src/mesa/sparc/norm.S

   1
   2 #include "sparc_matrix.h"
   3
   4         .register %g2, #scratch
   5         .register %g3, #scratch
   6
   7         .text
   8
   9 #ifdef __arch64__
  10 #define STACK_VAR_OFF   (2047 + (8 * 16))
  11 #else
  12 #define STACK_VAR_OFF   (4 * 16)
  13 #endif
  14
  15         /* Newton-Raphson approximation turns out to be slower
  16          * (and less accurate) than direct fsqrts/fdivs.
  17          */
  18 #define ONE_DOT_ZERO    0x3f800000
  19
  20         .globl  _mesa_sparc_transform_normalize_normals
  21 _mesa_sparc_transform_normalize_normals:
  22         /* o0=mat o1=scale o2=in o3=lengths o4=dest */
  23
  24         sethi   %hi(ONE_DOT_ZERO), %g2
  25         sub     %sp, 16, %sp
  26         st      %g2, [%sp + STACK_VAR_OFF+0x0]
  27         st      %o1, [%sp + STACK_VAR_OFF+0x4]
  28         ld      [%sp + STACK_VAR_OFF+0x0], %f12 ! f12 = 1.0f
  29         ld      [%sp + STACK_VAR_OFF+0x4], %f15 ! f15 = scale
  30         add     %sp, 16, %sp
  31
  32         LDPTR   [%o0 + MAT_INV], %o0            ! o0 = mat->inv
  33         LDPTR   [%o2 + V4F_START], %o5          ! o5 = 'from' in->start
  34         ld      [%o2 + V4F_COUNT], %g1          ! g1 = in->count
  35         ld      [%o2 + V4F_STRIDE], %g2         ! g2 = in->stride
  36         LDPTR   [%o4 + V4F_START], %g3          ! g3 = 'out' dest->start
  37
  38         LDMATRIX_0_1_2_4_5_6_8_9_10(%o0)
  39
  40         /* dest->count = in->count */
  41         st      %g1, [%o4 + V4F_COUNT]
  42
  43         cmp     %g1, 1
  44         bl      7f
  45          cmp    %o3, 0
  46         bne     4f
  47          clr    %o4                             ! 'i' for STRIDE_LOOP
  48
  49 1:      /* LENGTHS == NULL */
  50         ld      [%o5 + 0x00], %f0               ! ux = from[0]
  51         ld      [%o5 + 0x04], %f1               ! uy = from[1]
  52         ld      [%o5 + 0x08], %f2               ! uz = from[2]
  53         add     %o5, %g2, %o5                   ! STRIDE_F(from, stride)
  54         add     %o4, 1, %o4                     ! i++
  55
  56         /* tx (f3) = (ux * m0) + (uy * m1) + (uz * m2)
  57          * ty (f5) = (ux * m4) + (uy * m5) + (uz * m6)
  58          * tz (f7) = (ux * m8) + (uy * m9) + (uz * m10)
  59          */
  60         fmuls   %f0, M0, %f3                    ! FGM   Group
  61         fmuls   %f1, M1, %f4                    ! FGM   Group
  62         fmuls   %f0, M4, %f5                    ! FGM   Group
  63         fmuls   %f1, M5, %f6                    ! FGM   Group
  64         fmuls   %f0, M8, %f7                    ! FGM   Group   f3 available
  65         fmuls   %f1, M9, %f8                    ! FGM   Group   f4 available
  66         fadds   %f3, %f4, %f3                   ! FGA
  67         fmuls   %f2, M2, %f10                   ! FGM   Group   f5 available
  68         fmuls   %f2, M6, %f0                    ! FGM   Group   f6 available
  69         fadds   %f5, %f6, %f5                   ! FGA
  70         fmuls   %f2, M10, %f4                   ! FGM   Group   f7 available
  71         fadds   %f7, %f8, %f7                   ! FGA   Group   f8,f3 available
  72         fadds   %f3, %f10, %f3                  ! FGA   Group   f10 available
  73         fadds   %f5, %f0, %f5                   ! FGA   Group   stall f0,f5 available
  74         fadds   %f7, %f4, %f7                   ! FGA   Group   stall f4,f7 available
  75
  76         /* f3=tx, f5=ty, f7=tz */
  77
  78         /* len (f6) = (tx * tx) + (ty * ty) + (tz * tz) */
  79         fmuls   %f3, %f3, %f6                   ! FGM   Group   f3 available
  80         fmuls   %f5, %f5, %f8                   ! FGM   Group   f5 available
  81         fmuls   %f7, %f7, %f10                  ! FGM   Group   f7 available
  82         fadds   %f6, %f8, %f6                   ! FGA   Group   2cyc stall f6,f8 available
  83         fadds   %f6, %f10, %f6                  ! FGA   Group   4cyc stall f6,f10 available
  84
  85         /* scale (f6) = 1.0 / sqrt(len) */
  86         fsqrts  %f6, %f6                        ! FDIV  20 cycles
  87         fdivs   %f12, %f6, %f6                  ! FDIV  14 cycles
  88
  89         fmuls   %f3, %f6, %f3
  90         st      %f3, [%g3 + 0x00]               ! out[i][0] = tx * scale
  91         fmuls   %f5, %f6, %f5
  92         st      %f5, [%g3 + 0x04]               ! out[i][1] = ty * scale
  93         fmuls   %f7, %f6, %f7
  94         st      %f7, [%g3 + 0x08]               ! out[i][2] = tz * scale
  95
  96         cmp     %o4, %g1                        ! continue if (i < count)
  97         bl      1b
  98          add    %g3, 0x10, %g3                  ! advance out vector pointer
  99
 100         ba      7f
 101          nop
 102
 103 4:      /* LENGTHS != NULL */
 104         fmuls   M0, %f15, M0
 105         fmuls   M1, %f15, M1
 106         fmuls   M2, %f15, M2
 107         fmuls   M4, %f15, M4
 108         fmuls   M5, %f15, M5
 109         fmuls   M6, %f15, M6
 110         fmuls   M8, %f15, M8
 111         fmuls   M9, %f15, M9
 112         fmuls   M10, %f15, M10
 113
 114 5:
 115         ld      [%o5 + 0x00], %f0               ! ux = from[0]
 116         ld      [%o5 + 0x04], %f1               ! uy = from[1]
 117         ld      [%o5 + 0x08], %f2               ! uz = from[2]
 118         add     %o5, %g2, %o5                   ! STRIDE_F(from, stride)
 119         add     %o4, 1, %o4                     ! i++
 120
 121         /* tx (f3) = (ux * m0) + (uy * m1) + (uz * m2)
 122          * ty (f5) = (ux * m4) + (uy * m5) + (uz * m6)
 123          * tz (f7) = (ux * m8) + (uy * m9) + (uz * m10)
 124          */
 125         fmuls   %f0, M0, %f3                    ! FGM   Group
 126         fmuls   %f1, M1, %f4                    ! FGM   Group
 127         fmuls   %f0, M4, %f5                    ! FGM   Group
 128         fmuls   %f1, M5, %f6                    ! FGM   Group
 129         fmuls   %f0, M8, %f7                    ! FGM   Group   f3 available
 130         fmuls   %f1, M9, %f8                    ! FGM   Group   f4 available
 131         fadds   %f3, %f4, %f3                   ! FGA
 132         fmuls   %f2, M2, %f10                   ! FGM   Group   f5 available
 133         fmuls   %f2, M6, %f0                    ! FGM   Group   f6 available
 134         fadds   %f5, %f6, %f5                   ! FGA
 135         fmuls   %f2, M10, %f4                   ! FGM   Group   f7 available
 136         fadds   %f7, %f8, %f7                   ! FGA   Group   f8,f3 available
 137         fadds   %f3, %f10, %f3                  ! FGA   Group   f10 available
 138         ld      [%o3], %f13                     ! LSU
 139         fadds   %f5, %f0, %f5                   ! FGA   Group   stall f0,f5 available
 140         add     %o3, 4, %o3                     ! IEU0
 141         fadds   %f7, %f4, %f7                   ! FGA   Group   stall f4,f7 available
 142
 143         /* f3=tx, f5=ty, f7=tz, f13=lengths[i] */
 144
 145         fmuls   %f3, %f13, %f3
