[FFmpeg-cvslog] ARM: aacdec: fix constraints on inline asm
Mans Rullgard
git at videolan.org
Sun May 29 03:46:45 CEST 2011
ffmpeg | branch: master | Mans Rullgard <mans at mansr.com> | Fri May 27 22:52:49 2011 +0100| [7d8c17b5f62bc14d5f7c7c792630b23240b47eec] | committer: Mans Rullgard
ARM: aacdec: fix constraints on inline asm
This adds output operands for modified memory allowing the
volatile qualifiers to be dropped.
Signed-off-by: Mans Rullgard <mans at mansr.com>
> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=7d8c17b5f62bc14d5f7c7c792630b23240b47eec
---
libavcodec/arm/aac.h | 153 +++++++++++++++++++++++++------------------------
1 files changed, 78 insertions(+), 75 deletions(-)
diff --git a/libavcodec/arm/aac.h b/libavcodec/arm/aac.h
index 8355e9c..6d5df49 100644
--- a/libavcodec/arm/aac.h
+++ b/libavcodec/arm/aac.h
@@ -30,17 +30,17 @@ static inline float *VMUL2(float *dst, const float *v, unsigned idx,
const float *scale)
{
unsigned v0, v1;
- __asm__ volatile ("ubfx %0, %4, #0, #4 \n\t"
- "ubfx %1, %4, #4, #4 \n\t"
- "ldr %0, [%3, %0, lsl #2] \n\t"
- "ldr %1, [%3, %1, lsl #2] \n\t"
- "vld1.32 {d1[]}, [%5,:32] \n\t"
- "vmov d0, %0, %1 \n\t"
- "vmul.f32 d0, d0, d1 \n\t"
- "vst1.32 {d0}, [%2,:64]! \n\t"
- : "=&r"(v0), "=&r"(v1), "+r"(dst)
- : "r"(v), "r"(idx), "r"(scale)
- : "d0", "d1");
+ __asm__ ("ubfx %0, %6, #0, #4 \n\t"
+ "ubfx %1, %6, #4, #4 \n\t"
+ "ldr %0, [%5, %0, lsl #2] \n\t"
+ "ldr %1, [%5, %1, lsl #2] \n\t"
+ "vld1.32 {d1[]}, [%7,:32] \n\t"
+ "vmov d0, %0, %1 \n\t"
+ "vmul.f32 d0, d0, d1 \n\t"
+ "vst1.32 {d0}, [%2,:64]! \n\t"
+ : "=&r"(v0), "=&r"(v1), "+r"(dst), "=m"(dst[0]), "=m"(dst[1])
+ : "r"(v), "r"(idx), "r"(scale)
+ : "d0", "d1");
return dst;
}
@@ -49,22 +49,23 @@ static inline float *VMUL4(float *dst, const float *v, unsigned idx,
const float *scale)
{
unsigned v0, v1, v2, v3;
- __asm__ volatile ("ubfx %0, %6, #0, #2 \n\t"
- "ubfx %1, %6, #2, #2 \n\t"
- "ldr %0, [%5, %0, lsl #2] \n\t"
- "ubfx %2, %6, #4, #2 \n\t"
- "ldr %1, [%5, %1, lsl #2] \n\t"
- "ubfx %3, %6, #6, #2 \n\t"
- "ldr %2, [%5, %2, lsl #2] \n\t"
- "vmov d0, %0, %1 \n\t"
- "ldr %3, [%5, %3, lsl #2] \n\t"
