src/mesa/sparc/norm.S

   1 /* $Id: norm.S,v 1.1 2001/06/06 11:46:04 davem69 Exp $ */
   2
   3 #include "sparc_matrix.h"
   4
   5         .text
   6
   7 #ifdef __sparc_v9__
   8 #define STACK_VAR_OFF   (2047 + (8 * 16))
   9 #else
  10 #define STACK_VAR_OFF   (4 * 16)
  11 #endif
  12
  13         /* Newton-Raphson approximation turns out to be slower
  14          * (and less accurate) than direct fsqrts/fdivs.
  15          */
  16 #define ONE_DOT_ZERO    0x3f800000
  17
  18         .globl  _mesa_sparc_transform_normalize_normals
  19 _mesa_sparc_transform_normalize_normals:
  20         /* o0=mat o1=scale o2=in o3=lengths o4=dest */
  21
  22         sethi   %hi(ONE_DOT_ZERO), %g2
  23         sub     %sp, 16, %sp
  24         st      %g2, [%sp + STACK_VAR_OFF+0x0]
  25         st      %o1, [%sp + STACK_VAR_OFF+0x4]
  26         ld      [%sp + STACK_VAR_OFF+0x0], %f12 ! f12 = 1.0f
  27         ld      [%sp + STACK_VAR_OFF+0x4], %f15 ! f15 = scale
  28         add     %sp, 16, %sp
  29
  30         LDPTR   [%o0 + MAT_INV], %o0            ! o0 = mat->inv
  31         LDPTR   [%o2 + V4F_START], %o5          ! o5 = 'from' in->start
  32         ld      [%o2 + V4F_COUNT], %g1          ! g1 = in->count
  33         ld      [%o2 + V4F_STRIDE], %g2         ! g2 = in->stride
  34         LDPTR   [%o4 + V4F_START], %g3          ! g3 = 'out' dest->start
  35
  36         LDMATRIX_0_1_2_4_5_6_8_9_10(%o0)
  37
  38         /* dest->count = in->count */
  39         st      %g1, [%o4 + V4F_COUNT]
  40
  41         cmp     %g1, 1
  42         bl      7f
  43          cmp    %o3, 0
  44         bne     4f
  45          clr    %o4                             ! 'i' for STRIDE_LOOP
  46
  47 1:      /* LENGTHS == NULL */
  48         ld      [%o5 + 0x00], %f0               ! ux = from[0]
  49         ld      [%o5 + 0x04], %f1               ! uy = from[1]
  50         ld      [%o5 + 0x08], %f2               ! uz = from[2]
  51         add     %o5, %g2, %o5                   ! STRIDE_F(from, stride)
  52         add     %o4, 1, %o4                     ! i++
  53
  54         /* tx (f3) = (ux * m0) + (uy * m1) + (uz * m2)
  55          * ty (f5) = (ux * m4) + (uy * m5) + (uz * m6)
  56          * tz (f7) = (ux * m8) + (uy * m9) + (uz * m10)
  57          */
  58         fmuls   %f0, M0, %f3                    ! FGM   Group
  59         fmuls   %f1, M1, %f4                    ! FGM   Group
  60         fmuls   %f0, M4, %f5                    ! FGM   Group
  61         fmuls   %f1, M5, %f6                    ! FGM   Group
  62         fmuls   %f0, M8, %f7                    ! FGM   Group   f3 available
  63         fmuls   %f1, M9, %f8                    ! FGM   Group   f4 available
  64         fadds   %f3, %f4, %f3                   ! FGA
  65         fmuls   %f2, M2, %f10                   ! FGM   Group   f5 available
  66         fmuls   %f2, M6, %f0                    ! FGM   Group   f6 available
  67         fadds   %f5, %f6, %f5                   ! FGA
  68         fmuls   %f2, M10, %f4                   ! FGM   Group   f7 available
  69         fadds   %f7, %f8, %f7                   ! FGA   Group   f8,f3 available
  70         fadds   %f3, %f10, %f3                  ! FGA   Group   f10 available
  71         fadds   %f5, %f0, %f5                   ! FGA   Group   stall f0,f5 available
  72         fadds   %f7, %f4, %f7                   ! FGA   Group   stall f4,f7 available
  73
  74         /* f3=tx, f5=ty, f7=tz */
  75
  76         /* len (f6) = (tx * tx) + (ty * ty) + (tz * tz) */
  77         fmuls   %f3, %f3, %f6                   ! FGM   Group   f3 available
  78         fmuls   %f5, %f5, %f8                   ! FGM   Group   f5 available
  79         fmuls   %f7, %f7, %f10                  ! FGM   Group   f7 available
  80         fadds   %f6, %f8, %f6                   ! FGA   Group   2cyc stall f6,f8 available
  81         fadds   %f6, %f10, %f6                  ! FGA   Group   4cyc stall f6,f10 available
  82
  83         /* scale (f6) = 1.0 / sqrt(len) */
  84         fsqrts  %f6, %f6                        ! FDIV  20 cycles
  85         fdivs   %f12, %f6, %f6                  ! FDIV  14 cycles
  86
  87         fmuls   %f3, %f6, %f3
  88         st      %f3, [%g3 + 0x00]               ! out[i][0] = tx * scale
  89         fmuls   %f5, %f6, %f5
  90         st      %f5, [%g3 + 0x04]               ! out[i][1] = ty * scale
  91         fmuls   %f7, %f6, %f7
  92         st      %f7, [%g3 + 0x08]               ! out[i][2] = tz * scale
  93
  94         cmp     %o4, %g1                        ! continue if (i < count)
  95         bl      1b
  96          add    %g3, 0x0c, %g3                  ! advance out vector pointer
  97
  98         ba      7f
  99          nop
 100
 101 4:      /* LENGTHS != NULL */
 102         fmuls   M0, %f15, M0
 103         fmuls   M1, %f15, M1
 104         fmuls   M2, %f15, M2
 105         fmuls   M4, %f15, M4
 106         fmuls   M5, %f15, M5
 107         fmuls   M6, %f15, M6
 108         fmuls   M8, %f15, M8
 109         fmuls   M9, %f15, M9
 110         fmuls   M10, %f15, M10
 111
 112 5:
 113         ld      [%o5 + 0x00], %f0               ! ux = from[0]
 114         ld      [%o5 + 0x04], %f1               ! uy = from[1]
 115         ld      [%o5 + 0x08], %f2               ! uz = from[2]
 116         add     %o5, %g2, %o5                   ! STRIDE_F(from, stride)
 117         add     %o4, 1, %o4                     ! i++
 118
 119         /* tx (f3) = (ux * m0) + (uy * m1) + (uz * m2)
 120          * ty (f5) = (ux * m4) + (uy * m5) + (uz * m6)
 121          * tz (f7) = (ux * m8) + (uy * m9) + (uz * m10)
 122          */
 123         fmuls   %f0, M0, %f3                    ! FGM   Group
 124         fmuls   %f1, M1, %f4                    ! FGM   Group
 125         fmuls   %f0, M4, %f5                    ! FGM   Group
 126         fmuls   %f1, M5, %f6                    ! FGM   Group
 127         fmuls   %f0, M8, %f7                    ! FGM   Group   f3 available
 128         fmuls   %f1, M9, %f8                    ! FGM   Group   f4 available
 129         fadds   %f3, %f4, %f3                   ! FGA
 130         fmuls   %f2, M2, %f10                   ! FGM   Group   f5 available
 131         fmuls   %f2, M6, %f0                    ! FGM   Group   f6 available
 132         fadds   %f5, %f6, %f5                   ! FGA
 133         fmuls   %f2, M10, %f4                   ! FGM   Group   f7 available
 134         fadds   %f7, %f8, %f7                   ! FGA   Group   f8,f3 available
 135         fadds   %f3, %f10, %f3                  ! FGA   Group   f10 available
 136         ld      [%o3], %f13                     ! LSU
 137         fadds   %f5, %f0, %f5                   ! FGA   Group   stall f0,f5 available
 138         add     %o3, 4, %o3                     ! IEU0
 139         fadds   %f7, %f4, %f7                   ! FGA   Group   stall f4,f7 available
 140
 141         /* f3=tx, f5=ty, f7=tz, f13=lengths[i] */
 142
 143         fmuls   %f3, %f13, %f3
