mirror of
https://gitlab.freedesktop.org/mesa/mesa.git
synced 2026-05-19 02:48:07 +02:00
1f754b7aae
This allows it to be built independently of the gallium driver. Acked-by: Karol Herbst <kherbst@redhat.com> Reviewed-by: Emma Anholt <emma@anholt.net> Reviewed-by: Yusuf Khan<yusisamerican@gmail.com> Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16763>
107 lines
2.7 KiB
NASM
107 lines
2.7 KiB
NASM
.section #gf100_builtin_code
|
|
// DIV U32
|
|
//
|
|
// UNR recurrence (q = a / b):
|
|
// look for z such that 2^32 - b <= b * z < 2^32
|
|
// then q - 1 <= (a * z) / 2^32 <= q
|
|
//
|
|
// INPUT: $r0: dividend, $r1: divisor
|
|
// OUTPUT: $r0: result, $r1: modulus
|
|
// CLOBBER: $r2 - $r3, $p0 - $p1
|
|
// SIZE: 22 / 14 * 8 bytes
|
|
//
|
|
gf100_div_u32:
|
|
bfind u32 $r2 $r1
|
|
xor b32 $r2 $r2 0x1f
|
|
mov b32 $r3 0x1
|
|
shl b32 $r2 $r3 clamp $r2
|
|
cvt u32 $r1 neg u32 $r1
|
|
mul $r3 u32 $r1 u32 $r2
|
|
add $r2 (mul high u32 $r2 u32 $r3) $r2
|
|
mul $r3 u32 $r1 u32 $r2
|
|
add $r2 (mul high u32 $r2 u32 $r3) $r2
|
|
mul $r3 u32 $r1 u32 $r2
|
|
add $r2 (mul high u32 $r2 u32 $r3) $r2
|
|
mul $r3 u32 $r1 u32 $r2
|
|
add $r2 (mul high u32 $r2 u32 $r3) $r2
|
|
mul $r3 u32 $r1 u32 $r2
|
|
add $r2 (mul high u32 $r2 u32 $r3) $r2
|
|
mov b32 $r3 $r0
|
|
mul high $r0 u32 $r0 u32 $r2
|
|
cvt u32 $r2 neg u32 $r1
|
|
add $r1 (mul u32 $r1 u32 $r0) $r3
|
|
set $p0 0x1 ge u32 $r1 $r2
|
|
$p0 sub b32 $r1 $r1 $r2
|
|
$p0 add b32 $r0 $r0 0x1
|
|
$p0 set $p0 0x1 ge u32 $r1 $r2
|
|
$p0 sub b32 $r1 $r1 $r2
|
|
$p0 add b32 $r0 $r0 0x1
|
|
ret
|
|
|
|
// DIV S32, like DIV U32 after taking ABS(inputs)
|
|
//
|
|
// INPUT: $r0: dividend, $r1: divisor
|
|
// OUTPUT: $r0: result, $r1: modulus
|
|
// CLOBBER: $r2 - $r3, $p0 - $p3
|
|
//
|
|
gf100_div_s32:
|
|
set $p2 0x1 lt s32 $r0 0x0
|
|
set $p3 0x1 lt s32 $r1 0x0 xor $p2
|
|
cvt s32 $r0 abs s32 $r0
|
|
cvt s32 $r1 abs s32 $r1
|
|
bfind u32 $r2 $r1
|
|
xor b32 $r2 $r2 0x1f
|
|
mov b32 $r3 0x1
|
|
shl b32 $r2 $r3 clamp $r2
|
|
cvt u32 $r1 neg u32 $r1
|
|
mul $r3 u32 $r1 u32 $r2
|
|
add $r2 (mul high u32 $r2 u32 $r3) $r2
|
|
mul $r3 u32 $r1 u32 $r2
|
|
add $r2 (mul high u32 $r2 u32 $r3) $r2
|
|
mul $r3 u32 $r1 u32 $r2
|
|
add $r2 (mul high u32 $r2 u32 $r3) $r2
|
|
mul $r3 u32 $r1 u32 $r2
|
|
add $r2 (mul high u32 $r2 u32 $r3) $r2
|
|
mul $r3 u32 $r1 u32 $r2
|
|
add $r2 (mul high u32 $r2 u32 $r3) $r2
|
|
mov b32 $r3 $r0
|
|
mul high $r0 u32 $r0 u32 $r2
|
|
cvt u32 $r2 neg u32 $r1
|
|
add $r1 (mul u32 $r1 u32 $r0) $r3
|
|
set $p0 0x1 ge u32 $r1 $r2
|
|
$p0 sub b32 $r1 $r1 $r2
|
|
$p0 add b32 $r0 $r0 0x1
|
|
$p0 set $p0 0x1 ge u32 $r1 $r2
|
|
$p0 sub b32 $r1 $r1 $r2
|
|
$p0 add b32 $r0 $r0 0x1
|
|
$p3 cvt s32 $r0 neg s32 $r0
|
|
$p2 cvt s32 $r1 neg s32 $r1
|
|
ret
|
|
|
|
// RCP F64: Newton Raphson reciprocal(x): r_{i+1} = r_i * (2.0 - x * r_i)
|
|
//
|
|
// INPUT: $r0d (x)
|
|
// OUTPUT: $r0d (rcp(x))
|
|
// CLOBBER: $r2 - $r7
|
|
// SIZE: 9 * 8 bytes
|
|
//
|
|
gf100_rcp_f64:
|
|
nop
|
|
ret
|
|
|
|
// RSQ F64: Newton Raphson rsqrt(x): r_{i+1} = r_i * (1.5 - 0.5 * x * r_i * r_i)
|
|
//
|
|
// INPUT: $r0d (x)
|
|
// OUTPUT: $r0d (rsqrt(x))
|
|
// CLOBBER: $r2 - $r7
|
|
// SIZE: 14 * 8 bytes
|
|
//
|
|
gf100_rsq_f64:
|
|
nop
|
|
ret
|
|
|
|
.section #gf100_builtin_offsets
|
|
.b64 #gf100_div_u32
|
|
.b64 #gf100_div_s32
|
|
.b64 #gf100_rcp_f64
|
|
.b64 #gf100_rsq_f64
|