[FFmpeg-cvslog] arm: check if AS supports .dn

[Janne Grunau]

ffmpeg | branch: master | Janne Grunau <janne-libav at jannau.net> | Thu May 29 14:21:05 2014 +0200| [896a5bff64264f4d01ed98eacc97a67260c1e17e] | committer: Janne Grunau

arm: check if AS supports .dn

Move the GNU as check before the arch specific asm checks since the .dn
check requires gas compatible assembler.

Disable the VC-1 motion compensation NEON asm which is the only part
using that directive. The integrated assembler in the upcoming clang 3.5
does not support .dn/.qn without plans to change that. Too much effort
to implement it while it is rarely used.

http://llvm.org/bugs/show_bug.cgi?id=18199.

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=896a5bff64264f4d01ed98eacc97a67260c1e17e
---

 configure                         |   33 ++++++++++++++++++++-------------
 libavcodec/arm/vc1dsp_init_neon.c |    4 ++++
 libavcodec/arm/vc1dsp_neon.S      |    4 ++++
 3 files changed, 28 insertions(+), 13 deletions(-)

diff --git a/configure b/configure
index 9c6109e..13bcdfe 100755
--- a/configure
+++ b/configure
@@ -1467,6 +1467,7 @@ SYSTEM_FUNCS="
 "
 
 TOOLCHAIN_FEATURES="
+    as_dn_directive
     asm_mod_q
     attribute_may_alias
     attribute_packed
@@ -3721,6 +3722,20 @@ unsigned int endian = 'B' << 24 | 'I' << 16 | 'G' << 8 | 'E';
 EOF
 od -t x1 $TMPO | grep -q '42 *49 *47 *45' && enable bigendian
 
+
+if enabled asm; then
+    enabled     arm         && nogas=die
+    enabled_all ppc altivec && nogas=warn
+    as=${gas:=$as}
+    check_as <<EOF && enable gnu_as || \
+        $nogas "GNU assembler not found, install gas-preprocessor"
+.macro m n
+\n: .int 0
+.endm
+m x
+EOF
+fi
+
 check_inline_asm inline_asm_labels '"1:\n"'
 
 if enabled aarch64; then
@@ -3739,7 +3754,6 @@ elif enabled arm; then
 
     check_cpp_condition stddef.h "defined __thumb__" && enable_weak thumb
     enabled thumb && check_cflags -mthumb || check_cflags -marm
-    nogas=die
 
     if     check_cpp_condition stddef.h "defined __ARM_PCS_VFP"; then
         enable vfp_args
@@ -3767,6 +3781,11 @@ EOF
 
     check_inline_asm asm_mod_q '"add r0, %Q0, %R0" :: "r"((long long)0)'
 
+    check_as <<EOF && enable as_dn_directive
+ra .dn d0.i16
+.unreq ra
+EOF
+
     [ $target_os != win32 ] && enabled_all armv6t2 shared !pic && enable_weak_pic
 
 elif enabled mips; then
@@ -3792,7 +3811,6 @@ elif enabled ppc; then
 
     # AltiVec flags: The FSF version of GCC differs from the Apple version
     if enabled altivec; then
-        nogas=warn
         check_cflags -maltivec -mabi=altivec &&
         { check_header altivec.h && inc_altivec_h="#include <altivec.h>" ; } ||
         check_cflags -faltivec
@@ -3872,17 +3890,6 @@ EOF
 
 fi
 
-if enabled asm; then
-    as=${gas:=$as}
-    check_as <<EOF && enable gnu_as || \
-        $nogas "GNU assembler not found, install gas-preprocessor"
-.macro m n
-\n: .int 0
-.endm
-m x
-EOF
-fi
-
 check_ldflags -Wl,--as-needed
 
 if check_func dlopen; then
diff --git a/libavcodec/arm/vc1dsp_init_neon.c b/libavcodec/arm/vc1dsp_init_neon.c
index 912e33c..9ded7a2 100644
--- a/libavcodec/arm/vc1dsp_init_neon.c
+++ b/libavcodec/arm/vc1dsp_init_neon.c
@@ -22,6 +22,8 @@
 #include "libavcodec/vc1dsp.h"
 #include "vc1dsp.h"
 
+#include "config.h"
+
 void ff_vc1_inv_trans_8x8_neon(int16_t *block);
 void ff_vc1_inv_trans_4x8_neon(uint8_t *dest, int linesize, int16_t *block);
 void ff_vc1_inv_trans_8x4_neon(uint8_t *dest, int linesize, int16_t *block);
@@ -91,6 +93,7 @@ av_cold void ff_vc1dsp_init_neon(VC1DSPContext *dsp)
     dsp->vc1_inv_trans_4x4_dc = ff_vc1_inv_trans_4x4_dc_neon;
 
     dsp->put_vc1_mspel_pixels_tab[ 0] = ff_put_pixels8x8_neon;
+    if (HAVE_AS_DN_DIRECTIVE) {
     dsp->put_vc1_mspel_pixels_tab[ 1] = ff_put_vc1_mspel_mc10_neon;
     dsp->put_vc1_mspel_pixels_tab[ 2] = ff_put_vc1_mspel_mc20_neon;
     dsp->put_vc1_mspel_pixels_tab[ 3] = ff_put_vc1_mspel_mc30_neon;
@@ -106,6 +109,7 @@ av_cold void ff_vc1dsp_init_neon(VC1DSPContext *dsp)
     dsp->put_vc1_mspel_pixels_tab[13] = ff_put_vc1_mspel_mc13_neon;
     dsp->put_vc1_mspel_pixels_tab[14] = ff_put_vc1_mspel_mc23_neon;
     dsp->put_vc1_mspel_pixels_tab[15] = ff_put_vc1_mspel_mc33_neon;
+    }
 
     dsp->put_no_rnd_vc1_chroma_pixels_tab[0] = ff_put_vc1_chroma_mc8_neon;
     dsp->avg_no_rnd_vc1_chroma_pixels_tab[0] = ff_avg_vc1_chroma_mc8_neon;
diff --git a/libavcodec/arm/vc1dsp_neon.S b/libavcodec/arm/vc1dsp_neon.S
index d1b24cf..fa87ede 100644
--- a/libavcodec/arm/vc1dsp_neon.S
+++ b/libavcodec/arm/vc1dsp_neon.S
@@ -24,6 +24,8 @@
 #include "libavutil/arm/asm.S"
 #include "neon.S"
 
+#include "config.h"
+
 @ Transpose rows into columns of a matrix of 16-bit elements. For 4x4, pass
 @ double-word registers, for 8x4, pass quad-word registers.
 .macro transpose16 r0, r1, r2, r3
@@ -661,6 +663,7 @@ function ff_vc1_inv_trans_4x4_neon, export=1
         bx              lr
 endfunc
 
+#if HAVE_AS_DN_DIRECTIVE
 @ The absolute value of multiplication constants from vc1_mspel_filter and vc1_mspel_{ver,hor}_filter_16bits.
 @ The sign is embedded in the code below that carries out the multiplication (mspel_filter{,.16}).
 #define MSPEL_MODE_1_MUL_CONSTANTS  4 53 18 3
@@ -984,6 +987,7 @@ PUT_VC1_MSPEL_MC_V_ONLY(2)
 PUT_VC1_MSPEL_MC_V_ONLY(3)
 
 #undef PUT_VC1_MSPEL_MC_V_ONLY
+#endif
 
 function ff_put_pixels8x8_neon, export=1
         vld1.64         {d0}, [r1], r2