 144         st      %f3, [%g3 + 0x00]               ! out[i][0] = tx * len
 145         fmuls   %f5, %f13, %f5
 146         st      %f5, [%g3 + 0x04]               ! out[i][1] = ty * len
 147         fmuls   %f7, %f13, %f7
 148         st      %f7, [%g3 + 0x08]               ! out[i][2] = tz * len
 149
 150         cmp     %o4, %g1                        ! continue if (i < count)
 151         bl      5b
 152          add    %g3, 0x0c, %g3                  ! advance out vector pointer
 153
 154 7:      retl
 155          nop
 156
 157         .globl  _mesa_sparc_transform_normalize_normals_no_rot
 158 _mesa_sparc_transform_normalize_normals_no_rot:
 159         /* o0=mat o1=scale o2=in o3=lengths o4=dest */
 160
 161         sethi   %hi(ONE_DOT_ZERO), %g2
 162         sub     %sp, 16, %sp
 163         st      %g2, [%sp + STACK_VAR_OFF+0x0]
 164         st      %o1, [%sp + STACK_VAR_OFF+0x4]
 165         ld      [%sp + STACK_VAR_OFF+0x0], %f12 ! f12 = 1.0f
 166         ld      [%sp + STACK_VAR_OFF+0x4], %f15 ! f15 = scale
 167         add     %sp, 16, %sp
 168
 169         LDPTR   [%o0 + MAT_INV], %o0            ! o0 = mat->inv
 170         LDPTR   [%o2 + V4F_START], %o5          ! o5 = 'from' in->start
 171         ld      [%o2 + V4F_COUNT], %g1          ! g1 = in->count
 172         ld      [%o2 + V4F_STRIDE], %g2         ! g2 = in->stride
 173         LDPTR   [%o4 + V4F_START], %g3          ! g3 = 'out' dest->start
 174
 175         LDMATRIX_0_5_10(%o0)
 176
 177         /* dest->count = in->count */
 178         st      %g1, [%o4 + V4F_COUNT]
 179
 180         cmp     %g1, 1
 181         bl      7f
 182          cmp    %o3, 0
 183         bne     4f
 184          clr    %o4                             ! 'i' for STRIDE_LOOP
 185
 186 1:      /* LENGTHS == NULL */
 187         ld      [%o5 + 0x00], %f0               ! ux = from[0]
 188         ld      [%o5 + 0x04], %f1               ! uy = from[1]
 189         ld      [%o5 + 0x08], %f2               ! uz = from[2]
 190         add     %o5, %g2, %o5                   ! STRIDE_F(from, stride)
 191         add     %o4, 1, %o4                     ! i++
 192
 193         /* tx (f3) = (ux * m0)
 194          * ty (f5) = (uy * m5)
 195          * tz (f7) = (uz * m10)
 196          */
 197         fmuls   %f0, M0, %f3                    ! FGM   Group
 198         fmuls   %f1, M5, %f5                    ! FGM   Group
 199         fmuls   %f2, M10, %f7                   ! FGM   Group
 200
 201         /* f3=tx, f5=ty, f7=tz */
 202
 203         /* len (f6) = (tx * tx) + (ty * ty) + (tz * tz) */
 204         fmuls   %f3, %f3, %f6                   ! FGM   Group   stall, f3 available
 205         fmuls   %f5, %f5, %f8                   ! FGM   Group   f5 available
 206         fmuls   %f7, %f7, %f10                  ! FGM   Group   f7 available
 207         fadds   %f6, %f8, %f6                   ! FGA   Group   2cyc stall f6,f8 available
 208         fadds   %f6, %f10, %f6                  ! FGA   Group   4cyc stall f6,f10 available
 209
 210         /* scale (f6) = 1.0 / sqrt(len) */
 211         fsqrts  %f6, %f6                        ! FDIV  20 cycles
 212         fdivs   %f12, %f6, %f6                  ! FDIV  14 cycles
 213
 214         fmuls   %f3, %f6, %f3
 215         st      %f3, [%g3 + 0x00]               ! out[i][0] = tx * scale
 216         fmuls   %f5, %f6, %f5
 217         st      %f5, [%g3 + 0x04]               ! out[i][1] = ty * scale
 218         fmuls   %f7, %f6, %f7
