1.5x-1.8x faster on sandybridge
---
libavutil/lls.c | 3 +++
libavutil/lls.h | 1 +
libavutil/x86/lls.asm | 31 +++++++++++++++++++++++++++++++
libavutil/x86/lls_init.c | 6 +++++-
4 files changed, 40 insertions(+), 1 deletion(-)
diff --git a/libavutil/lls.c b/libavutil/lls.c
index eb500af..8f1aff1 100644
--- a/libavutil/lls.c
+++ b/libavutil/lls.c
@@ -119,6 +119,9 @@ double avpriv_evaluate_lls(LLSModel *m, double *param, int
order)
int i;
double out = 0;
+ if (m->evaluate_lls)
+ return m->evaluate_lls(m->coeff[order], param, order);
+
for (i = 0; i <= order; i++)
out += param[i] * m->coeff[order][i];
diff --git a/libavutil/lls.h b/libavutil/lls.h
index 76ff10c..8a4d318 100644
--- a/libavutil/lls.h
+++ b/libavutil/lls.h
@@ -39,6 +39,7 @@ typedef struct LLSModel {
double variance[MAX_VARS];
int indep_count;
void (*update_lls)(struct LLSModel *m, double *var, int order);
+ double (*evaluate_lls)(double *coefs, double *var, int order);
} LLSModel;
void avpriv_init_lls(LLSModel *m, int indep_count);
diff --git a/libavutil/x86/lls.asm b/libavutil/x86/lls.asm
index f44f7ae..b5e04d9 100644
--- a/libavutil/x86/lls.asm
+++ b/libavutil/x86/lls.asm
@@ -182,3 +182,34 @@ cglobal update_lls, 3,6,8, covar, var, count, i, j, count2
jle .loop2x1
.ret:
REP_RET
+
+
+INIT_XMM sse2
+cglobal evaluate_lls, 3,3,3, coefs, var, i
+ ; This function is often called on the same buffer as update_lls, but with
an offset. They can't both be aligned.
+ movsd m0, [varq]
+ movhpd m0, [varq+8]
+ mulpd m0, [coefsq]
+ lea coefsq, [coefsq+iq*8]
+ lea varq, [varq+iq*8]
+ neg iq
+ add iq, 2
+.loop:
+ movsd m1, [varq+iq*8]
+ movhpd m1, [varq+iq*8+8]
+ mulpd m1, [coefsq+iq*8]
+ addpd m0, m1
+ add iq, 2
+ jl .loop
+ jg .skip1
+ movsd m1, [varq+iq*8]
+ mulsd m1, [coefsq+iq*8]
+ addpd m0, m1
+.skip1:
+ movhlps m1, m0
+ addsd m0, m1
+%if ARCH_X86_32
+ movsd r0m, m0
+ fld qword r0m
+%endif
+ RET
diff --git a/libavutil/x86/lls_init.c b/libavutil/x86/lls_init.c
index 1215b14..d65ddc8 100644
--- a/libavutil/x86/lls_init.c
+++ b/libavutil/x86/lls_init.c
@@ -25,12 +25,16 @@
void ff_update_lls_sse2(LLSModel *m, double *var, int order);
void ff_update_lls_avx(LLSModel *m, double *var, int order);
+double ff_evaluate_lls_sse2(double *coefs, double *var, int order);
void avpriv_init_lls_x86(LLSModel *m)
{
int cpu_flags = av_get_cpu_flags();
- if (EXTERNAL_SSE2(cpu_flags))
+ if (EXTERNAL_SSE2(cpu_flags)) {
m->update_lls = ff_update_lls_sse2;
+ if (m->indep_count >= 4)
+ m->evaluate_lls = ff_evaluate_lls_sse2;
+ }
if (EXTERNAL_AVX(cpu_flags))
m->update_lls = ff_update_lls_avx;
}
--
1.8.1.5
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel
