Syntaxe des instructions gfx906

Introduction

Ce document décrit la syntaxe des instructions spécifiques à gfx906.

Pour une description des autres instructions gfx906, voir Syntaxe des instructions GFX9 essentielles.

Notation

La notation utilisée dans ce document est expliquée ici.

Aperçu

Un aperçu de la syntaxe générale et des autres fonctionnalités des instructions AMDGPU est disponible ici.

Instructions

VOP2

INSTRUCTION                    DST       SRC0      SRC1       MODIFIERS
————————————————————————————————————————————————————————————————————————————————————————————————————
v_fmac_f32                     vdst,     src0,     vsrc1
v_fmac_f32_dpp                 vdst,     vsrc0:m,  vsrc1:m    dpp_ctrl row_mask bank_mask bound_ctrl
v_xnor_b32                     vdst,     src0,     vsrc1
v_xnor_b32_dpp                 vdst,     vsrc0,    vsrc1      dpp_ctrl row_mask bank_mask bound_ctrl
v_xnor_b32_sdwa                vdst,     src0:m,   src1:m     dst_sel dst_unused src0_sel src1_sel

VOP3

INSTRUCTION                    DST       SRC0      SRC1           MODIFIERS
————————————————————————————————————————————————————————————————————————————
v_fmac_f32_e64                 vdst,     src0:m,   src1:m         clamp omod
v_xnor_b32_e64                 vdst,     src0,     src1

VOP3P

INSTRUCTION            DST      SRC0        SRC1        SRC2           MODIFIERS
—————————————————————————————————————————————————————————————————————————————————————————————————
v_dot2_f32_f16         vdst,    src0:f16x2, src1:f16x2, src2:f32       neg_lo neg_hi clamp
v_dot2_i32_i16         vdst,    src0:i16x2, src1:i16x2, src2:i32       clamp
v_dot2_u32_u16         vdst,    src0:u16x2, src1:u16x2, src2:u32       clamp
v_dot4_i32_i8          vdst,    src0:i8x4,  src1:i8x4,  src2:i32       clamp
v_dot4_u32_u8          vdst,    src0:u8x4,  src1:u8x4,  src2:u32       clamp
v_dot8_i32_i4          vdst,    src0:i4x8,  src1:i4x8,  src2:i32       clamp
v_dot8_u32_u4          vdst,    src0:u4x8,  src1:u4x8,  src2:u32       clamp
v_fma_mix_f32          vdst,    src0:m:fx,  src1:m:fx,  src2:m:fx      m_op_sel m_op_sel_hi clamp
v_fma_mixhi_f16        vdst,    src0:m:fx,  src1:m:fx,  src2:m:fx      m_op_sel m_op_sel_hi clamp
v_fma_mixlo_f16        vdst,    src0:m:fx,  src1:m:fx,  src2:m:fx      m_op_sel m_op_sel_hi clamp