 219         st      %f7, [%g3 + 0x08]               ! out[i][2] = tz * scale
 220
 221         cmp     %o4, %g1                        ! continue if (i < count)
 222         bl      1b
 223          add    %g3, 0x0c, %g3                  ! advance out vector pointer
 224
 225         ba      7f
 226          nop
 227
 228 4:      /* LENGTHS != NULL */
 229         fmuls   M0, %f15, M0
 230         fmuls   M5, %f15, M5
 231         fmuls   M10, %f15, M10
 232
 233 5:
 234         ld      [%o5 + 0x00], %f0               ! ux = from[0]
 235         ld      [%o5 + 0x04], %f1               ! uy = from[1]
 236         ld      [%o5 + 0x08], %f2               ! uz = from[2]
 237         add     %o5, %g2, %o5                   ! STRIDE_F(from, stride)
 238         add     %o4, 1, %o4                     ! i++
 239
 240         /* tx (f3) = (ux * m0)
 241          * ty (f5) = (uy * m5)
 242          * tz (f7) = (uz * m10)
 243          */
 244         fmuls   %f0, M0, %f3                    ! FGM   Group
 245         ld      [%o3], %f13                     ! LSU
 246         fmuls   %f1, M5, %f5                    ! FGM   Group
 247         add     %o3, 4, %o3                     ! IEU0
 248         fmuls   %f2, M10, %f7                   ! FGM   Group
 249
 250         /* f3=tx, f5=ty, f7=tz, f13=lengths[i] */
 251
 252         fmuls   %f3, %f13, %f3
 253         st      %f3, [%g3 + 0x00]               ! out[i][0] = tx * len
 254         fmuls   %f5, %f13, %f5
 255         st      %f5, [%g3 + 0x04]               ! out[i][1] = ty * len
 256         fmuls   %f7, %f13, %f7
 257         st      %f7, [%g3 + 0x08]               ! out[i][2] = tz * len
 258
 259         cmp     %o4, %g1                        ! continue if (i < count)
 260         bl      5b
 261          add    %g3, 0x0c, %g3                  ! advance out vector pointer
 262
 263 7:      retl
 264          nop
 265
 266         .globl  _mesa_sparc_transform_rescale_normals_no_rot
 267 _mesa_sparc_transform_rescale_normals_no_rot:
 268         /* o0=mat o1=scale o2=in o3=lengths o4=dest */
 269         sub     %sp, 16, %sp
 270         st      %o1, [%sp + STACK_VAR_OFF+0x0]
 271         ld      [%sp + STACK_VAR_OFF+0x0], %f15 ! f15 = scale
 272         add     %sp, 16, %sp
 273
 274         LDPTR   [%o0 + MAT_INV], %o0            ! o0 = mat->inv
 275         LDPTR   [%o2 + V4F_START], %o5          ! o5 = 'from' in->start
 276         ld      [%o2 + V4F_COUNT], %g1          ! g1 = in->count
 277         ld      [%o2 + V4F_STRIDE], %g2         ! g2 = in->stride
 278         LDPTR   [%o4 + V4F_START], %g3          ! g3 = 'out' dest->start
 279
 280         LDMATRIX_0_5_10(%o0)
 281
 282         /* dest->count = in->count */
 283         st      %g1, [%o4 + V4F_COUNT]
 284
 285         cmp     %g1, 1
 286         bl      7f
 287          clr    %o4                             ! 'i' for STRIDE_LOOP
 288
 289         fmuls   M0, %f15, M0
 290         fmuls   M5, %f15, M5
 291         fmuls   M10, %f15, M10
 292
 293 1:      ld      [%o5 + 0x00], %f0               ! ux = from[0]
 294         ld      [%o5 + 0x04], %f1               ! uy = from[1]
 295         ld      [%o5 + 0x08], %f2               ! uz = from[2]
 296         add     %o5, %g2, %o5                   ! STRIDE_F(from, stride)
 297         add     %o4, 1, %o4                     ! i++
 298
 299         /* tx (f3) = (ux * m0)
 300          * ty (f5) = (uy * m5)
 301          * tz (f7) = (uz * m10)
 302          */
 303         fmuls   %f0, M0, %f3                    ! FGM   Group