 146         st      %f3, [%g3 + 0x00]               ! out[i][0] = tx * len
 147         fmuls   %f5, %f13, %f5
 148         st      %f5, [%g3 + 0x04]               ! out[i][1] = ty * len
 149         fmuls   %f7, %f13, %f7
 150         st      %f7, [%g3 + 0x08]               ! out[i][2] = tz * len
 151
 152         cmp     %o4, %g1                        ! continue if (i < count)
 153         bl      5b
 154          add    %g3, 0x10, %g3                  ! advance out vector pointer
 155
 156 7:      retl
 157          nop
 158
 159         .globl  _mesa_sparc_transform_normalize_normals_no_rot
 160 _mesa_sparc_transform_normalize_normals_no_rot:
 161         /* o0=mat o1=scale o2=in o3=lengths o4=dest */
 162
 163         sethi   %hi(ONE_DOT_ZERO), %g2
 164         sub     %sp, 16, %sp
 165         st      %g2, [%sp + STACK_VAR_OFF+0x0]
 166         st      %o1, [%sp + STACK_VAR_OFF+0x4]
 167         ld      [%sp + STACK_VAR_OFF+0x0], %f12 ! f12 = 1.0f
 168         ld      [%sp + STACK_VAR_OFF+0x4], %f15 ! f15 = scale
 169         add     %sp, 16, %sp
 170
 171         LDPTR   [%o0 + MAT_INV], %o0            ! o0 = mat->inv
 172         LDPTR   [%o2 + V4F_START], %o5          ! o5 = 'from' in->start
 173         ld      [%o2 + V4F_COUNT], %g1          ! g1 = in->count
 174         ld      [%o2 + V4F_STRIDE], %g2         ! g2 = in->stride
 175         LDPTR   [%o4 + V4F_START], %g3          ! g3 = 'out' dest->start
 176
 177         LDMATRIX_0_5_10(%o0)
 178
 179         /* dest->count = in->count */
 180         st      %g1, [%o4 + V4F_COUNT]
 181
 182         cmp     %g1, 1
 183         bl      7f
 184          cmp    %o3, 0
 185         bne     4f
 186          clr    %o4                             ! 'i' for STRIDE_LOOP
 187
 188 1:      /* LENGTHS == NULL */
 189         ld      [%o5 + 0x00], %f0               ! ux = from[0]
 190         ld      [%o5 + 0x04], %f1               ! uy = from[1]
 191         ld      [%o5 + 0x08], %f2               ! uz = from[2]
 192         add     %o5, %g2, %o5                   ! STRIDE_F(from, stride)
 193         add     %o4, 1, %o4                     ! i++
 194
 195         /* tx (f3) = (ux * m0)
 196          * ty (f5) = (uy * m5)
 197          * tz (f7) = (uz * m10)
 198          */
 199         fmuls   %f0, M0, %f3                    ! FGM   Group
 200         fmuls   %f1, M5, %f5                    ! FGM   Group
 201         fmuls   %f2, M10, %f7                   ! FGM   Group
 202
 203         /* f3=tx, f5=ty, f7=tz */
 204
 205         /* len (f6) = (tx * tx) + (ty * ty) + (tz * tz) */
 206         fmuls   %f3, %f3, %f6                   ! FGM   Group   stall, f3 available
 207         fmuls   %f5, %f5, %f8                   ! FGM   Group   f5 available
 208         fmuls   %f7, %f7, %f10                  ! FGM   Group   f7 available
 209         fadds   %f6, %f8, %f6                   ! FGA   Group   2cyc stall f6,f8 available
 210         fadds   %f6, %f10, %f6                  ! FGA   Group   4cyc stall f6,f10 available
 211
 212         /* scale (f6) = 1.0 / sqrt(len) */
 213         fsqrts  %f6, %f6                        ! FDIV  20 cycles
 214         fdivs   %f12, %f6, %f6                  ! FDIV  14 cycles
 215
 216         fmuls   %f3, %f6, %f3
 217         st      %f3, [%g3 + 0x00]               ! out[i][0] = tx * scale
 218         fmuls   %f5, %f6, %f5
 219         st      %f5, [%g3 + 0x04]               ! out[i][1] = ty * scale
 220         fmuls   %f7, %f6, %f7
