diff --git a/common/set.c b/common/set.c
index 6f4f471a2c51505384287fc0e4061ecb47c5d5b2..859a249d94d06486a2634ce4bf4739545788dfa2 100644
--- a/common/set.c
+++ b/common/set.c
@@ -158,6 +158,9 @@ int x264_cqm_init( x264_t *h )
quant8_mf[i_list][q][i] = DIV(def_quant8[q][i] * 16, h->sps->scaling_list[4+i_list][i]);
}
}
+
+#define MAX_MF X264_MIN( 0xffff, (1 << (25 - BIT_DEPTH)) - 1 )
+
for( int q = 0; q <= QP_MAX_SPEC; q++ )
{
int j;
@@ -174,9 +177,9 @@ int x264_cqm_init( x264_t *h )
// round to nearest, unless that would cause the deadzone to be negative
h->quant4_bias[i_list][q][i] = X264_MIN( DIV(deadzone[i_list]<<10, j), (1<<15)/j );
h->quant4_bias0[i_list][q][i] = (1<<15)/j;
- if( j > 0xffff && q > max_qp_err && (i_list == CQM_4IY || i_list == CQM_4PY) )
+ if( j > MAX_MF && q > max_qp_err && (i_list == CQM_4IY || i_list == CQM_4PY) )
max_qp_err = q;
- if( j > 0xffff && q > max_chroma_qp_err && (i_list == CQM_4IC || i_list == CQM_4PC) )
+ if( j > MAX_MF && q > max_chroma_qp_err && (i_list == CQM_4IC || i_list == CQM_4PC) )
max_chroma_qp_err = q;
}
if( h->param.analyse.b_transform_8x8 )
@@ -194,9 +197,9 @@ int x264_cqm_init( x264_t *h )
}
h->quant8_bias[i_list][q][i] = X264_MIN( DIV(deadzone[i_list]<<10, j), (1<<15)/j );
h->quant8_bias0[i_list][q][i] = (1<<15)/j;
- if( j > 0xffff && q > max_qp_err && (i_list == CQM_8IY || i_list == CQM_8PY) )
+ if( j > MAX_MF && q > max_qp_err && (i_list == CQM_8IY || i_list == CQM_8PY) )
max_qp_err = q;
- if( j > 0xffff && q > max_chroma_qp_err && (i_list == CQM_8IC || i_list == CQM_8PC) )
+ if( j > MAX_MF && q > max_chroma_qp_err && (i_list == CQM_8IC || i_list == CQM_8PC) )
max_chroma_qp_err = q;
}
}
diff --git a/common/x86/quant-a.asm b/common/x86/quant-a.asm
index 7a0dcd0a4c58fca2a02f2d7712de3631ccde1fca..af9c7236f6d1083de639ac0572de86d25d7b67bb 100644
--- a/common/x86/quant-a.asm
+++ b/common/x86/quant-a.asm
@@ -167,7 +167,7 @@ cextern popcnt_table
ABSD m1, m0
paddd m1, %3
pmulld m1, %2
- psrad m1, 16
+ psrld m1, 16
%else ; !sse4
mova m0, [%1]
ABSD m1, m0
@@ -195,8 +195,8 @@ cextern popcnt_table
paddd m3, %3
pmulld m2, %2
pmulld m3, %2
- psrad m2, 16
- psrad m3, 16
+ psrld m2, 16
+ psrld m3, 16
PSIGND m2, m0
PSIGND m3, m1
mova [%1 ], m2
@@ -222,7 +222,7 @@ cextern popcnt_table
pmuludq m3, m4
psllq m3, 32
paddd m1, m3
- psrad m1, 16
+ psrld m1, 16
PSIGND m1, m0
mova [%1], m1
ACCUM por, %5, 1, %4
@@ -238,8 +238,8 @@ cextern popcnt_table
paddd m3, [%3+mmsize]
pmulld m2, [%2 ]
pmulld m3, [%2+mmsize]
- psrad m2, 16
- psrad m3, 16
+ psrld m2, 16
+ psrld m3, 16
PSIGND m2, m0
PSIGND m3, m1
mova [%1 ], m2
diff --git a/tools/checkasm.c b/tools/checkasm.c
index 392d8696639c775bb680d0984285738b9e432e68..3a9c812f1775af75895c66670c46abf043c7b4e5 100644
--- a/tools/checkasm.c
+++ b/tools/checkasm.c
@@ -881,7 +881,7 @@ static int check_dct( uint32_t cpu_ref, uint32_t cpu_new )
h->param.analyse.i_luma_deadzone[0] = 0;
