| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | #include <asm/ppc_asm.h> |
| 2 | #include <asm/processor.h> |
| 3 | |
| 4 | /* |
| 5 | * The routines below are in assembler so we can closely control the |
| 6 | * usage of floating-point registers. These routines must be called |
| 7 | * with preempt disabled. |
| 8 | */ |
| 9 | .data |
| 10 | fpzero: |
| 11 | .long 0 |
| 12 | fpone: |
| 13 | .long 0x3f800000 /* 1.0 in single-precision FP */ |
| 14 | fphalf: |
| 15 | .long 0x3f000000 /* 0.5 in single-precision FP */ |
| 16 | |
| 17 | .text |
| 18 | /* |
| 19 | * Internal routine to enable floating point and set FPSCR to 0. |
| 20 | * Don't call it from C; it doesn't use the normal calling convention. |
| 21 | */ |
| 22 | fpenable: |
| 23 | mfmsr r10 |
| 24 | ori r11,r10,MSR_FP |
| 25 | mtmsr r11 |
| 26 | isync |
| 27 | stfd fr0,24(r1) |
| 28 | stfd fr1,16(r1) |
| 29 | stfd fr31,8(r1) |
| 30 | lis r11,fpzero@ha |
| 31 | mffs fr31 |
| 32 | lfs fr1,fpzero@l(r11) |
| 33 | mtfsf 0xff,fr1 |
| 34 | blr |
| 35 | |
| 36 | fpdisable: |
| 37 | mtfsf 0xff,fr31 |
| 38 | lfd fr31,8(r1) |
| 39 | lfd fr1,16(r1) |
| 40 | lfd fr0,24(r1) |
| 41 | mtmsr r10 |
| 42 | isync |
| 43 | blr |
| 44 | |
| 45 | /* |
| 46 | * Vector add, floating point. |
| 47 | */ |
| 48 | .globl vaddfp |
| 49 | vaddfp: |
| 50 | stwu r1,-32(r1) |
| 51 | mflr r0 |
| 52 | stw r0,36(r1) |
| 53 | bl fpenable |
| 54 | li r0,4 |
| 55 | mtctr r0 |
| 56 | li r6,0 |
| 57 | 1: lfsx fr0,r4,r6 |
| 58 | lfsx fr1,r5,r6 |
| 59 | fadds fr0,fr0,fr1 |
| 60 | stfsx fr0,r3,r6 |
| 61 | addi r6,r6,4 |
| 62 | bdnz 1b |
| 63 | bl fpdisable |
| 64 | lwz r0,36(r1) |
| 65 | mtlr r0 |
| 66 | addi r1,r1,32 |
| 67 | blr |
| 68 | |
| 69 | /* |
| 70 | * Vector subtract, floating point. |
| 71 | */ |
| 72 | .globl vsubfp |
| 73 | vsubfp: |
| 74 | stwu r1,-32(r1) |
| 75 | mflr r0 |
| 76 | stw r0,36(r1) |
| 77 | bl fpenable |
| 78 | li r0,4 |
| 79 | mtctr r0 |
| 80 | li r6,0 |
| 81 | 1: lfsx fr0,r4,r6 |
| 82 | lfsx fr1,r5,r6 |
| 83 | fsubs fr0,fr0,fr1 |
| 84 | stfsx fr0,r3,r6 |
| 85 | addi r6,r6,4 |
| 86 | bdnz 1b |
| 87 | bl fpdisable |
| 88 | lwz r0,36(r1) |
| 89 | mtlr r0 |
| 90 | addi r1,r1,32 |
| 91 | blr |
| 92 | |
| 93 | /* |
| 94 | * Vector multiply and add, floating point. |
| 95 | */ |
| 96 | .globl vmaddfp |
| 97 | vmaddfp: |
| 98 | stwu r1,-48(r1) |
| 99 | mflr r0 |
| 100 | stw r0,52(r1) |
| 101 | bl fpenable |
| 102 | stfd fr2,32(r1) |
| 103 | li r0,4 |
| 104 | mtctr r0 |
| 105 | li r7,0 |
| 106 | 1: lfsx fr0,r4,r7 |
| 107 | lfsx fr1,r5,r7 |
| 108 | lfsx fr2,r6,r7 |
| 109 | fmadds fr0,fr0,fr2,fr1 |
| 110 | stfsx fr0,r3,r7 |
| 111 | addi r7,r7,4 |
| 112 | bdnz 1b |
| 113 | lfd fr2,32(r1) |
| 114 | bl fpdisable |
| 115 | lwz r0,52(r1) |
| 116 | mtlr r0 |