 304         st      %f3, [%g3 + 0x00]               ! LSU
 305         fmuls   %f1, M5, %f5                    ! FGM   Group
 306         st      %f5, [%g3 + 0x04]               ! LSU
 307         fmuls   %f2, M10, %f7                   ! FGM   Group
 308         st      %f7, [%g3 + 0x08]               ! LSU
 309
 310         cmp     %o4, %g1                        ! continue if (i < count)
 311         bl      1b
 312          add    %g3, 0x0c, %g3                  ! advance out vector pointer
 313
 314 7:      retl
 315          nop
 316
 317         .globl  _mesa_sparc_transform_rescale_normals
 318 _mesa_sparc_transform_rescale_normals:
 319         /* o0=mat o1=scale o2=in o3=lengths o4=dest */
 320         sub     %sp, 16, %sp
 321         st      %o1, [%sp + STACK_VAR_OFF+0x0]
 322         ld      [%sp + STACK_VAR_OFF+0x0], %f15 ! f15 = scale
 323         add     %sp, 16, %sp
 324
 325         LDPTR   [%o0 + MAT_INV], %o0            ! o0 = mat->inv
 326         LDPTR   [%o2 + V4F_START], %o5          ! o5 = 'from' in->start
 327         ld      [%o2 + V4F_COUNT], %g1          ! g1 = in->count
 328         ld      [%o2 + V4F_STRIDE], %g2         ! g2 = in->stride
 329         LDPTR   [%o4 + V4F_START], %g3          ! g3 = 'out' dest->start
 330
 331         LDMATRIX_0_1_2_4_5_6_8_9_10(%o0)
 332
 333         /* dest->count = in->count */
 334         st      %g1, [%o4 + V4F_COUNT]
 335
 336         cmp     %g1, 1
 337         bl      7f
 338          clr    %o4                             ! 'i' for STRIDE_LOOP
 339
 340         fmuls   M0, %f15, M0
 341         fmuls   M1, %f15, M1
 342         fmuls   M2, %f15, M2
 343         fmuls   M4, %f15, M4
 344         fmuls   M5, %f15, M5
 345         fmuls   M6, %f15, M6
 346         fmuls   M8, %f15, M8
 347         fmuls   M9, %f15, M9
 348         fmuls   M10, %f15, M10
 349
 350 1:      ld      [%o5 + 0x00], %f0               ! ux = from[0]
 351         ld      [%o5 + 0x04], %f1               ! uy = from[1]
 352         ld      [%o5 + 0x08], %f2               ! uz = from[2]
 353         add     %o5, %g2, %o5                   ! STRIDE_F(from, stride)
 354         add     %o4, 1, %o4                     ! i++
 355
 356         fmuls   %f0, M0, %f3                    ! FGM   Group
 357         fmuls   %f1, M1, %f4                    ! FGM   Group
 358         fmuls   %f0, M4, %f5                    ! FGM   Group
 359         fmuls   %f1, M5, %f6                    ! FGM   Group
 360         fmuls   %f0, M8, %f7                    ! FGM   Group   f3 available
 361         fmuls   %f1, M9, %f8                    ! FGM   Group   f4 available
 362         fadds   %f3, %f4, %f3                   ! FGA
 363         fmuls   %f2, M2, %f10                   ! FGM   Group   f5 available
 364         fmuls   %f2, M6, %f0                    ! FGM   Group   f6 available
 365         fadds   %f5, %f6, %f5                   ! FGA
 366         fmuls   %f2, M10, %f4                   ! FGM   Group   f7 available
 367         fadds   %f7, %f8, %f7                   ! FGA   Group   f8,f3 available
 368         fadds   %f3, %f10, %f3                  ! FGA   Group   f10 available
 369         st      %f3, [%g3 + 0x00]               ! LSU
 370         fadds   %f5, %f0, %f5                   ! FGA   Group   stall f0,f5 available
 371         st      %f5, [%g3 + 0x04]               ! LSU
 372         fadds   %f7, %f4, %f7                   ! FGA   Group   stall f4,f7 available
 373         st      %f7, [%g3 + 0x08]               ! LSU
