diff ppc/fft_altivec.c @ 12089:5638941ec8ef libavcodec

PPC: convert Altivec FFT to pure assembler On PPC a leaf function has a 288-byte red zone below the stack pointer, sparing these functions the chore of setting up a full stack frame. When a function call is disguised within an inline asm block, the compiler might not adjust the stack pointer as required before a function call, resulting in the red zone being clobbered. Moving the entire function to pure asm avoids this problem and also results in somewhat better code.

--- a/ppc/fft_altivec.c	Sun Jul 04 18:33:43 2010 +0000
+++ b/ppc/fft_altivec.c	Sun Jul 04 18:33:47 2010 +0000
@@ -22,7 +22,6 @@
 #include "libavcodec/fft.h"
 #include "util_altivec.h"
 #include "types_altivec.h"
-#include "regs.h"
 
 /**
  * Do a complex FFT with the parameters defined in ff_fft_init(). The
@@ -33,51 +32,10 @@
  * It also assumes all FFTComplex are 8 bytes-aligned pair of float
  */
 
-// Pointers to functions. Not using function pointer syntax, because
-// that involves an extra level of indirection on some PPC ABIs.
-extern void *ff_fft_dispatch_altivec[2][15];
+void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
+void ff_fft_calc_interleave_altivec(FFTContext *s, FFTComplex *z);
 
 #if HAVE_GNU_AS
-static av_always_inline void fft_dispatch(FFTContext *s, FFTComplex *z, int do_swizzle)
-{
-    register vec_f  v14 __asm__("v14") = {0,0,0,0};
-    register vec_f  v15 __asm__("v15") = *(const vec_f*)ff_cos_16;
-    register vec_f  v16 __asm__("v16") = {0, 0.38268343, M_SQRT1_2, 0.92387953};
-    register vec_f  v17 __asm__("v17") = {-M_SQRT1_2, M_SQRT1_2, M_SQRT1_2,-M_SQRT1_2};
-    register vec_f  v18 __asm__("v18") = { M_SQRT1_2, M_SQRT1_2, M_SQRT1_2, M_SQRT1_2};
-    register vec_u8 v19 __asm__("v19") = vcprm(s0,3,2,1);
-    register vec_u8 v20 __asm__("v20") = vcprm(0,1,s2,s1);
-    register vec_u8 v21 __asm__("v21") = vcprm(2,3,s0,s3);
-    register vec_u8 v22 __asm__("v22") = vcprm(2,s3,3,s2);
-    register vec_u8 v23 __asm__("v23") = vcprm(0,1,s0,s1);
-    register vec_u8 v24 __asm__("v24") = vcprm(2,3,s2,s3);
-    register vec_u8 v25 __asm__("v25") = vcprm(2,3,0,1);
-    register vec_u8 v26 __asm__("v26") = vcprm(1,2,s3,s0);
-    register vec_u8 v27 __asm__("v27") = vcprm(0,3,s2,s1);
-    register vec_u8 v28 __asm__("v28") = vcprm(0,2,s1,s3);
-    register vec_u8 v29 __asm__("v29") = vcprm(1,3,s0,s2);
-    register FFTSample *const*cos_tabs __asm__("r12") = ff_cos_tabs;
-    register FFTComplex *zarg __asm__("r3") = z;
-    __asm__(
-        "mtctr %0               \n"
-        "li   "r(9)", 16        \n"
-        "subi "r(1)","r(1) ",%1 \n"
-        "bctrl                  \n"
-        "addi "r(1)","r(1) ",%1 \n"
-        ::"r"(ff_fft_dispatch_altivec[do_swizzle][s->nbits-2]), "i"(12*sizeof(void*)),
-          "r"(zarg), "r"(cos_tabs),
-          "v"(v14),"v"(v15),"v"(v16),"v"(v17),"v"(v18),"v"(v19),"v"(v20),"v"(v21),
-          "v"(v22),"v"(v23),"v"(v24),"v"(v25),"v"(v26),"v"(v27),"v"(v28),"v"(v29)
-        : "lr","ctr","r0","r4","r5","r6","r7","r8","r9","r10","r11",
-          "v0","v1","v2","v3","v4","v5","v6","v7","v8","v9","v10","v11","v12","v13"
-    );
-}
-
-static void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z)
-{
-    fft_dispatch(s, z, 1);
-}
-
 static void ff_imdct_half_altivec(FFTContext *s, FFTSample *output, const FFTSample *input)
 {
     int j, k;
@@ -132,7 +90,7 @@
         k--;
     } while(k >= 0);
 
-    fft_dispatch(s, (FFTComplex*)output, 0);
+    ff_fft_calc_altivec(s, (FFTComplex*)output);
 
     /* post rotation + reordering */
     j = -n32;
@@ -182,7 +140,7 @@
 av_cold void ff_fft_init_altivec(FFTContext *s)
 {
 #if HAVE_GNU_AS
-    s->fft_calc = ff_fft_calc_altivec;
+    s->fft_calc   = ff_fft_calc_interleave_altivec;
     s->imdct_calc = ff_imdct_calc_altivec;
     s->imdct_half = ff_imdct_half_altivec;
 #endif