| 117 | addi r1,r1,48 |
| 118 | blr |
| 119 | |
| 120 | /* |
| 121 | * Vector negative multiply and subtract, floating point. |
| 122 | */ |
| 123 | .globl vnmsubfp |
| 124 | vnmsubfp: |
| 125 | stwu r1,-48(r1) |
| 126 | mflr r0 |
| 127 | stw r0,52(r1) |
| 128 | bl fpenable |
| 129 | stfd fr2,32(r1) |
| 130 | li r0,4 |
| 131 | mtctr r0 |
| 132 | li r7,0 |
| 133 | 1: lfsx fr0,r4,r7 |
| 134 | lfsx fr1,r5,r7 |
| 135 | lfsx fr2,r6,r7 |
| 136 | fnmsubs fr0,fr0,fr2,fr1 |
| 137 | stfsx fr0,r3,r7 |
| 138 | addi r7,r7,4 |
| 139 | bdnz 1b |
| 140 | lfd fr2,32(r1) |
| 141 | bl fpdisable |
| 142 | lwz r0,52(r1) |
| 143 | mtlr r0 |
| 144 | addi r1,r1,48 |
| 145 | blr |
| 146 | |
| 147 | /* |
| 148 | * Vector reciprocal estimate. We just compute 1.0/x. |
| 149 | * r3 -> destination, r4 -> source. |
| 150 | */ |
| 151 | .globl vrefp |
| 152 | vrefp: |
| 153 | stwu r1,-32(r1) |
| 154 | mflr r0 |
| 155 | stw r0,36(r1) |
| 156 | bl fpenable |
| 157 | lis r9,fpone@ha |
| 158 | li r0,4 |
| 159 | lfs fr1,fpone@l(r9) |
| 160 | mtctr r0 |
| 161 | li r6,0 |
| 162 | 1: lfsx fr0,r4,r6 |
| 163 | fdivs fr0,fr1,fr0 |
| 164 | stfsx fr0,r3,r6 |
| 165 | addi r6,r6,4 |
| 166 | bdnz 1b |
| 167 | bl fpdisable |
| 168 | lwz r0,36(r1) |
| 169 | mtlr r0 |
| 170 | addi r1,r1,32 |
| 171 | blr |
| 172 | |
| 173 | /* |
| 174 | * Vector reciprocal square-root estimate, floating point. |
| 175 | * We use the frsqrte instruction for the initial estimate followed |
| 176 | * by 2 iterations of Newton-Raphson to get sufficient accuracy. |
| 177 | * r3 -> destination, r4 -> source. |
| 178 | */ |
| 179 | .globl vrsqrtefp |
| 180 | vrsqrtefp: |
| 181 | stwu r1,-48(r1) |
| 182 | mflr r0 |
| 183 | stw r0,52(r1) |
| 184 | bl fpenable |
| 185 | stfd fr2,32(r1) |
| 186 | stfd fr3,40(r1) |
| 187 | stfd fr4,48(r1) |
| 188 | stfd fr5,56(r1) |
| 189 | lis r9,fpone@ha |
| 190 | lis r8,fphalf@ha |
| 191 | li r0,4 |
| 192 | lfs fr4,fpone@l(r9) |
| 193 | lfs fr5,fphalf@l(r8) |
| 194 | mtctr r0 |
| 195 | li r6,0 |
| 196 | 1: lfsx fr0,r4,r6 |
| 197 | frsqrte fr1,fr0 /* r = frsqrte(s) */ |
| 198 | fmuls fr3,fr1,fr0 /* r * s */ |
| 199 | fmuls fr2,fr1,fr5 /* r * 0.5 */ |
| 200 | fnmsubs fr3,fr1,fr3,fr4 /* 1 - s * r * r */ |
| 201 | fmadds fr1,fr2,fr3,fr1 /* r = r + 0.5 * r * (1 - s * r * r) */ |
| 202 | fmuls fr3,fr1,fr0 /* r * s */ |
| 203 | fmuls fr2,fr1,fr5 /* r * 0.5 */ |
| 204 | fnmsubs fr3,fr1,fr3,fr4 /* 1 - s * r * r */ |
| 205 | fmadds fr1,fr2,fr3,fr1 /* r = r + 0.5 * r * (1 - s * r * r) */ |
| 206 | stfsx fr1,r3,r6 |
| 207 | addi r6,r6,4 |
| 208 | bdnz 1b |
| 209 | lfd fr5,56(r1) |
| 210 | lfd fr4,48(r1) |
| 211 | lfd fr3,40(r1) |
| 212 | lfd fr2,32(r1) |
| 213 | bl fpdisable |
| 214 | lwz r0,36(r1) |
| 215 | mtlr r0 |
| 216 | addi r1,r1,32 |
| 217 | blr |