Build only on x86_32 targets.
Signed-off-by: James Almer jamr...@gmail.com
---
libavcodec/x86/dcadsp.asm| 55 +---
libavcodec/x86/dcadsp_init.c | 44 +--
2 files changed, 69 insertions(+), 30 deletions(-)
diff --git a/libavcodec/x86/dcadsp.asm b/libavcodec/x86/dcadsp.asm
index 56039ba..970ec3d 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])
-pshufdm5, [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])
-pshufdm6, [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
subj, 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
-pshufdm0, m0, 0
+SPLATDm0
; 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 jr9q
-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, \
subi, (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 3821892..f8dd9b1 100644
--- a/libavcodec/x86/dcadsp_init.c
+++ b/libavcodec/x86/dcadsp_init.c
@@ -56,29 +56,39 @@ 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);
+#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