 221         st      %f7, [%g3 + 0x08]               ! out[i][2] = tz * scale
 222
 223         cmp     %o4, %g1                        ! continue if (i < count)
 224         bl      1b
 225          add    %g3, 0x10, %g3                  ! advance out vector pointer
 226
 227         ba      7f
 228          nop
 229
 230 4:      /* LENGTHS != NULL */
 231         fmuls   M0, %f15, M0
 232         fmuls   M5, %f15, M5
 233         fmuls   M10, %f15, M10
 234
 235 5:
 236         ld      [%o5 + 0x00], %f0               ! ux = from[0]
 237         ld      [%o5 + 0x04], %f1               ! uy = from[1]
 238         ld      [%o5 + 0x08], %f2               ! uz = from[2]
 239         add     %o5, %g2, %o5                   ! STRIDE_F(from, stride)
 240         add     %o4, 1, %o4                     ! i++
 241
 242         /* tx (f3) = (ux * m0)
 243          * ty (f5) = (uy * m5)
 244          * tz (f7) = (uz * m10)
 245          */
 246         fmuls   %f0, M0, %f3                    ! FGM   Group
 247         ld      [%o3], %f13                     ! LSU
 248         fmuls   %f1, M5, %f5                    ! FGM   Group
 249         add     %o3, 4, %o3                     ! IEU0
 250         fmuls   %f2, M10, %f7                   ! FGM   Group
 251
 252         /* f3=tx, f5=ty, f7=tz, f13=lengths[i] */
 253
 254         fmuls   %f3, %f13, %f3
 255         st      %f3, [%g3 + 0x00]               ! out[i][0] = tx * len
 256         fmuls   %f5, %f13, %f5
 257         st      %f5, [%g3 + 0x04]               ! out[i][1] = ty * len
 258         fmuls   %f7, %f13, %f7
 259         st      %f7, [%g3 + 0x08]               ! out[i][2] = tz * len
 260
 261         cmp     %o4, %g1                        ! continue if (i < count)
 262         bl      5b
 263          add    %g3, 0x10, %g3                  ! advance out vector pointer
 264
 265 7:      retl
 266          nop
 267
 268         .globl  _mesa_sparc_transform_rescale_normals_no_rot
 269 _mesa_sparc_transform_rescale_normals_no_rot:
 270         /* o0=mat o1=scale o2=in o3=lengths o4=dest */
 271         sub     %sp, 16, %sp
 272         st      %o1, [%sp + STACK_VAR_OFF+0x0]
 273         ld      [%sp + STACK_VAR_OFF+0x0], %f15 ! f15 = scale
 274         add     %sp, 16, %sp
 275
 276         LDPTR   [%o0 + MAT_INV], %o0            ! o0 = mat->inv
 277         LDPTR   [%o2 + V4F_START], %o5          ! o5 = 'from' in->start
 278         ld      [%o2 + V4F_COUNT], %g1          ! g1 = in->count
 279         ld      [%o2 + V4F_STRIDE], %g2         ! g2 = in->stride
 280         LDPTR   [%o4 + V4F_START], %g3          ! g3 = 'out' dest->start
 281
 282         LDMATRIX_0_5_10(%o0)
 283
 284         /* dest->count = in->count */
 285         st      %g1, [%o4 + V4F_COUNT]
 286
 287         cmp     %g1, 1
 288         bl      7f
 289          clr    %o4                             ! 'i' for STRIDE_LOOP
 290
 291         fmuls   M0, %f15, M0
 292         fmuls   M5, %f15, M5
 293         fmuls   M10, %f15, M10
 294
 295 1:      ld      [%o5 + 0x00], %f0               ! ux = from[0]
 296         ld      [%o5 + 0x04], %f1               ! uy = from[1]
 297         ld      [%o5 + 0x08], %f2               ! uz = from[2]
 298         add     %o5, %g2, %o5                   ! STRIDE_F(from, stride)
 299         add     %o4, 1, %o4                     ! i++
 300
 301         /* tx (f3) = (ux * m0)
 302          * ty (f5) = (uy * m5)
 303          * tz (f7) = (uz * m10)
 304          */
 305         fmuls   %f0, M0, %f3                    ! FGM   Group
