aboutsummaryrefslogtreecommitdiffstats
path: root/libavcodec/x86
diff options
context:
space:
mode:
authorJames Almer <jamrial@gmail.com>2014-03-20 15:37:54 -0300
committerAnton Khirnov <anton@khirnov.net>2014-04-04 17:40:51 +0200
commit2025d8026fe507c3f82c3440ecdf09641c577163 (patch)
tree7ba6a320addfb8c583e7134cca352c26d6a03128 /libavcodec/x86
parentb97f2c7c49e35724f3963677f8aeee28089dc64e (diff)
downloadffmpeg-2025d8026fe507c3f82c3440ecdf09641c577163.tar.gz
x86/synth_filter: add synth_filter_sse
Build only on x86_32 targets. Signed-off-by: James Almer <jamrial@gmail.com> Signed-off-by: Anton Khirnov <anton@khirnov.net>
Diffstat (limited to 'libavcodec/x86')
-rw-r--r--libavcodec/x86/dcadsp.asm55
-rw-r--r--libavcodec/x86/dcadsp_init.c45
2 files changed, 70 insertions, 30 deletions
diff --git a/libavcodec/x86/dcadsp.asm b/libavcodec/x86/dcadsp.asm
index 56039baa6f..970ec3dacb 100644
--- a/libavcodec/x86/dcadsp.asm
+++ b/libavcodec/x86/dcadsp.asm
@@ -199,15 +199,31 @@ INIT_XMM sse
DCA_LFE_FIR 0
DCA_LFE_FIR 1
-INIT_XMM sse2
+%macro SETZERO 1
+%if cpuflag(sse2)
+ pxor %1, %1
+%else
+ xorps %1, %1, %1
+%endif
+%endmacro
+
+%macro SHUF 2
+%if cpuflag(sse2)
+ pshufd %1, %2, q0123
+%else
+ mova %1, %2
+ shufps %1, %1, q0123
+%endif
+%endmacro
+
%macro INNER_LOOP 1
; reading backwards: ptr1 = synth_buf + j + i; ptr2 = synth_buf + j - i
;~ a += window[i + j] * (-synth_buf[15 - i + j])
;~ b += window[i + j + 16] * (synth_buf[i + j])
- pshufd m5, [ptr2 + j + (15 - 3) * 4], q0123
+ SHUF m5, [ptr2 + j + (15 - 3) * 4]
mova m6, [ptr1 + j]
%if ARCH_X86_64
- pshufd m11, [ptr2 + j + (15 - 3) * 4 - mmsize], q0123
+ SHUF m11, [ptr2 + j + (15 - 3) * 4 - mmsize]
mova m12, [ptr1 + j + mmsize]
%endif
mulps m6, [win + %1 + j + 16 * 4]
@@ -224,10 +240,10 @@ INIT_XMM sse2
%endif
;~ c += window[i + j + 32] * (synth_buf[16 + i + j])
;~ d += window[i + j + 48] * (synth_buf[31 - i + j])
- pshufd m6, [ptr2 + j + (31 - 3) * 4], q0123
+ SHUF m6, [ptr2 + j + (31 - 3) * 4]
mova m5, [ptr1 + j + 16 * 4]
%if ARCH_X86_64
- pshufd m12, [ptr2 + j + (31 - 3) * 4 - mmsize], q0123
+ SHUF m12, [ptr2 + j + (31 - 3) * 4 - mmsize]
mova m11, [ptr1 + j + mmsize + 16 * 4]
%endif
mulps m5, [win + %1 + j + 32 * 4]
@@ -245,20 +261,25 @@ INIT_XMM sse2
sub j, 64 * 4
%endmacro
-; void ff_synth_filter_inner_sse2(float *synth_buf, float synth_buf2[32],
-; const float window[512], float out[32],
-; intptr_t offset, float scale)
+; void ff_synth_filter_inner_<opt>(float *synth_buf, float synth_buf2[32],
+; const float window[512], float out[32],
+; intptr_t offset, float scale)
+%macro SYNTH_FILTER 0
cglobal synth_filter_inner, 0, 6 + 4 * ARCH_X86_64, 7 + 6 * ARCH_X86_64, \
synth_buf, synth_buf2, window, out, off, scale
%define scale m0
%if ARCH_X86_32 || WIN64
+%if cpuflag(sse2)
movd scale, scalem
+%else
+ movss scale, scalem
+%endif
; Make sure offset is in a register and not on the stack
%define OFFQ r4q
%else
%define OFFQ offq
%endif