 374
 375         cmp     %o4, %g1                        ! continue if (i < count)
 376         bl      1b
 377          add    %g3, 0x0c, %g3                  ! advance out vector pointer
 378
 379 7:      retl
 380          nop
 381
 382         .globl  _mesa_sparc_transform_normals_no_rot
 383 _mesa_sparc_transform_normals_no_rot:
 384         /* o0=mat o1=scale o2=in o3=lengths o4=dest */
 385         LDPTR   [%o0 + MAT_INV], %o0            ! o0 = mat->inv
 386         LDPTR   [%o2 + V4F_START], %o5          ! o5 = 'from' in->start
 387         ld      [%o2 + V4F_COUNT], %g1          ! g1 = in->count
 388         ld      [%o2 + V4F_STRIDE], %g2         ! g2 = in->stride
 389         LDPTR   [%o4 + V4F_START], %g3          ! g3 = 'out' dest->start
 390
 391         LDMATRIX_0_5_10(%o0)
 392
 393         /* dest->count = in->count */
 394         st      %g1, [%o4 + V4F_COUNT]
 395
 396         cmp     %g1, 1
 397         bl      7f
 398          clr    %o4                             ! 'i' for STRIDE_LOOP
 399
 400 1:      ld      [%o5 + 0x00], %f0               ! ux = from[0]
 401         ld      [%o5 + 0x04], %f1               ! uy = from[1]
 402         ld      [%o5 + 0x08], %f2               ! uz = from[2]
 403         add     %o5, %g2, %o5                   ! STRIDE_F(from, stride)
 404         add     %o4, 1, %o4                     ! i++
 405
 406         /* tx (f3) = (ux * m0)
 407          * ty (f5) = (uy * m5)
 408          * tz (f7) = (uz * m10)
 409          */
 410         fmuls   %f0, M0, %f3                    ! FGM   Group
 411         st      %f3, [%g3 + 0x00]               ! LSU
 412         fmuls   %f1, M5, %f5                    ! FGM   Group
 413         st      %f5, [%g3 + 0x04]               ! LSU
 414         fmuls   %f2, M10, %f7                   ! FGM   Group
 415         st      %f7, [%g3 + 0x08]               ! LSU
 416
 417         cmp     %o4, %g1                        ! continue if (i < count)
 418         bl      1b
 419          add    %g3, 0x0c, %g3                  ! advance out vector pointer
 420
 421 7:      retl
 422          nop
 423
 424         .globl  _mesa_sparc_transform_normals
 425 _mesa_sparc_transform_normals:
 426         /* o0=mat o1=scale o2=in o3=lengths o4=dest */
 427         LDPTR   [%o0 + MAT_INV], %o0            ! o0 = mat->inv
 428         LDPTR   [%o2 + V4F_START], %o5          ! o5 = 'from' in->start
 429         ld      [%o2 + V4F_COUNT], %g1          ! g1 = in->count
 430         ld      [%o2 + V4F_STRIDE], %g2         ! g2 = in->stride
 431         LDPTR   [%o4 + V4F_START], %g3          ! g3 = 'out' dest->start
 432
 433         LDMATRIX_0_1_2_4_5_6_8_9_10(%o0)
 434
 435         /* dest->count = in->count */
 436         st      %g1, [%o4 + V4F_COUNT]
 437
 438         cmp     %g1, 1
 439         bl      7f
 440          clr    %o4                             ! 'i' for STRIDE_LOOP
 441
 442 1:      ld      [%o5 + 0x00], %f0               ! ux = from[0]
 443         ld      [%o5 + 0x04], %f1               ! uy = from[1]
 444         ld      [%o5 + 0x08], %f2               ! uz = from[2]
 445         add     %o5, %g2, %o5                   ! STRIDE_F(from, stride)
 446         add     %o4, 1, %o4                     ! i++
 447
 448         fmuls   %f0, M0, %f3                    ! FGM   Group
 449         fmuls   %f1, M1, %f4                    ! FGM   Group
 450         fmuls   %f0, M4, %f5                    ! FGM   Group