 306         st      %f3, [%g3 + 0x00]               ! LSU
 307         fmuls   %f1, M5, %f5                    ! FGM   Group
 308         st      %f5, [%g3 + 0x04]               ! LSU
 309         fmuls   %f2, M10, %f7                   ! FGM   Group
 310         st      %f7, [%g3 + 0x08]               ! LSU
 311
 312         cmp     %o4, %g1                        ! continue if (i < count)
 313         bl      1b
 314          add    %g3, 0x10, %g3                  ! advance out vector pointer
 315
 316 7:      retl
 317          nop
 318
 319         .globl  _mesa_sparc_transform_rescale_normals
 320 _mesa_sparc_transform_rescale_normals:
 321         /* o0=mat o1=scale o2=in o3=lengths o4=dest */
 322         sub     %sp, 16, %sp
 323         st      %o1, [%sp + STACK_VAR_OFF+0x0]
 324         ld      [%sp + STACK_VAR_OFF+0x0], %f15 ! f15 = scale
 325         add     %sp, 16, %sp
 326
 327         LDPTR   [%o0 + MAT_INV], %o0            ! o0 = mat->inv
 328         LDPTR   [%o2 + V4F_START], %o5          ! o5 = 'from' in->start
 329         ld      [%o2 + V4F_COUNT], %g1          ! g1 = in->count
 330         ld      [%o2 + V4F_STRIDE], %g2         ! g2 = in->stride
 331         LDPTR   [%o4 + V4F_START], %g3          ! g3 = 'out' dest->start
 332
 333         LDMATRIX_0_1_2_4_5_6_8_9_10(%o0)
 334
 335         /* dest->count = in->count */
 336         st      %g1, [%o4 + V4F_COUNT]
 337
 338         cmp     %g1, 1
 339         bl      7f
 340          clr    %o4                             ! 'i' for STRIDE_LOOP
 341
 342         fmuls   M0, %f15, M0
 343         fmuls   M1, %f15, M1
 344         fmuls   M2, %f15, M2
 345         fmuls   M4, %f15, M4
 346         fmuls   M5, %f15, M5
 347         fmuls   M6, %f15, M6
 348         fmuls   M8, %f15, M8
 349         fmuls   M9, %f15, M9
 350         fmuls   M10, %f15, M10
 351
 352 1:      ld      [%o5 + 0x00], %f0               ! ux = from[0]
 353         ld      [%o5 + 0x04], %f1               ! uy = from[1]
 354         ld      [%o5 + 0x08], %f2               ! uz = from[2]
 355         add     %o5, %g2, %o5                   ! STRIDE_F(from, stride)
 356         add     %o4, 1, %o4                     ! i++
 357
 358         fmuls   %f0, M0, %f3                    ! FGM   Group
 359         fmuls   %f1, M1, %f4                    ! FGM   Group
 360         fmuls   %f0, M4, %f5                    ! FGM   Group
 361         fmuls   %f1, M5, %f6                    ! FGM   Group
 362         fmuls   %f0, M8, %f7                    ! FGM   Group   f3 available
 363         fmuls   %f1, M9, %f8                    ! FGM   Group   f4 available
 364         fadds   %f3, %f4, %f3                   ! FGA
 365         fmuls   %f2, M2, %f10                   ! FGM   Group   f5 available
 366         fmuls   %f2, M6, %f0                    ! FGM   Group   f6 available
 367         fadds   %f5, %f6, %f5                   ! FGA
 368         fmuls   %f2, M10, %f4                   ! FGM   Group   f7 available
 369         fadds   %f7, %f8, %f7                   ! FGA   Group   f8,f3 available
 370         fadds   %f3, %f10, %f3                  ! FGA   Group   f10 available
 371         st      %f3, [%g3 + 0x00]               ! LSU
 372         fadds   %f5, %f0, %f5                   ! FGA   Group   stall f0,f5 available
 373         st      %f5, [%g3 + 0x04]               ! LSU
 374         fadds   %f7, %f4, %f7                   ! FGA   Group   stall f4,f7 available
