Skip to content

GitLab

Commit b8c2a4eb authored by Jian Zhou's avatar Jian Zhou Committed by Gerrit Code Review
Browse files

Merge "Code clean of highbd_tm_predictor_32x32"

parents dbe2d8c3 26a6ce4c
Hide whitespace changes
Inline Side-by-side
test/vp9_intrapred_test.cc
View file @ b8c2a4eb
......@@ -132,7 +132,6 @@ using std::tr1::make_tuple;
#if HAVE_SSE2
#if CONFIG_VP9_HIGHBITDEPTH
#if CONFIG_USE_X86INC
#if ARCH_X86_64
INSTANTIATE_TEST_CASE_P(SSE2_TO_C_8, VP9IntraPredTest,
::testing::Values(
make_tuple(&vpx_highbd_dc_predictor_32x32_sse2,
......@@ -159,34 +158,7 @@ INSTANTIATE_TEST_CASE_P(SSE2_TO_C_8, VP9IntraPredTest,
&vpx_highbd_tm_predictor_4x4_c, 4, 8),
make_tuple(&vpx_highbd_tm_predictor_8x8_sse2,
&vpx_highbd_tm_predictor_8x8_c, 8, 8)));
#else
INSTANTIATE_TEST_CASE_P(SSE2_TO_C_8, VP9IntraPredTest,
::testing::Values(
make_tuple(&vpx_highbd_dc_predictor_32x32_sse2,
&vpx_highbd_dc_predictor_32x32_c, 32, 8),
make_tuple(&vpx_highbd_tm_predictor_16x16_sse2,
&vpx_highbd_tm_predictor_16x16_c, 16, 8),
make_tuple(&vpx_highbd_dc_predictor_4x4_sse2,
&vpx_highbd_dc_predictor_4x4_c, 4, 8),
make_tuple(&vpx_highbd_dc_predictor_8x8_sse2,
&vpx_highbd_dc_predictor_8x8_c, 8, 8),
make_tuple(&vpx_highbd_dc_predictor_16x16_sse2,
&vpx_highbd_dc_predictor_16x16_c, 16, 8),
make_tuple(&vpx_highbd_v_predictor_4x4_sse2,
&vpx_highbd_v_predictor_4x4_c, 4, 8),
make_tuple(&vpx_highbd_v_predictor_8x8_sse2,
&vpx_highbd_v_predictor_8x8_c, 8, 8),
make_tuple(&vpx_highbd_v_predictor_16x16_sse2,
&vpx_highbd_v_predictor_16x16_c, 16, 8),
make_tuple(&vpx_highbd_v_predictor_32x32_sse2,
&vpx_highbd_v_predictor_32x32_c, 32, 8),
make_tuple(&vpx_highbd_tm_predictor_4x4_sse2,
&vpx_highbd_tm_predictor_4x4_c, 4, 8),
make_tuple(&vpx_highbd_tm_predictor_8x8_sse2,
&vpx_highbd_tm_predictor_8x8_c, 8, 8)));
#endif // !ARCH_X86_64
#if ARCH_X86_64
INSTANTIATE_TEST_CASE_P(SSE2_TO_C_10, VP9IntraPredTest,
::testing::Values(
make_tuple(&vpx_highbd_dc_predictor_32x32_sse2,
......@@ -219,37 +191,7 @@ INSTANTIATE_TEST_CASE_P(SSE2_TO_C_10, VP9IntraPredTest,
&vpx_highbd_tm_predictor_4x4_c, 4, 10),
make_tuple(&vpx_highbd_tm_predictor_8x8_sse2,
&vpx_highbd_tm_predictor_8x8_c, 8, 10)));
#else
INSTANTIATE_TEST_CASE_P(SSE2_TO_C_10, VP9IntraPredTest,
::testing::Values(
make_tuple(&vpx_highbd_dc_predictor_32x32_sse2,
&vpx_highbd_dc_predictor_32x32_c, 32,
10),
make_tuple(&vpx_highbd_tm_predictor_16x16_sse2,
&vpx_highbd_tm_predictor_16x16_c, 16,
10),
make_tuple(&vpx_highbd_dc_predictor_4x4_sse2,
&vpx_highbd_dc_predictor_4x4_c, 4, 10),
make_tuple(&vpx_highbd_dc_predictor_8x8_sse2,
&vpx_highbd_dc_predictor_8x8_c, 8, 10),
make_tuple(&vpx_highbd_dc_predictor_16x16_sse2,