- "vld1.32 {d2[],d3[]},[%7,:32] \n\t"
- "vmov d1, %2, %3 \n\t"
- "vmul.f32 q0, q0, q1 \n\t"
- "vst1.32 {q0}, [%4,:128]! \n\t"
- : "=&r"(v0), "=&r"(v1), "=&r"(v2), "=&r"(v3), "+r"(dst)
- : "r"(v), "r"(idx), "r"(scale)
- : "d0", "d1", "d2", "d3");
+ __asm__ ("ubfx %0, %10, #0, #2 \n\t"
+ "ubfx %1, %10, #2, #2 \n\t"
+ "ldr %0, [%9, %0, lsl #2] \n\t"
+ "ubfx %2, %10, #4, #2 \n\t"
+ "ldr %1, [%9, %1, lsl #2] \n\t"
+ "ubfx %3, %10, #6, #2 \n\t"
+ "ldr %2, [%9, %2, lsl #2] \n\t"
+ "vmov d0, %0, %1 \n\t"
+ "ldr %3, [%9, %3, lsl #2] \n\t"
+ "vld1.32 {d2[],d3[]},[%11,:32] \n\t"
+ "vmov d1, %2, %3 \n\t"
+ "vmul.f32 q0, q0, q1 \n\t"
+ "vst1.32 {q0}, [%4,:128]! \n\t"
+ : "=&r"(v0), "=&r"(v1), "=&r"(v2), "=&r"(v3), "+r"(dst),
+ "=m"(dst[0]), "=m"(dst[1]), "=m"(dst[2]), "=m"(dst[3])
+ : "r"(v), "r"(idx), "r"(scale)
+ : "d0", "d1", "d2", "d3");
return dst;
}
@@ -73,22 +74,23 @@ static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
unsigned sign, const float *scale)
{
unsigned v0, v1, v2, v3;
- __asm__ volatile ("ubfx %0, %6, #0, #4 \n\t"
- "ubfx %1, %6, #4, #4 \n\t"
- "ldr %0, [%5, %0, lsl #2] \n\t"
- "lsl %2, %8, #30 \n\t"
- "ldr %1, [%5, %1, lsl #2] \n\t"
- "lsl %3, %8, #31 \n\t"
- "vmov d0, %0, %1 \n\t"
- "bic %2, %2, #1<<30 \n\t"
- "vld1.32 {d1[]}, [%7,:32] \n\t"
- "vmov d2, %2, %3 \n\t"
- "veor d0, d0, d2 \n\t"
- "vmul.f32 d0, d0, d1 \n\t"
- "vst1.32 {d0}, [%4,:64]! \n\t"
- : "=&r"(v0), "=&r"(v1), "=&r"(v2), "=&r"(v3), "+r"(dst)
- : "r"(v), "r"(idx), "r"(scale), "r"(sign)
- : "d0", "d1", "d2");
+ __asm__ ("ubfx %0, %8, #0, #4 \n\t"
+ "ubfx %1, %8, #4, #4 \n\t"
+ "ldr %0, [%7, %0, lsl #2] \n\t"
+ "lsl %2, %10, #30 \n\t"
+ "ldr %1, [%7, %1, lsl #2] \n\t"
+ "lsl %3, %10, #31 \n\t"
+ "vmov d0, %0, %1 \n\t"
+ "bic %2, %2, #1<<30 \n\t"
+ "vld1.32 {d1[]}, [%9,:32] \n\t"
+ "vmov d2, %2, %3 \n\t"
+ "veor d0, d0, d2 \n\t"
+ "vmul.f32 d0, d0, d1 \n\t"
+ "vst1.32 {d0}, [%4,:64]! \n\t"
+ : "=&r"(v0), "=&r"(v1), "=&r"(v2), "=&r"(v3), "+r"(dst),
+ "=m"(dst[0]), "=m"(dst[1])
+ : "r"(v), "r"(idx), "r"(scale), "r"(sign)
+ : "d0", "d1", "d2");
return dst;
}
@@ -97,38 +99,39 @@ static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
