src/gallium/drivers/nouveau/codegen/target_lib_nvc0.asm

   1 //
   2 // DIV U32
   3 //
   4 // UNR recurrence (q = a / b):
   5 // look for z such that 2^32 - b <= b * z < 2^32
   6 // then q - 1 <= (a * z) / 2^32 <= q
   7 //
   8 // INPUT:   $r0: dividend, $r1: divisor
   9 // OUTPUT:  $r0: result, $r1: modulus
  10 // CLOBBER: $r2 - $r3, $p0 - $p1
  11 // SIZE:    22 / 14 * 8 bytes
  12 //
  13 bfind u32 $r2 $r1
  14 xor b32 $r2 $r2 0x1f
  15 mov b32 $r3 0x1
  16 shl b32 $r2 $r3 clamp $r2
  17 cvt u32 $r1 neg u32 $r1
  18 mul $r3 u32 $r1 u32 $r2
  19 add $r2 (mul high u32 $r2 u32 $r3) $r2
  20 mul $r3 u32 $r1 u32 $r2
  21 add $r2 (mul high u32 $r2 u32 $r3) $r2
  22 mul $r3 u32 $r1 u32 $r2
  23 add $r2 (mul high u32 $r2 u32 $r3) $r2
  24 mul $r3 u32 $r1 u32 $r2
  25 add $r2 (mul high u32 $r2 u32 $r3) $r2
  26 mul $r3 u32 $r1 u32 $r2
  27 add $r2 (mul high u32 $r2 u32 $r3) $r2
  28 mov b32 $r3 $r0
  29 mul high $r0 u32 $r0 u32 $r2
  30 cvt u32 $r2 neg u32 $r1
  31 add $r1 (mul u32 $r1 u32 $r0) $r3
  32 set $p0 0x1 ge u32 $r1 $r2
  33 $p0 sub b32 $r1 $r1 $r2
  34 $p0 add b32 $r0 $r0 0x1
  35 $p0 set $p0 0x1 ge u32 $r1 $r2
  36 $p0 sub b32 $r1 $r1 $r2
  37 $p0 add b32 $r0 $r0 0x1
  38 ret
  39 //
  40 // DIV S32, like DIV U32 after taking ABS(inputs)
  41 //
  42 // INPUT:   $r0: dividend, $r1: divisor
  43 // OUTPUT:  $r0: result, $r1: modulus
  44 // CLOBBER: $r2 - $r3, $p0 - $p3
  45 //
  46 set $p2 0x1 lt s32 $r0 0x0
  47 set $p3 0x1 lt s32 $r1 0x0 xor $p2
  48 cvt s32 $r0 abs s32 $r0
  49 cvt s32 $r1 abs s32 $r1
  50 bfind u32 $r2 $r1
  51 xor b32 $r2 $r2 0x1f
  52 mov b32 $r3 0x1
  53 shl b32 $r2 $r3 clamp $r2
  54 cvt u32 $r1 neg u32 $r1
  55 mul $r3 u32 $r1 u32 $r2
  56 add $r2 (mul high u32 $r2 u32 $r3) $r2
  57 mul $r3 u32 $r1 u32 $r2
  58 add $r2 (mul high u32 $r2 u32 $r3) $r2
  59 mul $r3 u32 $r1 u32 $r2
  60 add $r2 (mul high u32 $r2 u32 $r3) $r2
  61 mul $r3 u32 $r1 u32 $r2
  62 add $r2 (mul high u32 $r2 u32 $r3) $r2
  63 mul $r3 u32 $r1 u32 $r2
  64 add $r2 (mul high u32 $r2 u32 $r3) $r2
  65 mov b32 $r3 $r0
  66 mul high $r0 u32 $r0 u32 $r2
  67 cvt u32 $r2 neg u32 $r1
  68 add $r1 (mul u32 $r1 u32 $r0) $r3
  69 set $p0 0x1 ge u32 $r1 $r2
  70 $p0 sub b32 $r1 $r1 $r2
  71 $p0 add b32 $r0 $r0 0x1
  72 $p0 set $p0 0x1 ge u32 $r1 $r2
  73 $p0 sub b32 $r1 $r1 $r2
  74 $p0 add b32 $r0 $r0 0x1
  75 $p3 cvt s32 $r0 neg s32 $r0
  76 $p2 cvt s32 $r1 neg s32 $r1
  77 ret
  78 //
  79 // RCP F64: Newton Raphson reciprocal(x): r_{i+1} = r_i * (2.0 - x * r_i)
  80 //
  81 // INPUT:   $r0d (x)
  82 // OUTPUT:  $r0d (rcp(x))
  83 // CLOBBER: $r2 - $r7
  84 // SIZE:    9 * 8 bytes
  85 //
  86 nop
  87 ret
  88 // RSQ F64: Newton Raphson rsqrt(x): r_{i+1} = r_i * (1.5 - 0.5 * x * r_i * r_i)
  89 //
  90 // INPUT:   $r0d (x)
  91 // OUTPUT:  $r0d (rsqrt(x))
  92 // CLOBBER: $r2 - $r7
  93 // SIZE:    14 * 8 bytes
  94 //
  95 nop
  96 ret