- pshufd m0, m0, 0
+ SPLATD m0
; prepare inner counter limit 1
mov r5q, 480
sub r5q, offmp
@@ -274,8 +295,8 @@ cglobal synth_filter_inner, 0, 6 + 4 * ARCH_X86_64, 7 + 6 * ARCH_X86_64, \
%endif
.mainloop
; m1 = a m2 = b m3 = c m4 = d
- pxor m3, m3
- pxor m4, m4
+ SETZERO m3
+ SETZERO m4
mova m1, [buf2 + i]
mova m2, [buf2 + i + 16 * 4]
%if ARCH_X86_32
@@ -292,8 +313,8 @@ cglobal synth_filter_inner, 0, 6 + 4 * ARCH_X86_64, 7 + 6 * ARCH_X86_64, \
%define ptr2 r7q ; must be loaded
%define win r8q
%define j r9q
- pxor m9, m9
- pxor m10, m10
+ SETZERO m9
+ SETZERO m10
mova m7, [buf2 + i + mmsize]
mova m8, [buf2 + i + mmsize + 16 * 4]
lea win, [windowq + i]
@@ -350,3 +371,11 @@ cglobal synth_filter_inner, 0, 6 + 4 * ARCH_X86_64, 7 + 6 * ARCH_X86_64, \
sub i, (ARCH_X86_64 + 1) * mmsize
jge .mainloop
RET
+%endmacro
+
+%if ARCH_X86_32
+INIT_XMM sse
+SYNTH_FILTER
+%endif
+INIT_XMM sse2
+SYNTH_FILTER
diff --git a/libavcodec/x86/dcadsp_init.c b/libavcodec/x86/dcadsp_init.c
index 65e3db5c6c..5b77985569 100644
--- a/libavcodec/x86/dcadsp_init.c
+++ b/libavcodec/x86/dcadsp_init.c
@@ -56,25 +56,31 @@ av_cold void ff_dcadsp_init_x86(DCADSPContext *s)
}
}
-void ff_synth_filter_inner_sse2(float *synth_buf_ptr, float synth_buf2[32],
- const float window[512],
- float out[32], intptr_t offset, float scale);
-#if HAVE_YASM
-static void synth_filter_sse2(FFTContext *imdct,
- float *synth_buf_ptr, int *synth_buf_offset,
- float synth_buf2[32], const float window[512],
- float out[32], const float in[32], float scale)
-{
- float *synth_buf= synth_buf_ptr + *synth_buf_offset;
-
- imdct->imdct_half(imdct, synth_buf, in);
+#define SYNTH_FILTER_FUNC(opt) \
+void ff_synth_filter_inner_##opt(float *synth_buf_ptr, float synth_buf2[32], \
+ const float window[512], \
+ float out[32], intptr_t offset, float scale); \
+static void synth_filter_##opt(FFTContext *imdct, \
+ float *synth_buf_ptr, int *synth_buf_offset, \
+ float synth_buf2[32], const float window[512], \
+ float out[32], const float in[32], float scale) \
+{ \
+ float *synth_buf= synth_buf_ptr + *synth_buf_offset; \
+ \
+ imdct->imdct_half(imdct, synth_buf, in); \
+ \
+ ff_synth_filter_inner_##opt(synth_buf, synth_buf2, window, \
+ out, *synth_buf_offset, scale); \
+ \
+ *synth_buf_offset = (*synth_buf_offset - 32) & 511; \
+} \
- ff_synth_filter_inner_sse2(synth_buf, synth_buf2, window,
- out, *synth_buf_offset, scale);
-
- *synth_buf_offset = (*synth_buf_offset - 32) & 511;
-}
+#if HAVE_YASM
+#if ARCH_X86_32
+SYNTH_FILTER_FUNC(sse)
+#endif
+SYNTH_FILTER_FUNC(sse2)
#endif /* HAVE_YASM */
av_cold void ff_synth_filter_init_x86(SynthFilterContext *s)
@@ -82,6 +88,11 @@ av_cold void ff_synth_filter_init_x86(SynthFilterContext *s)
#if HAVE_YASM
int cpu_flags = av_get_cpu_flags();
+#if ARCH_X86_32
+ if (EXTERNAL_SSE(cpu_flags)) {
+ s->synth_filter_float = synth_filter_sse;
+ }
+#endif
if (EXTERNAL_SSE2(cpu_flags)) {
s->synth_filter_float = synth_filter_sse2;
}