1
0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2024-12-07 11:13:41 +02:00
FFmpeg/libavutil/riscv/asm.S
Rémi Denis-Courmont ee1526c05f lavu/riscv: add assembler macros for adjusting vector LMUL
vtype_vli computes the VTYPE value with the optimal LMUL for a given
element width, tail and mask policies and a run-time vector length.

vtype_ivli does the same, but with the compile-time constant vector
length.

vwtypei and vntypei can be used to widen or narrow a VTYPE value for
use in mixed-width vector-optimised functions.
2024-05-19 18:37:33 +03:00

241 lines
6.5 KiB
ArmAsm

/*
* Copyright © 2022 Rémi Denis-Courmont.
* Loosely based on earlier work copyrighted by Måns Rullgård, 2008.
*
* This file is part of FFmpeg.
*
* FFmpeg is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* FFmpeg is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with FFmpeg; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#if defined (__riscv_float_abi_soft)
#define NOHWF
#define NOHWD
#define HWF #
#define HWD #
#elif defined (__riscv_float_abi_single)
#define NOHWF #
#define NOHWD
#define HWF
#define HWD #
#else
#define NOHWF #
#define NOHWD #
#define HWF
#define HWD
#endif
.macro func sym, ext1=, ext2=
.text
.align 2
.option push
.ifnb \ext1
.option arch, +\ext1
.ifnb \ext2
.option arch, +\ext2
.endif
.endif
.global \sym
.hidden \sym
.type \sym, %function
\sym:
.macro endfunc
.size \sym, . - \sym
.option pop
.previous
.purgem endfunc
.endm
.endm
.macro const sym, align=3, relocate=0
.if \relocate
.pushsection .data.rel.ro
.else
.pushsection .rodata
.endif
.align \align
\sym:
.macro endconst
.size \sym, . - \sym
.popsection
.purgem endconst
.endm
.endm
#if !defined (__riscv_zba)
/* SH{1,2,3}ADD definitions for pre-Zba assemblers */
.macro shnadd n, rd, rs1, rs2
.insn r OP, 2 * \n, 16, \rd, \rs1, \rs2
.endm
.macro sh1add rd, rs1, rs2
shnadd 1, \rd, \rs1, \rs2
.endm
.macro sh2add rd, rs1, rs2
shnadd 2, \rd, \rs1, \rs2
.endm
.macro sh3add rd, rs1, rs2
shnadd 3, \rd, \rs1, \rs2
.endm
#endif
#if defined (__riscv_v_elen)
# define RV_V_ELEN __riscv_v_elen
#else
/* Run-time detection of the V extension implies ELEN >= 64. */
# define RV_V_ELEN 64
#endif
#if RV_V_ELEN == 32
# define VSEW_MAX 2
#else
# define VSEW_MAX 3
#endif
.macro parse_vtype ew, tp, mp
.ifc \ew,e8
.equ vsew, 0
.else
.ifc \ew,e16
.equ vsew, 1
.else
.ifc \ew,e32
.equ vsew, 2
.else
.ifc \ew,e64
.equ vsew, 3
.else
.error "Unknown element width \ew"
.endif
.endif
.endif
.endif
.ifc \tp,tu
.equ tp, 0
.else
.ifc \tp,ta
.equ tp, 1
.else
.error "Unknown tail policy \tp"
.endif
.endif
.ifc \mp,mu
.equ mp, 0
.else
.ifc \mp,ma
.equ mp, 1
.else
.error "Unknown mask policy \mp"
.endif
.endif
.endm
/**
* Gets the vector type with the smallest suitable LMUL value.
* @param[out] rd vector type destination register
* @param vl vector length constant
* @param ew element width: e8, e16, e32 or e64
* @param tp tail policy: tu or ta
* @param mp mask policty: mu or ma
*/
.macro vtype_ivli rd, avl, ew, tp=tu, mp=mu
.if \avl <= 1
.equ log2vl, 0
.elseif \avl <= 2
.equ log2vl, 1
.elseif \avl <= 4
.equ log2vl, 2
.elseif \avl <= 8
.equ log2vl, 3
.elseif \avl <= 16
.equ log2vl, 4
.elseif \avl <= 32
.equ log2vl, 5
.elseif \avl <= 64
.equ log2vl, 6
.elseif \avl <= 128
.equ log2vl, 7
.else
.error "Vector length \avl out of range"
.endif
parse_vtype \ew, \tp, \mp
csrr \rd, vlenb
clz \rd, \rd
addi \rd, \rd, log2vl + 1 + VSEW_MAX - __riscv_xlen
max \rd, \rd, zero // VLMUL must be >= VSEW - VSEW_MAX
.if vsew < VSEW_MAX
addi \rd, \rd, vsew - VSEW_MAX
andi \rd, \rd, 7
.endif
ori \rd, \rd, (vsew << 3) | (tp << 6) | (mp << 7)
.endm
/**
* Gets the vector type with the smallest suitable LMUL value.
* @param[out] rd vector type destination register
* @param rs vector length source register
* @param[out] tmp temporary register to be clobbered
* @param ew element width: e8, e16, e32 or e64
* @param tp tail policy: tu or ta
* @param mp mask policty: mu or ma
*/
.macro vtype_vli rd, rs, tmp, ew, tp=tu, mp=mu
parse_vtype \ew, \tp, \mp
/*
* The difference between the CLZ's notionally equals the VLMUL value
* for 4-bit elements. But we want the value for SEW_MAX-bit elements.
*/
slli \tmp, \rs, 1 + VSEW_MAX
csrr \rd, vlenb
addi \tmp, \tmp, -1
clz \rd, \rd
clz \tmp, \tmp
sub \rd, \rd, \tmp
max \rd, \rd, zero // VLMUL must be >= VSEW - VSEW_MAX
.if vsew < VSEW_MAX
addi \rd, \rd, vsew - VSEW_MAX
andi \rd, \rd, 7
.endif
ori \rd, \rd, (vsew << 3) | (tp << 6) | (mp << 7)
.endm
/**
* Widens a vector type.
* @param[out] rd widened vector type destination register
* @param rs vector type source register
* @param n number of times to widen (once by default)
*/
.macro vwtypei rd, rs, n=1
xori \rd, \rs, 4
addi \rd, \rd, (\n) * 011
xori \rd, \rd, 4
.endm
/**
* Narrows a vector type.
* @param[out] rd narrowed vector type destination register
* @param rs vector type source register
* @param n number of times to narrow (once by default)
*/
.macro vntypei rd, rs, n=1
vwtypei \rd, \rs, -(\n)
.endm