From 2b6a8187a6ae7407498888298e8d59f90cf94f9d Mon Sep 17 00:00:00 2001 From: Mirjana Vulin Date: Mon, 28 Jan 2013 17:47:53 +0100 Subject: mips: optimization for float aac decoder (core module) Signed-off-by: Mirjana Vulin Signed-off-by: Michael Niedermayer --- libavutil/mips/float_dsp_mips.c | 117 ++++++++++++++++++++++++++++++++++++++++ 1 file changed, 117 insertions(+) (limited to 'libavutil') diff --git a/libavutil/mips/float_dsp_mips.c b/libavutil/mips/float_dsp_mips.c index e39a4bf2bb..5211a265d4 100644 --- a/libavutil/mips/float_dsp_mips.c +++ b/libavutil/mips/float_dsp_mips.c @@ -106,6 +106,43 @@ static void vector_fmul_mips(float *dst, const float *src0, const float *src1, } } +static void vector_fmul_scalar_mips(float *dst, const float *src, float mul, + int len) +{ + float temp0, temp1, temp2, temp3; + float *local_src = (float*)src; + float *end = local_src + len; + + /* loop unrolled 4 times */ + __asm__ volatile( + ".set push \n\t" + ".set noreorder \n\t" + "1: \n\t" + "lwc1 %[temp0], 0(%[src]) \n\t" + "lwc1 %[temp1], 4(%[src]) \n\t" + "lwc1 %[temp2], 8(%[src]) \n\t" + "lwc1 %[temp3], 12(%[src]) \n\t" + "addiu %[dst], %[dst], 16 \n\t" + "mul.s %[temp0], %[temp0], %[mul] \n\t" + "mul.s %[temp1], %[temp1], %[mul] \n\t" + "mul.s %[temp2], %[temp2], %[mul] \n\t" + "mul.s %[temp3], %[temp3], %[mul] \n\t" + "addiu %[src], %[src], 16 \n\t" + "swc1 %[temp0], -16(%[dst]) \n\t" + "swc1 %[temp1], -12(%[dst]) \n\t" + "swc1 %[temp2], -8(%[dst]) \n\t" + "bne %[src], %[end], 1b \n\t" + " swc1 %[temp3], -4(%[dst]) \n\t" + ".set pop \n\t" + + : [temp0]"=&f"(temp0), [temp1]"=&f"(temp1), + [temp2]"=&f"(temp2), [temp3]"=&f"(temp3), + [dst]"+r"(dst), [src]"+r"(local_src) + : [end]"r"(end), [mul]"f"(mul) + : "memory" + ); +} + static void vector_fmul_window_mips(float *dst, const float *src0, const float *src1, const float *win, int len) { @@ -216,11 +253,91 @@ static void vector_fmul_window_mips(float *dst, const float *src0, ); } } + +static void butterflies_float_mips(float *av_restrict v1, float *av_restrict v2, + int len) +{ + float temp0, temp1, temp2, temp3, temp4; + float temp5, temp6, temp7, temp8, temp9; + float temp10, temp11, temp12, temp13, temp14, temp15; + int pom; + pom = (len >> 2)-1; + + /* loop unrolled 4 times */ + __asm__ volatile ( + "lwc1 %[temp0], 0(%[v1]) \n\t" + "lwc1 %[temp1], 4(%[v1]) \n\t" + "lwc1 %[temp2], 8(%[v1]) \n\t" + "lwc1 %[temp3], 12(%[v1]) \n\t" + "lwc1 %[temp4], 0(%[v2]) \n\t" + "lwc1 %[temp5], 4(%[v2]) \n\t" + "lwc1 %[temp6], 8(%[v2]) \n\t" + "lwc1 %[temp7], 12(%[v2]) \n\t" + "beq %[pom], $zero, 2f \n\t" + "1: \n\t" + "sub.s %[temp8], %[temp0], %[temp4] \n\t" + "add.s %[temp9], %[temp0], %[temp4] \n\t" + "sub.s %[temp10], %[temp1], %[temp5] \n\t" + "add.s %[temp11], %[temp1], %[temp5] \n\t" + "sub.s %[temp12], %[temp2], %[temp6] \n\t" + "add.s %[temp13], %[temp2], %[temp6] \n\t" + "sub.s %[temp14], %[temp3], %[temp7] \n\t" + "add.s %[temp15], %[temp3], %[temp7] \n\t" + "addiu %[v1], %[v1], 16 \n\t" + "addiu %[v2], %[v2], 16 \n\t" + "addiu %[pom], %[pom], -1 \n\t" + "lwc1 %[temp0], 0(%[v1]) \n\t" + "lwc1 %[temp1], 4(%[v1]) \n\t" + "lwc1 %[temp2], 8(%[v1]) \n\t" + "lwc1 %[temp3], 12(%[v1]) \n\t" + "lwc1 %[temp4], 0(%[v2]) \n\t" + "lwc1 %[temp5], 4(%[v2]) \n\t" + "lwc1 %[temp6], 8(%[v2]) \n\t" + "lwc1 %[temp7], 12(%[v2]) \n\t" + "swc1 %[temp9], -16(%[v1]) \n\t" + "swc1 %[temp8], -16(%[v2]) \n\t" + "swc1 %[temp11], -12(%[v1]) \n\t" + "swc1 %[temp10], -12(%[v2]) \n\t" + "swc1 %[temp13], -8(%[v1]) \n\t" + "swc1 %[temp12], -8(%[v2]) \n\t" + "swc1 %[temp15], -4(%[v1]) \n\t" + "swc1 %[temp14], -4(%[v2]) \n\t" + "bgtz %[pom], 1b \n\t" + "2: \n\t" + "sub.s %[temp8], %[temp0], %[temp4] \n\t" + "add.s %[temp9], %[temp0], %[temp4] \n\t" + "sub.s %[temp10], %[temp1], %[temp5] \n\t" + "add.s %[temp11], %[temp1], %[temp5] \n\t" + "sub.s %[temp12], %[temp2], %[temp6] \n\t" + "add.s %[temp13], %[temp2], %[temp6] \n\t" + "sub.s %[temp14], %[temp3], %[temp7] \n\t" + "add.s %[temp15], %[temp3], %[temp7] \n\t" + "swc1 %[temp9], 0(%[v1]) \n\t" + "swc1 %[temp8], 0(%[v2]) \n\t" + "swc1 %[temp11], 4(%[v1]) \n\t" + "swc1 %[temp10], 4(%[v2]) \n\t" + "swc1 %[temp13], 8(%[v1]) \n\t" + "swc1 %[temp12], 8(%[v2]) \n\t" + "swc1 %[temp15], 12(%[v1]) \n\t" + "swc1 %[temp14], 12(%[v2]) \n\t" + + : [v1]"+r"(v1), [v2]"+r"(v2), [pom]"+r"(pom), [temp0] "=&f" (temp0), + [temp1]"=&f"(temp1), [temp2]"=&f"(temp2), [temp3]"=&f"(temp3), + [temp4]"=&f"(temp4), [temp5]"=&f"(temp5), [temp6]"=&f"(temp6), + [temp7]"=&f"(temp7), [temp8]"=&f"(temp8), [temp9]"=&f"(temp9), + [temp10]"=&f"(temp10), [temp11]"=&f"(temp11), [temp12]"=&f"(temp12), + [temp13]"=&f"(temp13), [temp14]"=&f"(temp14), [temp15]"=&f"(temp15) + : + : "memory" + ); +} #endif /* HAVE_INLINE_ASM && HAVE_MIPSFPU */ void ff_float_dsp_init_mips(AVFloatDSPContext *fdsp) { #if HAVE_INLINE_ASM && HAVE_MIPSFPU fdsp->vector_fmul = vector_fmul_mips; + fdsp->vector_fmul_scalar = vector_fmul_scalar_mips; fdsp->vector_fmul_window = vector_fmul_window_mips; + fdsp->butterflies_float = butterflies_float_mips; #endif /* HAVE_INLINE_ASM && HAVE_MIPSFPU */ } -- cgit v1.2.3