&vpx_highbd_dc_predictor_16x16_c, 16,
10),
make_tuple(&vpx_highbd_v_predictor_4x4_sse2,
&vpx_highbd_v_predictor_4x4_c, 4, 10),
make_tuple(&vpx_highbd_v_predictor_8x8_sse2,
&vpx_highbd_v_predictor_8x8_c, 8, 10),
make_tuple(&vpx_highbd_v_predictor_16x16_sse2,
&vpx_highbd_v_predictor_16x16_c, 16, 10),
make_tuple(&vpx_highbd_v_predictor_32x32_sse2,
&vpx_highbd_v_predictor_32x32_c, 32, 10),
make_tuple(&vpx_highbd_tm_predictor_4x4_sse2,
&vpx_highbd_tm_predictor_4x4_c, 4, 10),
make_tuple(&vpx_highbd_tm_predictor_8x8_sse2,
&vpx_highbd_tm_predictor_8x8_c, 8, 10)));
#endif // !ARCH_X86_64
#if ARCH_X86_64
INSTANTIATE_TEST_CASE_P(SSE2_TO_C_12, VP9IntraPredTest,
::testing::Values(
make_tuple(&vpx_highbd_dc_predictor_32x32_sse2,
......@@ -282,35 +224,7 @@ INSTANTIATE_TEST_CASE_P(SSE2_TO_C_12, VP9IntraPredTest,
&vpx_highbd_tm_predictor_4x4_c, 4, 12),
make_tuple(&vpx_highbd_tm_predictor_8x8_sse2,
&vpx_highbd_tm_predictor_8x8_c, 8, 12)));
#else
INSTANTIATE_TEST_CASE_P(SSE2_TO_C_12, VP9IntraPredTest,
::testing::Values(
make_tuple(&vpx_highbd_dc_predictor_32x32_sse2,
&vpx_highbd_dc_predictor_32x32_c, 32,
12),
make_tuple(&vpx_highbd_tm_predictor_16x16_sse2,
&vpx_highbd_tm_predictor_16x16_c, 16,
12),
make_tuple(&vpx_highbd_dc_predictor_4x4_sse2,
&vpx_highbd_dc_predictor_4x4_c, 4, 12),
make_tuple(&vpx_highbd_dc_predictor_8x8_sse2,
&vpx_highbd_dc_predictor_8x8_c, 8, 12),
make_tuple(&vpx_highbd_dc_predictor_16x16_sse2,
&vpx_highbd_dc_predictor_16x16_c, 16,
12),
make_tuple(&vpx_highbd_v_predictor_4x4_sse2,
&vpx_highbd_v_predictor_4x4_c, 4, 12),
make_tuple(&vpx_highbd_v_predictor_8x8_sse2,
&vpx_highbd_v_predictor_8x8_c, 8, 12),
make_tuple(&vpx_highbd_v_predictor_16x16_sse2,
&vpx_highbd_v_predictor_16x16_c, 16, 12),
make_tuple(&vpx_highbd_v_predictor_32x32_sse2,
&vpx_highbd_v_predictor_32x32_c, 32, 12),
make_tuple(&vpx_highbd_tm_predictor_4x4_sse2,
&vpx_highbd_tm_predictor_4x4_c, 4, 12),
make_tuple(&vpx_highbd_tm_predictor_8x8_sse2,
&vpx_highbd_tm_predictor_8x8_c, 8, 12)));
#endif // !ARCH_X86_64
#endif // CONFIG_USE_X86INC
#endif // CONFIG_VP9_HIGHBITDEPTH
#endif // HAVE_SSE2
......
vpx_dsp/vpx_dsp_rtcd_defs.pl
View file @ b8c2a4eb
......@@ -435,7 +435,7 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
specialize qw/vpx_highbd_v_predictor_32x32/, "$sse2_x86inc";
add_proto qw/void vpx_highbd_tm_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
specialize qw/vpx_highbd_tm_predictor_32x32/, "$sse2_x86_64_x86inc";
specialize qw/vpx_highbd_tm_predictor_32x32/, "$sse2_x86inc";
add_proto qw/void vpx_highbd_dc_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
specialize qw/vpx_highbd_dc_predictor_32x32/, "$sse2_x86inc";
......
