Re: [libav-devel] [PATCH 9/9] audiodsp/x86: yasmify vector_clipf_sse

2016-09-06 Thread Henrik Gramner
On Tue, Sep 6, 2016 at 11:39 AM, Anton Khirnov wrote: >> Use 3-arg maxps instead of mova. > > Isn't that AVX-only? It is, x86inc will simply convert it to mova+minps when assembling it as non-AVX code but it reduces the line count. It's certainly not worth to go into

Re: [libav-devel] [PATCH 9/9] audiodsp/x86: yasmify vector_clipf_sse

2016-09-06 Thread Anton Khirnov
Quoting Henrik Gramner (2016-09-05 15:15:14) > On Mon, Sep 5, 2016 at 1:02 PM, Anton Khirnov wrote: > > +cglobal vector_clipf, 3, 3, 6, dst, src, len, min, max > > +%if ARCH_X86_32 > > +VBROADCASTSS m0, minm > > +VBROADCASTSS m1, maxm > > +%else > > +VBROADCASTSS

Re: [libav-devel] [PATCH 9/9] audiodsp/x86: yasmify vector_clipf_sse

2016-09-06 Thread Diego Biurrun
On Mon, Sep 05, 2016 at 01:02:43PM +0200, Anton Khirnov wrote: > --- > libavcodec/x86/Makefile | 1 - > libavcodec/x86/audiodsp.asm | 42 +++ > libavcodec/x86/audiodsp_mmx.c | 58 > --- > 3 files changed, 42

Re: [libav-devel] [PATCH 9/9] audiodsp/x86: yasmify vector_clipf_sse

2016-09-05 Thread James Almer
On 9/5/2016 11:40 AM, James Almer wrote: > On 9/5/2016 10:15 AM, Henrik Gramner wrote: >> On Mon, Sep 5, 2016 at 1:02 PM, Anton Khirnov wrote: +cglobal vector_clipf, 3, 3, 6, dst, src, len, min, max +%if ARCH_X86_32 +VBROADCASTSS m0, minm +

Re: [libav-devel] [PATCH 9/9] audiodsp/x86: yasmify vector_clipf_sse

2016-09-05 Thread James Almer
On 9/5/2016 8:02 AM, Anton Khirnov wrote: > --- > libavcodec/x86/Makefile | 1 - > libavcodec/x86/audiodsp.asm | 42 +++ > libavcodec/x86/audiodsp_mmx.c | 58 > --- > 3 files changed, 42 insertions(+), 59 deletions(-) >

Re: [libav-devel] [PATCH 9/9] audiodsp/x86: yasmify vector_clipf_sse

2016-09-05 Thread James Almer
On 9/5/2016 10:15 AM, Henrik Gramner wrote: > On Mon, Sep 5, 2016 at 1:02 PM, Anton Khirnov wrote: >> > +cglobal vector_clipf, 3, 3, 6, dst, src, len, min, max >> > +%if ARCH_X86_32 >> > +VBROADCASTSS m0, minm >> > +VBROADCASTSS m1, maxm >> > +%else >> > +

Re: [libav-devel] [PATCH 9/9] audiodsp/x86: yasmify vector_clipf_sse

2016-09-05 Thread Henrik Gramner
On Mon, Sep 5, 2016 at 1:02 PM, Anton Khirnov wrote: > +shl lenq, 2 You could also skip this shift and just use 4*lenq instead in the memory operands, multiplying by 2, 4, or 8 in memory args is free. ___ libav-devel mailing list

Re: [libav-devel] [PATCH 9/9] audiodsp/x86: yasmify vector_clipf_sse

2016-09-05 Thread Henrik Gramner
On Mon, Sep 5, 2016 at 1:02 PM, Anton Khirnov wrote: > +cglobal vector_clipf, 3, 3, 6, dst, src, len, min, max > +%if ARCH_X86_32 > +VBROADCASTSS m0, minm > +VBROADCASTSS m1, maxm > +%else > +VBROADCASTSS m0, m0 > +VBROADCASTSS m1, m1 > +%endif This will fail

[libav-devel] [PATCH 9/9] audiodsp/x86: yasmify vector_clipf_sse

2016-09-05 Thread Anton Khirnov
--- libavcodec/x86/Makefile | 1 - libavcodec/x86/audiodsp.asm | 42 +++ libavcodec/x86/audiodsp_mmx.c | 58 --- 3 files changed, 42 insertions(+), 59 deletions(-) delete mode 100644 libavcodec/x86/audiodsp_mmx.c diff