 375         st      %f7, [%g3 + 0x08]               ! LSU
 376
 377         cmp     %o4, %g1                        ! continue if (i < count)
 378         bl      1b
 379          add    %g3, 0x10, %g3                  ! advance out vector pointer
 380
 381 7:      retl
 382          nop
 383
 384         .globl  _mesa_sparc_transform_normals_no_rot
 385 _mesa_sparc_transform_normals_no_rot:
 386         /* o0=mat o1=scale o2=in o3=lengths o4=dest */
 387         LDPTR   [%o0 + MAT_INV], %o0            ! o0 = mat->inv
 388         LDPTR   [%o2 + V4F_START], %o5          ! o5 = 'from' in->start
 389         ld      [%o2 + V4F_COUNT], %g1          ! g1 = in->count
 390         ld      [%o2 + V4F_STRIDE], %g2         ! g2 = in->stride
 391         LDPTR   [%o4 + V4F_START], %g3          ! g3 = 'out' dest->start
 392
 393         LDMATRIX_0_5_10(%o0)
 394
 395         /* dest->count = in->count */
 396         st      %g1, [%o4 + V4F_COUNT]
 397
 398         cmp     %g1, 1
 399         bl      7f
 400          clr    %o4                             ! 'i' for STRIDE_LOOP
 401
 402 1:      ld      [%o5 + 0x00], %f0               ! ux = from[0]
 403         ld      [%o5 + 0x04], %f1               ! uy = from[1]
 404         ld      [%o5 + 0x08], %f2               ! uz = from[2]
 405         add     %o5, %g2, %o5                   ! STRIDE_F(from, stride)
 406         add     %o4, 1, %o4                     ! i++
 407
 408         /* tx (f3) = (ux * m0)
 409          * ty (f5) = (uy * m5)
 410          * tz (f7) = (uz * m10)
 411          */
 412         fmuls   %f0, M0, %f3                    ! FGM   Group
 413         st      %f3, [%g3 + 0x00]               ! LSU
 414         fmuls   %f1, M5, %f5                    ! FGM   Group
 415         st      %f5, [%g3 + 0x04]               ! LSU
 416         fmuls   %f2, M10, %f7                   ! FGM   Group
 417         st      %f7, [%g3 + 0x08]               ! LSU
 418
 419         cmp     %o4, %g1                        ! continue if (i < count)
 420         bl      1b
 421          add    %g3, 0x10, %g3                  ! advance out vector pointer
 422
 423 7:      retl
 424          nop
 425
 426         .globl  _mesa_sparc_transform_normals
 427 _mesa_sparc_transform_normals:
 428         /* o0=mat o1=scale o2=in o3=lengths o4=dest */
 429         LDPTR   [%o0 + MAT_INV], %o0            ! o0 = mat->inv
 430         LDPTR   [%o2 + V4F_START], %o5          ! o5 = 'from' in->start
 431         ld      [%o2 + V4F_COUNT], %g1          ! g1 = in->count
 432         ld      [%o2 + V4F_STRIDE], %g2         ! g2 = in->stride
 433         LDPTR   [%o4 + V4F_START], %g3          ! g3 = 'out' dest->start
 434
 435         LDMATRIX_0_1_2_4_5_6_8_9_10(%o0)
 436
 437         /* dest->count = in->count */
 438         st      %g1, [%o4 + V4F_COUNT]
 439
 440         cmp     %g1, 1
 441         bl      7f
 442          clr    %o4                             ! 'i' for STRIDE_LOOP
 443
 444 1:      ld      [%o5 + 0x00], %f0               ! ux = from[0]
 445         ld      [%o5 + 0x04], %f1               ! uy = from[1]
 446         ld      [%o5 + 0x08], %f2               ! uz = from[2]
 447         add     %o5, %g2, %o5                   ! STRIDE_F(from, stride)
 448         add     %o4, 1, %o4                     ! i++
 449
 450         fmuls   %f0, M0, %f3                    ! FGM   Group
 451         fmuls   %f1, M1, %f4                    ! FGM   Group
 452         fmuls   %f0, M4, %f5                    ! FGM   Group