 451         fmuls   %f1, M5, %f6                    ! FGM   Group
 452         fmuls   %f0, M8, %f7                    ! FGM   Group   f3 available
 453         fmuls   %f1, M9, %f8                    ! FGM   Group   f4 available
 454         fadds   %f3, %f4, %f3                   ! FGA
 455         fmuls   %f2, M2, %f10                   ! FGM   Group   f5 available
 456         fmuls   %f2, M6, %f0                    ! FGM   Group   f6 available
 457         fadds   %f5, %f6, %f5                   ! FGA
 458         fmuls   %f2, M10, %f4                   ! FGM   Group   f7 available
 459         fadds   %f7, %f8, %f7                   ! FGA   Group   f8,f3 available
 460         fadds   %f3, %f10, %f3                  ! FGA   Group   f10 available
 461         st      %f3, [%g3 + 0x00]               ! LSU
 462         fadds   %f5, %f0, %f5                   ! FGA   Group   stall f0,f5 available
 463         st      %f5, [%g3 + 0x04]               ! LSU
 464         fadds   %f7, %f4, %f7                   ! FGA   Group   stall f4,f7 available
 465         st      %f7, [%g3 + 0x08]               ! LSU
 466
 467         cmp     %o4, %g1                        ! continue if (i < count)
 468         bl      1b
 469          add    %g3, 0x0c, %g3                  ! advance out vector pointer
 470
 471 7:      retl
 472          nop
 473
 474         .globl  _mesa_sparc_normalize_normals
 475 _mesa_sparc_normalize_normals:
 476         /* o0=mat o1=scale o2=in o3=lengths o4=dest */
 477
 478         sethi   %hi(ONE_DOT_ZERO), %g2
 479         sub     %sp, 16, %sp
 480         st      %g2, [%sp + STACK_VAR_OFF+0x0]
 481         ld      [%sp + STACK_VAR_OFF+0x0], %f12 ! f12 = 1.0f
 482         add     %sp, 16, %sp
 483
 484         LDPTR   [%o2 + V4F_START], %o5          ! o5 = 'from' in->start
 485         ld      [%o2 + V4F_COUNT], %g1          ! g1 = in->count
 486         ld      [%o2 + V4F_STRIDE], %g2         ! g2 = in->stride
 487         LDPTR   [%o4 + V4F_START], %g3          ! g3 = 'out' dest->start
 488
 489         /* dest->count = in->count */
 490         st      %g1, [%o4 + V4F_COUNT]
 491
 492         cmp     %g1, 1
 493         bl      7f
 494          cmp    %o3, 0
 495         bne     4f
 496          clr    %o4                             ! 'i' for STRIDE_LOOP
 497
 498 1:      /* LENGTHS == NULL */
 499         ld      [%o5 + 0x00], %f3               ! ux = from[0]
 500         ld      [%o5 + 0x04], %f5               ! uy = from[1]
 501         ld      [%o5 + 0x08], %f7               ! uz = from[2]
 502         add     %o5, %g2, %o5                   ! STRIDE_F(from, stride)
 503         add     %o4, 1, %o4                     ! i++
 504
 505         /* f3=tx, f5=ty, f7=tz */
 506
 507         /* len (f6) = (tx * tx) + (ty * ty) + (tz * tz) */
 508         fmuls   %f3, %f3, %f6                   ! FGM   Group   f3 available
 509         fmuls   %f5, %f5, %f8                   ! FGM   Group   f5 available
 510         fmuls   %f7, %f7, %f10                  ! FGM   Group   f7 available
 511         fadds   %f6, %f8, %f6                   ! FGA   Group   2cyc stall f6,f8 available
 512         fadds   %f6, %f10, %f6                  ! FGA   Group   4cyc stall f6,f10 available
 513
 514         /* scale (f6) = 1.0 / sqrt(len) */
 515         fsqrts  %f6, %f6                        ! FDIV  20 cycles
 516         fdivs   %f12, %f6, %f6                  ! FDIV  14 cycles
 517
 518         fmuls   %f3, %f6, %f3
 519         st      %f3, [%g3 + 0x00]               ! out[i][0] = tx * scale