vpx_dsp/x86/highbd_intrapred_sse2.asm
View file @ b8c2a4eb
......@@ -385,9 +385,8 @@ cglobal highbd_tm_predictor_16x16, 5, 5, 8, dst, stride, above, left, bps
jnz .loop
REP_RET
%if ARCH_X86_64
INIT_XMM sse2
cglobal highbd_tm_predictor_32x32, 5, 6, 12, dst, stride, above, left, bps, one
cglobal highbd_tm_predictor_32x32, 5, 5, 8, dst, stride, above, left, bps
movd m0, [aboveq-2]
mova m1, [aboveq]
mova m2, [aboveq+16]
......@@ -395,70 +394,60 @@ cglobal highbd_tm_predictor_32x32, 5, 6, 12, dst, stride, above, left, bps, one
mova m4, [aboveq+48]
pshuflw m0, m0, 0x0
; Get the values to compute the maximum value at this bit depth
mov oned, 1
pxor m10, m10
pxor m11, m11
pinsrw m10, oned, 0
pinsrw m11, bpsd, 0
pshuflw m10, m10, 0x0
pcmpeqw m5, m5
movd m6, bpsd
psllw m5, m6
pcmpeqw m7, m7
pxor m6, m6 ; min possible value
pxor m5, m7 ; max possible value
punpcklqdq m0, m0
DEFINE_ARGS dst, stride, line, left
punpcklqdq m10, m10
mov lineq, -16
mova m5, m10
punpcklqdq m0, m0
psllw m10, m11
add leftq, 64
psubw m10, m5 ; max possible value
pxor m11, m11 ; min possible value
psubw m1, m0
psubw m2, m0
psubw m3, m0
psubw m4, m0
.loop:
movd m5, [leftq+lineq*4]
movd m6, [leftq+lineq*4+2]
pshuflw m5, m5, 0x0
pshuflw m6, m6, 0x0
punpcklqdq m5, m5
punpcklqdq m6, m6
paddw m7, m5, m1
paddw m8, m5, m2
paddw m9, m5, m3
paddw m5, m4
;Clamp these values to the bit-depth
pminsw m7, m10
pminsw m8, m10
pminsw m9, m10
pminsw m5, m10
pmaxsw m7, m11
pmaxsw m8, m11
pmaxsw m9, m11
pmaxsw m5, m11
;Store these values
mova [dstq ], m7
mova [dstq +16], m8
mova [dstq +32], m9
mova [dstq +48], m5
paddw m7, m6, m1
paddw m8, m6, m2
paddw m9, m6, m3
paddw m6, m4
;Clamp these values to the bit-depth
pminsw m7, m10
pminsw m8, m10
pminsw m9, m10
pminsw m6, m10
pmaxsw m7, m11
pmaxsw m8, m11
pmaxsw m9, m11
pmaxsw m6, m11
;Store these values
mova [dstq+strideq*2 ], m7
mova [dstq+strideq*2+16], m8
mova [dstq+strideq*2+32], m9
mova [dstq+strideq*2+48], m6
movd m7, [leftq]
pshuflw m7, m7, 0x0
punpcklqdq m7, m7 ; l1 l1 l1 l1 l1 l1 l1 l1
paddw m0, m7, m1
pminsw m0, m5
pmaxsw m0, m6
mova [dstq ], m0
paddw m0, m7, m2
pminsw m0, m5
pmaxsw m0, m6
mova [dstq +16], m0
paddw m0, m7, m3
pminsw m0, m5
pmaxsw m0, m6
mova [dstq +32], m0
paddw m0, m7, m4
pminsw m0, m5
pmaxsw m0, m6
mova [dstq +48], m0
movd m7, [leftq+2]
pshuflw m7, m7, 0x0
punpcklqdq m7, m7 ; l2 l2 l2 l2 l2 l2 l2 l2
paddw m0, m7, m1
pminsw m0, m5
pmaxsw m0, m6
mova [dstq+strideq*2 ], m0
paddw m0, m7, m2
pminsw m0, m5
pmaxsw m0, m6
mova [dstq+strideq*2+16], m0
paddw m0, m7, m3
pminsw m0, m5
pmaxsw m0, m6
mova [dstq+strideq*2+32], m0
paddw m0, m7, m4
pminsw m0, m5
pmaxsw m0, m6
mova [dstq+strideq*2+48], m0
lea dstq, [dstq+strideq*4]
lea leftq, [leftq+4]
inc lineq
jnz .loop
REP_RET
%endif
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment