From c9311a12d5f948cbfab1732edff087e61ab9b8d4 Mon Sep 17 00:00:00 2001 From: Måns Rullgård Date: Thu, 28 May 2009 17:19:28 +0000 Subject: ARM: add some PLD in NEON IDCT Originally committed as revision 18972 to svn://svn.ffmpeg.org/ffmpeg/trunk --- libavcodec/arm/simple_idct_neon.S | 17 +++++++++++++++-- 1 file changed, 15 insertions(+), 2 deletions(-) diff --git a/libavcodec/arm/simple_idct_neon.S b/libavcodec/arm/simple_idct_neon.S index e7099a221e..08824812fa 100644 --- a/libavcodec/arm/simple_idct_neon.S +++ b/libavcodec/arm/simple_idct_neon.S @@ -68,6 +68,19 @@ .text .align 6 +function idct_row4_pld_neon + pld [r0] + add r3, r0, r1, lsl #2 + pld [r0, r1] + pld [r0, r1, lsl #1] + pld [r3, -r1] + pld [r3] + pld [r3, r1] + add r3, r3, r1, lsl #1 + pld [r3] + pld [r3, r1] + .endfunc + function idct_row4_neon vmov.i32 q15, #(1<<(ROW_SHIFT-1)) vld1.64 {d2-d5}, [r2,:128]! @@ -252,7 +265,7 @@ idct_coeff_neon: function ff_simple_idct_put_neon, export=1 idct_start r2 - bl idct_row4_neon + bl idct_row4_pld_neon bl idct_row4_neon add r2, r2, #-128 bl idct_col4_neon @@ -307,7 +320,7 @@ function idct_col4_add8_neon function ff_simple_idct_add_neon, export=1 idct_start r2 - bl idct_row4_neon + bl idct_row4_pld_neon bl idct_row4_neon add r2, r2, #-128 bl idct_col4_neon -- cgit v1.2.3