h->param.analyse.i_luma_deadzone[1] = 0;
h->param.analyse.b_transform_8x8 = 1;
- for( int i = 0; i < 6; i++ )
+ for( int i = 0; i < 8; i++ )
h->sps->scaling_list[i] = x264_cqm_flat16;
x264_cqm_init( h );
x264_quant_init( h, 0, &qf );
@@ -2054,30 +2054,65 @@ static int check_quant( uint32_t cpu_ref, uint32_t cpu_new )
h->chroma_qp_table = i_chroma_qp_table + 12;
h->param.analyse.b_transform_8x8 = 1;
- for( int i_cqm = 0; i_cqm < 4; i_cqm++ )
+ static const uint8_t cqm_test4[16] =
+ {
+ 6,4,6,4,
+ 4,3,4,3,
+ 6,4,6,4,
+ 4,3,4,3
+ };
+ static const uint8_t cqm_test8[64] =
+ {
+ 3,3,4,3,3,3,4,3,
+ 3,3,4,3,3,3,4,3,
+ 4,4,5,4,4,4,5,4,
+ 3,3,4,3,3,3,4,3,
+ 3,3,4,3,3,3,4,3,
+ 3,3,4,3,3,3,4,3,
+ 4,4,5,4,4,4,5,4,
+ 3,3,4,3,3,3,4,3
+ };
+
+ for( int i_cqm = 0; i_cqm < 6; i_cqm++ )
{
if( i_cqm == 0 )
{
- for( int i = 0; i < 6; i++ )
+ for( int i = 0; i < 8; i++ )
h->sps->scaling_list[i] = x264_cqm_flat16;
h->param.i_cqm_preset = h->sps->i_cqm_preset = X264_CQM_FLAT;
}
else if( i_cqm == 1 )
{
- for( int i = 0; i < 6; i++ )
+ for( int i = 0; i < 8; i++ )
h->sps->scaling_list[i] = x264_cqm_jvt[i];
h->param.i_cqm_preset = h->sps->i_cqm_preset = X264_CQM_JVT;
}
+ else if( i_cqm == 2 )
+ {
+ for( int i = 0; i < 4; i++ )
+ h->sps->scaling_list[i] = cqm_test4;
+ for( int i = 4; i < 8; i++ )
+ h->sps->scaling_list[i] = x264_cqm_flat16;
+ h->param.i_cqm_preset = h->sps->i_cqm_preset = X264_CQM_CUSTOM;
+ }
+ else if( i_cqm == 3 )
+ {
+ for( int i = 0; i < 4; i++ )
+ h->sps->scaling_list[i] = x264_cqm_flat16;
+ for( int i = 4; i < 8; i++ )
+ h->sps->scaling_list[i] = cqm_test8;
+ h->param.i_cqm_preset = h->sps->i_cqm_preset = X264_CQM_CUSTOM;
+ }
else
{
int max_scale = BIT_DEPTH < 10 ? 255 : 228;
- if( i_cqm == 2 )
+ if( i_cqm == 4 )
for( int i = 0; i < 64; i++ )
cqm_buf[i] = 10 + rand() % (max_scale - 9);
else
for( int i = 0; i < 64; i++ )
cqm_buf[i] = 1;
- for( int i = 0; i < 6; i++ )
+ for( int i = 0; i < 8; i++ )
h->sps->scaling_list[i] = cqm_buf;
h->param.i_cqm_preset = h->sps->i_cqm_preset = X264_CQM_CUSTOM;
}
@@ -2094,8 +2129,8 @@ static int check_quant( uint32_t cpu_ref, uint32_t cpu_new )
static const int scale1d[8] = {32,31,24,31,32,31,24,31}; \
for( int i = 0; i < max; i++ ) \
{ \
- unsigned int scale = (255*scale1d[(i>>3)&7]*scale1d[i&7])/16; \
- dct1[i] = dct2[i] = (j>>(i>>6))&1 ? (rand()%(2*scale+1))-scale : 0; \
+ int scale = (PIXEL_MAX*scale1d[(i>>3)&7]*scale1d[i&7])/16; \
+ dct1[i] = dct2[i] = (j>>(i>>6))&1 ? (rand30()%(2*scale+1))-scale : 0; \
} \
}
@@ -2104,8 +2139,8 @@ static int check_quant( uint32_t cpu_ref, uint32_t cpu_new )
static const int scale1d[4] = {4,6,4,6}; \
for( int i = 0; i < max; i++ ) \
{ \
- unsigned int scale = 255*scale1d[(i>>2)&3]*scale1d[i&3]; \
- dct1[i] = dct2[i] = (j>>(i>>4))&1 ? (rand()%(2*scale+1))-scale : 0; \
+ int scale = PIXEL_MAX*scale1d[(i>>2)&3]*scale1d[i&3]; \
+ dct1[i] = dct2[i] = (j>>(i>>4))&1 ? (rand30()%(2*scale+1))-scale : 0; \
} \
}