unsigned sign, const float *scale)
{
unsigned v0, v1, v2, v3, nz;
- __asm__ volatile ("vld1.32 {d2[],d3[]},[%9,:32] \n\t"
- "ubfx %0, %8, #0, #2 \n\t"
- "ubfx %1, %8, #2, #2 \n\t"
- "ldr %0, [%7, %0, lsl #2] \n\t"
- "ubfx %2, %8, #4, #2 \n\t"
- "ldr %1, [%7, %1, lsl #2] \n\t"
- "ubfx %3, %8, #6, #2 \n\t"
- "ldr %2, [%7, %2, lsl #2] \n\t"
- "vmov d0, %0, %1 \n\t"
- "ldr %3, [%7, %3, lsl #2] \n\t"
- "lsr %6, %8, #12 \n\t"
- "rbit %6, %6 \n\t"
- "vmov d1, %2, %3 \n\t"
- "lsls %6, %6, #1 \n\t"
- "and %0, %5, #1<<31 \n\t"
- "lslcs %5, %5, #1 \n\t"
- "lsls %6, %6, #1 \n\t"
- "and %1, %5, #1<<31 \n\t"
- "lslcs %5, %5, #1 \n\t"
- "lsls %6, %6, #1 \n\t"
- "and %2, %5, #1<<31 \n\t"
- "lslcs %5, %5, #1 \n\t"
- "vmov d4, %0, %1 \n\t"
- "and %3, %5, #1<<31 \n\t"
- "vmov d5, %2, %3 \n\t"
- "veor q0, q0, q2 \n\t"
- "vmul.f32 q0, q0, q1 \n\t"
- "vst1.32 {q0}, [%4,:128]! \n\t"
- : "=&r"(v0), "=&r"(v1), "=&r"(v2), "=&r"(v3), "+r"(dst),
- "+r"(sign), "=r"(nz)
- : "r"(v), "r"(idx), "r"(scale)
- : "cc", "d0", "d1", "d2", "d3", "d4", "d5");
+ __asm__ ("vld1.32 {d2[],d3[]},[%13,:32] \n\t"
+ "ubfx %0, %12, #0, #2 \n\t"
+ "ubfx %1, %12, #2, #2 \n\t"
+ "ldr %0, [%11,%0, lsl #2] \n\t"
+ "ubfx %2, %12, #4, #2 \n\t"
+ "ldr %1, [%11,%1, lsl #2] \n\t"
+ "ubfx %3, %12, #6, #2 \n\t"
+ "ldr %2, [%11,%2, lsl #2] \n\t"
+ "vmov d0, %0, %1 \n\t"
+ "ldr %3, [%11,%3, lsl #2] \n\t"
+ "lsr %6, %12, #12 \n\t"
+ "rbit %6, %6 \n\t"
+ "vmov d1, %2, %3 \n\t"
+ "lsls %6, %6, #1 \n\t"
+ "and %0, %5, #1<<31 \n\t"
+ "lslcs %5, %5, #1 \n\t"
+ "lsls %6, %6, #1 \n\t"
+ "and %1, %5, #1<<31 \n\t"
+ "lslcs %5, %5, #1 \n\t"
+ "lsls %6, %6, #1 \n\t"
+ "and %2, %5, #1<<31 \n\t"
+ "lslcs %5, %5, #1 \n\t"
+ "vmov d4, %0, %1 \n\t"
+ "and %3, %5, #1<<31 \n\t"
+ "vmov d5, %2, %3 \n\t"
+ "veor q0, q0, q2 \n\t"
+ "vmul.f32 q0, q0, q1 \n\t"
+ "vst1.32 {q0}, [%4,:128]! \n\t"
+ : "=&r"(v0), "=&r"(v1), "=&r"(v2), "=&r"(v3), "+r"(dst),
+ "+r"(sign), "=r"(nz),
+ "=m"(dst[0]), "=m"(dst[1]), "=m"(dst[2]), "=m"(dst[3])
+ : "r"(v), "r"(idx), "r"(scale)
+ : "cc", "d0", "d1", "d2", "d3", "d4", "d5");
return dst;
}
More information about the ffmpeg-cvslog
mailing list