 453         fmuls   %f1, M5, %f6                    ! FGM   Group
 454         fmuls   %f0, M8, %f7                    ! FGM   Group   f3 available
 455         fmuls   %f1, M9, %f8                    ! FGM   Group   f4 available
 456         fadds   %f3, %f4, %f3                   ! FGA
 457         fmuls   %f2, M2, %f10                   ! FGM   Group   f5 available
 458         fmuls   %f2, M6, %f0                    ! FGM   Group   f6 available
 459         fadds   %f5, %f6, %f5                   ! FGA
 460         fmuls   %f2, M10, %f4                   ! FGM   Group   f7 available
 461         fadds   %f7, %f8, %f7                   ! FGA   Group   f8,f3 available
 462         fadds   %f3, %f10, %f3                  ! FGA   Group   f10 available
 463         st      %f3, [%g3 + 0x00]               ! LSU
 464         fadds   %f5, %f0, %f5                   ! FGA   Group   stall f0,f5 available
 465         st      %f5, [%g3 + 0x04]               ! LSU
 466         fadds   %f7, %f4, %f7                   ! FGA   Group   stall f4,f7 available
 467         st      %f7, [%g3 + 0x08]               ! LSU
 468
 469         cmp     %o4, %g1                        ! continue if (i < count)
 470         bl      1b
 471          add    %g3, 0x10, %g3                  ! advance out vector pointer
 472
 473 7:      retl
 474          nop
 475
 476         .globl  _mesa_sparc_normalize_normals
 477 _mesa_sparc_normalize_normals:
 478         /* o0=mat o1=scale o2=in o3=lengths o4=dest */
 479
 480         sethi   %hi(ONE_DOT_ZERO), %g2
 481         sub     %sp, 16, %sp
 482         st      %g2, [%sp + STACK_VAR_OFF+0x0]
 483         ld      [%sp + STACK_VAR_OFF+0x0], %f12 ! f12 = 1.0f
 484         add     %sp, 16, %sp
 485
 486         LDPTR   [%o2 + V4F_START], %o5          ! o5 = 'from' in->start
 487         ld      [%o2 + V4F_COUNT], %g1          ! g1 = in->count
 488         ld      [%o2 + V4F_STRIDE], %g2         ! g2 = in->stride
 489         LDPTR   [%o4 + V4F_START], %g3          ! g3 = 'out' dest->start
 490
 491         /* dest->count = in->count */
 492         st      %g1, [%o4 + V4F_COUNT]
 493
 494         cmp     %g1, 1
 495         bl      7f
 496          cmp    %o3, 0
 497         bne     4f
 498          clr    %o4                             ! 'i' for STRIDE_LOOP
 499
 500 1:      /* LENGTHS == NULL */
 501         ld      [%o5 + 0x00], %f3               ! ux = from[0]
 502         ld      [%o5 + 0x04], %f5               ! uy = from[1]
 503         ld      [%o5 + 0x08], %f7               ! uz = from[2]
 504         add     %o5, %g2, %o5                   ! STRIDE_F(from, stride)
 505         add     %o4, 1, %o4                     ! i++
 506
 507         /* f3=tx, f5=ty, f7=tz */
 508
 509         /* len (f6) = (tx * tx) + (ty * ty) + (tz * tz) */
 510         fmuls   %f3, %f3, %f6                   ! FGM   Group   f3 available
 511         fmuls   %f5, %f5, %f8                   ! FGM   Group   f5 available
 512         fmuls   %f7, %f7, %f10                  ! FGM   Group   f7 available
 513         fadds   %f6, %f8, %f6                   ! FGA   Group   2cyc stall f6,f8 available
 514         fadds   %f6, %f10, %f6                  ! FGA   Group   4cyc stall f6,f10 available
 515
 516         /* scale (f6) = 1.0 / sqrt(len) */
 517         fsqrts  %f6, %f6                        ! FDIV  20 cycles
 518         fdivs   %f12, %f6, %f6                  ! FDIV  14 cycles
 519
 520         fmuls   %f3, %f6, %f3
 521         st      %f3, [%g3 + 0x00]               ! out[i][0] = tx * scale
