7 +394,7 @@
#if CONFIG_INTERNAL_STATS
vp8_ssimpf_fn_t ssimpf_8x8;
- vp8_ssimpf_fn_t ssimpf_16x16;
+ vp8_ssimpf_fn_t ssimpf;
#endif
} vp8_variance_rtcd_vtable_t;
@@ -417,10 +417,8 @@
#if CONFIG_RUNTIME_CPU_DETECT
#define VARIANCE_INVOKE(ctx,fn) (ctx)->fn
-#define SSIMPF_INVOKE(ctx,fn) (ctx)->fn
#else
#define VARIANCE_INVOKE(ctx,fn) vp8_variance_##fn
-#define SSIMPF_INVOKE(ctx,fn) vp8_ssimpf_##fn
#endif
#endif
--- a/vp8/encoder/x86/ssim_opt.asm
+++ b/vp8/encoder/x86/ssim_opt.asm
@@ -44,7 +44,7 @@
paddd %1, xmm1
SUM_ACROSS_Q %1
%endmacro
-;void ssim_parms_sse2(
+;void ssim_parms_sse3(
; unsigned char *s,
; int sp,
; unsigned char *r,
@@ -61,8 +61,8 @@
; or pavgb At this point this is just meant to be first pass for calculating
; all the parms needed for 16x16 ssim so we can play with dssim as distortion
; in mode selection code.
-global sym(vp8_ssim_parms_16x16_sse2)
-sym(vp8_ssim_parms_16x16_sse2):
+global sym(vp8_ssim_parms_16x16_sse3)
+sym(vp8_ssim_parms_16x16_sse3):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 9
@@ -134,7 +134,7 @@
pop rbp
ret
-;void ssim_parms_sse2(
+;void ssim_parms_sse3(
; unsigned char *s,
; int sp,
; unsigned char *r,
@@ -151,8 +151,8 @@
; or pavgb At this point this is just meant to be first pass for calculating
; all the parms needed for 16x16 ssim so we can play with dssim as distortion
; in mode selection code.
-global sym(vp8_ssim_parms_8x8_sse2)
-sym(vp8_ssim_parms_8x8_sse2):
+global sym(vp8_ssim_parms_8x8_sse3)
+sym(vp8_ssim_parms_8x8_sse3):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 9
--- a/vp8/encoder/x86/variance_x86.h
+++ b/vp8/encoder/x86/variance_x86.h
@@ -140,8 +140,6 @@
extern prototype_variance(vp8_mse16x16_wmt);
extern prototype_variance2(vp8_get8x8var_sse2);
extern prototype_variance2(vp8_get16x16var_sse2);
-extern prototype_ssimpf(vp8_ssim_parms_8x8_sse2)
-extern prototype_ssimpf(vp8_ssim_parms_16x16_sse2)
#if !CONFIG_RUNTIME_CPU_DETECT
#undef vp8_variance_sad4x4
@@ -209,14 +207,6 @@
#undef vp8_variance_mse16x16
#define vp8_variance_mse16x16 vp8_mse16x16_wmt
-
-#if ARCH_X86_64
-#undef vp8_ssimpf_8x8
-#define vp8_ssimpf_8x8 vp8_ssim_parms_8x8_sse2
-
-#undef vp8_ssimpf_16x16
-#define vp8_ssimpf_16x16 vp8_ssim_parms_16x16_sse2
-#endif
#endif
#endif
--- a/vp8/encoder/x86/x86_csystemdependent.c
+++ b/vp8/encoder/x86/x86_csystemdependent.c
@@ -111,6 +111,29 @@
#endif
+#if HAVE_SSSE3
+#if CONFIG_INTERNAL_STATS
+#if ARCH_X86_64
+typedef void ssimpf
+(
+ unsigned char *s,
+ int sp,
+ unsigned char *r,
+ int rp,
+ unsigned long *sum_s,
+ unsigned long *sum_r,
+ unsigned long *sum_sq_s,
+ unsigned long *sum_sq_r,
+ unsigned long *sum_sxr
+);
+
+extern ssimpf vp8_ssim_parms_16x16_sse3;
+extern ssimpf vp8_ssim_parms_8x8_sse3;
+#endif
+#endif
+#endif
+
+
void vp8_arch_x86_encoder_init(VP8_COMP *cpi)
{
#if CONFIG_RUNTIME_CPU_DETECT
@@ -223,13 +246,6 @@
#if !(CONFIG_REALTIME_ONLY)
cpi->rtcd.temporal.apply = vp8_temporal_filter_apply_sse2;
#endif
-
-#if CONFIG_INTERNAL_STATS
-#if ARCH_X86_64
- cpi->rtcd.variance.ssimpf_8x8 = vp8_ssim_parms_8x8_sse2;
- cpi->rtcd.variance.ssimpf_16x16 = vp8_ssim_parms_16x16_sse2;
-#endif
-#endif
}
#endif
@@ -264,6 +280,14 @@
cpi->rtcd.variance.subpixvar16x16 = vp8_sub_pixel_variance16x16_ssse3;
cpi->rtcd.quantize.fastquantb = vp8_fast_quantize_b_ssse3;
+
+#if CONFIG_INTERNAL_STATS
+#if ARCH_X86_64
+ cpi->rtcd.variance.ssimpf_8x8 = vp8_ssim_parms_8x8_sse3;
+ cpi->rtcd.variance.ssimpf = vp8_ssim_parms_16x16_sse3;
+#endif
+#endif
+
}
#endif