docs/SimpleV Grouping Proposal.md

   1 # Proposal for Element-Grouping in SimpleV on RV64GC
   2
   3 #### Notes
   4
   5 i*N* is used to denote a sign-agnostic *N*-bit integer,
   6 similarly f*N* is a *N*-bit floating-point number. *VL* is used to denote the current vector length.
   7
   8 For the unused elements in an integer register, the used element closest to the MSB is sign-extended on write and the unused elements are ignored on read.
   9 The unused elements in a floating-point register are treated as-if they are set to all ones on write and are ignored on read, matching the existing standard for storing smaller FP values in larger registers.
  10
  11 For grouped modes, *VL* denotes the number of groups, not the number of elements.
  12
  13 ### Group Size 1
  14
  15 | Mode       | # Regs | Regs/Group | Elms/Reg | Packed/Non-packed SIMD   |
  16 |------------|--------|------------|----------|--------------------------|
  17 | i8x1x*VL*  | 1x*VL* | 1          | 1        | i64x1x*VL* (Non-packed)  |
  18 | i16x1x*VL* | 1x*VL* | 1          | 1        | i64x1x*VL* (Non-packed)  |
  19 | i32x1x*VL* | 1x*VL* | 1          | 1        | i64x1x*VL* (Non-packed)  |
  20 | i64x1x*VL* | 1x*VL* | 1          | 1        | i64x1x*VL* (Non-packed)  |
  21 | f16x1x*VL* | 1x*VL* | 1          | 1        | f16x1x*VL* (Non-packed)  |
  22 | f32x1x*VL* | 1x*VL* | 1          | 1        | f32x1x*VL* (Non-packed)  |
  23 | f64x1x*VL* | 1x*VL* | 1          | 1        | f64x1x*VL* (Non-packed)  |
  24
  25 ### Group Size 2
  26
  27 | Mode       | # Regs | Regs/Group | Elms/Reg | Packed/Non-packed SIMD   |
  28 |------------|--------|------------|----------|--------------------------|
  29 | i8x2x*VL*  | 1x*VL* | 1          | 2        | Not supported            |
  30 | i16x2x*VL* | 1x*VL* | 1          | 2        | Not supported            |
  31 | i32x2x*VL* | 1x*VL* | 1          | 2        | i32x2x*VL* (Packed)      |
  32 | i64x2x*VL* | 2x*VL* | 2          | 1        | i64x2x*VL* (Non-packed)* |
  33 | f16x2x*VL* | 1x*VL* | 1          | 2        | Not supported            |
  34 | f32x2x*VL* | 1x*VL* | 1          | 2        | f32x2x*VL* (Packed)      |
  35 | f64x2x*VL* | 2x*VL* | 2          | 1        | f64x2x*VL* (Non-packed)* |
  36
  37 \* Not supported unless *VL* is changed
  38
  39 ### Group Size 3
  40
  41 | Mode       | # Regs | Regs/Group | Elms/Reg | Packed/Non-packed SIMD   |
  42 |------------|--------|------------|----------|--------------------------|
  43 | i8x3x*VL*  | 1x*VL* | 1          | 3        | Not supported            |
  44 | i16x3x*VL* | 1x*VL* | 1          | 3        | Not supported            |
  45 | i32x3x*VL* | 2x*VL* | 2          | 2        | Not supported            |
  46 | i64x3x*VL* | 3x*VL* | 3          | 1        | i64x3x*VL* (Non-packed)* |
  47 | f16x3x*VL* | 1x*VL* | 1          | 3        | Not supported            |
  48 | f32x3x*VL* | 2x*VL* | 2          | 2        | Not supported            |
  49 | f64x3x*VL* | 3x*VL* | 3          | 1        | f64x3x*VL* (Non-packed)* |
  50
  51 \* Not supported unless *VL* is changed
  52
  53 ### Group Size 4
  54
  55 | Mode       | # Regs | Regs/Group | Elms/Reg | Packed/Non-packed SIMD   |
  56 |------------|--------|------------|----------|--------------------------|
  57 | i8x4x*VL*  | 1x*VL* | 1          | 4        | Not supported            |
  58 | i16x4x*VL* | 1x*VL* | 1          | 4        | i16x4x*VL* (Packed)      |
  59 | i32x4x*VL* | 2x*VL* | 2          | 2        | i32x4x*VL* (Packed)*     |
  60 | i64x4x*VL* | 4x*VL* | 4          | 1        | i64x4x*VL* (Non-packed)* |
  61 | f16x4x*VL* | 1x*VL* | 1          | 4        | f16x4x*VL* (Packed)      |
  62 | f32x4x*VL* | 2x*VL* | 2          | 2        | f32x4x*VL* (Packed)*     |
  63 | f64x4x*VL* | 4x*VL* | 4          | 1        | f64x4x*VL* (Non-packed)* |
  64
  65 \* Not supported unless *VL* is changed
  66
  67 ...
  68
  69 ### Group Size 7
  70
  71 | Mode       | # Regs | Regs/Group | Elms/Reg | Packed/Non-packed SIMD   |
  72 |------------|--------|------------|----------|--------------------------|
  73 | i8x7x*VL*  | 1x*VL* | 1          | 7        | Not supported            |
  74 | i16x7x*VL* | 2x*VL* | 2          | 4        | Not supported            |
  75 | i32x7x*VL* | 4x*VL* | 4          | 2        | Not supported            |
  76 | i64x7x*VL* | 7x*VL* | 7          | 1        | i64x7x*VL* (Non-packed)* |
  77 | f16x7x*VL* | 2x*VL* | 2          | 4        | Not supported            |
  78 | f32x7x*VL* | 4x*VL* | 4          | 2        | Not supported            |
  79 | f64x7x*VL* | 7x*VL* | 7          | 1        | f64x7x*VL* (Non-packed)* |
  80
  81 \* Not supported unless *VL* is changed
  82
  83 ### Group Size 8
  84
  85 | Mode       | # Regs | Regs/Group | Elms/Reg | Packed/Non-packed SIMD   |
  86 |------------|--------|------------|----------|--------------------------|
  87 | i8x8x*VL*  | 1x*VL* | 1          | 8        | i8x8x*VL*  (Packed)      |
  88 | i16x8x*VL* | 2x*VL* | 2          | 4        | i16x8x*VL* (Packed)*     |
  89 | i32x8x*VL* | 4x*VL* | 4          | 2        | i32x8x*VL* (Packed)*     |
  90 | i64x8x*VL* | 8x*VL* | 8          | 1        | i64x8x*VL* (Non-packed)* |
  91 | f16x8x*VL* | 2x*VL* | 2          | 4        | f16x8x*VL* (Packed)*     |
  92 | f32x8x*VL* | 4x*VL* | 4          | 2        | f32x8x*VL* (Packed)*     |
  93 | f64x8x*VL* | 8x*VL* | 8          | 1        | f64x8x*VL* (Non-packed)* |
  94
  95 \* Not supported unless *VL* is changed