 522         fmuls   %f5, %f6, %f5
 523         st      %f5, [%g3 + 0x04]               ! out[i][1] = ty * scale
 524         fmuls   %f7, %f6, %f7
 525         st      %f7, [%g3 + 0x08]               ! out[i][2] = tz * scale
 526
 527         cmp     %o4, %g1                        ! continue if (i < count)
 528         bl      1b
 529          add    %g3, 0x10, %g3                  ! advance out vector pointer
 530
 531         ba      7f
 532          nop
 533
 534 4:      /* LENGTHS != NULL */
 535
 536 5:
 537         ld      [%o5 + 0x00], %f3               ! ux = from[0]
 538         ld      [%o5 + 0x04], %f5               ! uy = from[1]
 539         ld      [%o5 + 0x08], %f7               ! uz = from[2]
 540         add     %o5, %g2, %o5                   ! STRIDE_F(from, stride)
 541         add     %o4, 1, %o4                     ! i++
 542
 543         ld      [%o3], %f13                     ! LSU
 544         add     %o3, 4, %o3                     ! IEU0
 545
 546         /* f3=tx, f5=ty, f7=tz, f13=lengths[i] */
 547
 548         fmuls   %f3, %f13, %f3
 549         st      %f3, [%g3 + 0x00]               ! out[i][0] = tx * len
 550         fmuls   %f5, %f13, %f5
 551         st      %f5, [%g3 + 0x04]               ! out[i][1] = ty * len
 552         fmuls   %f7, %f13, %f7
 553         st      %f7, [%g3 + 0x08]               ! out[i][2] = tz * len
 554
 555         cmp     %o4, %g1                        ! continue if (i < count)
 556         bl      5b
 557          add    %g3, 0x10, %g3                  ! advance out vector pointer
 558
 559 7:      retl
 560          nop
 561
 562         .globl  _mesa_sparc_rescale_normals
 563 _mesa_sparc_rescale_normals:
 564         /* o0=mat o1=scale o2=in o3=lengths o4=dest */
 565
 566         sethi   %hi(ONE_DOT_ZERO), %g2
 567         sub     %sp, 16, %sp
 568         st      %o1, [%sp + STACK_VAR_OFF+0x0]
 569         ld      [%sp + STACK_VAR_OFF+0x0], %f15 ! f15 = scale
 570         add     %sp, 16, %sp
 571
 572         LDPTR   [%o2 + V4F_START], %o5          ! o5 = 'from' in->start
 573         ld      [%o2 + V4F_COUNT], %g1          ! g1 = in->count
 574         ld      [%o2 + V4F_STRIDE], %g2         ! g2 = in->stride
 575         LDPTR   [%o4 + V4F_START], %g3          ! g3 = 'out' dest->start
 576
 577         /* dest->count = in->count */
 578         st      %g1, [%o4 + V4F_COUNT]
 579
 580         cmp     %g1, 1
 581         bl      7f
 582          clr    %o4                             ! 'i' for STRIDE_LOOP
 583
 584 1:
 585         ld      [%o5 + 0x00], %f3               ! ux = from[0]
 586         ld      [%o5 + 0x04], %f5               ! uy = from[1]
 587         ld      [%o5 + 0x08], %f7               ! uz = from[2]
 588         add     %o5, %g2, %o5                   ! STRIDE_F(from, stride)
 589         add     %o4, 1, %o4                     ! i++
 590
 591         /* f3=tx, f5=ty, f7=tz */
 592
 593         fmuls   %f3, %f15, %f3
 594         st      %f3, [%g3 + 0x00]               ! out[i][0] = tx * scale
 595         fmuls   %f5, %f15, %f5
 596         st      %f5, [%g3 + 0x04]               ! out[i][1] = ty * scale
 597         fmuls   %f7, %f15, %f7
 598         st      %f7, [%g3 + 0x08]               ! out[i][2] = tz * scale
 599
 600         cmp     %o4, %g1                        ! continue if (i < count)
 601         bl      1b
 602          add    %g3, 0x10, %g3                  ! advance out vector pointer
 603
 604 7:      retl
 605          nop