| 5e993f12 |
1 | Index: codecs/gsm/src/k6opt.s |
| 2 | =================================================================== |
| 3 | --- codecs/gsm/src/k6opt.s (revision 49815) |
| 4 | +++ codecs/gsm/src/k6opt.s (working copy) |
| 5 | @@ -1,23 +1,24 @@ |
| 6 | .file "k6opt.s" |
| 7 | .version "01.01" |
| 8 | -/* gcc2_compiled.: */ |
| 9 | -.section .rodata |
| 10 | - .align 4 |
| 11 | - .type coefs,@object |
| 12 | - .size coefs,24 |
| 13 | -coefs: |
| 14 | - .value -134 |
| 15 | - .value -374 |
| 16 | - .value 0 |
| 17 | - .value 2054 |
| 18 | - .value 5741 |
| 19 | - .value 8192 |
| 20 | - .value 5741 |
| 21 | - .value 2054 |
| 22 | - .value 0 |
| 23 | - .value -374 |
| 24 | - .value -134 |
| 25 | - .value 0 |
| 26 | + |
| 27 | +/* load a mask into a mmx register ... use the stack to avoid TEXTRELS */ |
| 28 | +.macro load_immq mask reg |
| 29 | + pushl $h_\mask |
| 30 | + pushl $l_\mask |
| 31 | + movq (%esp), \reg |
| 32 | +.endm |
| 33 | +.macro cleanup_immq num |
| 34 | + addl \num * 8, %esp |
| 35 | +.endm |
| 36 | + |
| 37 | +/* some coefficients for us to play with */ |
| 38 | +.set h_coefs1, 0xFE8AFF7A /* -134 -374 */ |
| 39 | +.set l_coefs1, 0x08060000 /* 0 2504 */ |
| 40 | +.set h_coefs2, 0x2000166D /* 5741 8192 */ |
| 41 | +.set l_coefs2, 0x0806166D /* 5741 2054 */ |
| 42 | +.set h_coefs3, 0xFE8A0000 /* 0 -374 */ |
| 43 | +.set l_coefs3, 0x0000FF7A /* -134 0 */ |
| 44 | + |
| 45 | .text |
| 46 | .align 4 |
| 47 | /* void Weighting_filter (const short *e, short *x) */ |
| 48 | @@ -34,9 +35,10 @@ Weighting_filter: |
| 49 | addl $-10,%ebx |
| 50 | emms |
| 51 | movl $0x1000,%eax; movd %eax,%mm5 /* for rounding */ |
| 52 | - movq coefs,%mm1 |
| 53 | - movq coefs+8,%mm2 |
| 54 | - movq coefs+16,%mm3 |
| 55 | + load_immq coefs1, %mm1 |
| 56 | + load_immq coefs2, %mm2 |
| 57 | + load_immq coefs3, %mm3 |
| 58 | + cleanup_immq 3 |
| 59 | xorl %esi,%esi |
| 60 | .p2align 2 |
| 61 | .L21: |
| 62 | @@ -229,6 +231,8 @@ k6iprod: |
| 63 | |
| 64 | .align 4 |
| 65 | /* void k6vsraw P3((short *p, int n, int bits) */ |
| 66 | +.set h_ones, 0x00010001 |
| 67 | +.set l_ones, 0x00010001 |
| 68 | .globl k6vsraw |
| 69 | .type k6vsraw,@function |
| 70 | k6vsraw: |
| 71 | @@ -242,7 +246,8 @@ k6vsraw: |
| 72 | leal -16(%esi,%eax,2),%edx /* edx = top - 16 */ |
| 73 | emms |
| 74 | movd %ecx,%mm3 |
| 75 | - movq ones,%mm2 |
| 76 | + load_immq ones, %mm2 |
| 77 | + cleanup_immq 1 |
| 78 | psllw %mm3,%mm2; psrlw $1,%mm2 |
| 79 | cmpl %edx,%esi; ja .L306 |
| 80 | |
| 81 | @@ -362,22 +367,10 @@ k6vsllw: |
| 82 | .size k6vsllw,.Lfe5-k6vsllw |
| 83 | |
| 84 | |
| 85 | -.section .rodata |
| 86 | - .align 4 |
| 87 | - .type extremes,@object |
| 88 | - .size extremes,8 |
| 89 | -extremes: |
| 90 | - .long 0x80008000 |
| 91 | - .long 0x7fff7fff |
| 92 | - .type ones,@object |
| 93 | - .size ones,8 |
| 94 | -ones: |
| 95 | - .long 0x00010001 |
| 96 | - .long 0x00010001 |
| 97 | - |
| 98 | -.text |
| 99 | .align 4 |
| 100 | /* long k6maxmin (const short *p, int n, short *out) */ |
| 101 | +.set extremes_1, 0x80008000 |
| 102 | +.set extremes_2, 0x7fff7fff |
| 103 | .globl k6maxmin |
| 104 | .type k6maxmin,@function |
| 105 | k6maxmin: |
| 106 | @@ -391,8 +384,10 @@ k6maxmin: |
| 107 | |
| 108 | cmpl %edx,%esi |
| 109 | jbe .L52 |
| 110 | - movd extremes,%mm0 |
| 111 | - movd extremes+4,%mm1 |
| 112 | + mov $extremes_1, %ecx |
| 113 | + movd %ecx, %mm0 |
| 114 | + mov $extremes_2, %ecx |
| 115 | + movd %ecx, %mm1 |
| 116 | jmp .L58 |
| 117 | |
| 118 | .p2align 2 |
git.wh0rd.org / patches.git / blame