 520         fmuls   %f5, %f6, %f5
 521         st      %f5, [%g3 + 0x04]               ! out[i][1] = ty * scale
 522         fmuls   %f7, %f6, %f7
 523         st      %f7, [%g3 + 0x08]               ! out[i][2] = tz * scale
 524
 525         cmp     %o4, %g1                        ! continue if (i < count)
 526         bl      1b
 527          add    %g3, 0x0c, %g3                  ! advance out vector pointer
 528
 529         ba      7f
 530          nop
 531
 532 4:      /* LENGTHS != NULL */
 533
 534 5:
 535         ld      [%o5 + 0x00], %f3               ! ux = from[0]
 536         ld      [%o5 + 0x04], %f5               ! uy = from[1]
 537         ld      [%o5 + 0x08], %f7               ! uz = from[2]
 538         add     %o5, %g2, %o5                   ! STRIDE_F(from, stride)
 539         add     %o4, 1, %o4                     ! i++
 540
 541         ld      [%o3], %f13                     ! LSU
 542         add     %o3, 4, %o3                     ! IEU0
 543
 544         /* f3=tx, f5=ty, f7=tz, f13=lengths[i] */
 545
 546         fmuls   %f3, %f13, %f3
 547         st      %f3, [%g3 + 0x00]               ! out[i][0] = tx * len
 548         fmuls   %f5, %f13, %f5
 549         st      %f5, [%g3 + 0x04]               ! out[i][1] = ty * len
 550         fmuls   %f7, %f13, %f7
 551         st      %f7, [%g3 + 0x08]               ! out[i][2] = tz * len
 552
 553         cmp     %o4, %g1                        ! continue if (i < count)
 554         bl      5b
 555          add    %g3, 0x0c, %g3                  ! advance out vector pointer
 556
 557 7:      retl
 558          nop
 559
 560         .globl  _mesa_sparc_rescale_normals
 561 _mesa_sparc_rescale_normals:
 562         /* o0=mat o1=scale o2=in o3=lengths o4=dest */
 563
 564         sethi   %hi(ONE_DOT_ZERO), %g2
 565         sub     %sp, 16, %sp
 566         st      %o1, [%sp + STACK_VAR_OFF+0x0]
 567         ld      [%sp + STACK_VAR_OFF+0x0], %f15 ! f15 = scale
 568         add     %sp, 16, %sp
 569
 570         LDPTR   [%o2 + V4F_START], %o5          ! o5 = 'from' in->start
 571         ld      [%o2 + V4F_COUNT], %g1          ! g1 = in->count
 572         ld      [%o2 + V4F_STRIDE], %g2         ! g2 = in->stride
 573         LDPTR   [%o4 + V4F_START], %g3          ! g3 = 'out' dest->start
 574
 575         /* dest->count = in->count */
 576         st      %g1, [%o4 + V4F_COUNT]
 577
 578         cmp     %g1, 1
 579         bl      7f
 580          clr    %o4                             ! 'i' for STRIDE_LOOP
 581
 582 1:
 583         ld      [%o5 + 0x00], %f3               ! ux = from[0]
 584         ld      [%o5 + 0x04], %f5               ! uy = from[1]
 585         ld      [%o5 + 0x08], %f7               ! uz = from[2]
 586         add     %o5, %g2, %o5                   ! STRIDE_F(from, stride)
 587         add     %o4, 1, %o4                     ! i++
 588
 589         /* f3=tx, f5=ty, f7=tz */
 590
 591         fmuls   %f3, %f15, %f3
 592         st      %f3, [%g3 + 0x00]               ! out[i][0] = tx * scale
 593         fmuls   %f5, %f15, %f5
 594         st      %f5, [%g3 + 0x04]               ! out[i][1] = ty * scale
 595         fmuls   %f7, %f15, %f7
 596         st      %f7, [%g3 + 0x08]               ! out[i][2] = tz * scale
 597
 598         cmp     %o4, %g1                        ! continue if (i < count)
 599         bl      1b
 600          add    %g3, 0x0c, %g3                  ! advance out vector pointer
 601
 602 7:      retl
 603          nop