mirror of
https://github.com/pound-emu/ballistic.git
synced 2026-01-31 01:15:21 +01:00
185 lines
11 KiB
XML
185 lines
11 KiB
XML
<?xml version="1.0" encoding="utf-8"?>
|
|
<?xml-stylesheet type="text/xsl" encoding="UTF-8" href="iform.xsl" version="1.0"?>
|
|
<!DOCTYPE instructionsection PUBLIC "-//ARM//DTD instructionsection //EN" "iform-p.dtd">
|
|
<!-- Copyright (c) 2010-2022 Arm Limited or its affiliates. All rights reserved. -->
|
|
<!-- This document is Non-Confidential. This document may only be used and distributed in accordance with the terms of the agreement entered into by Arm and the party that Arm delivered this document to. -->
|
|
|
|
<instructionsection id="bfvdot_za_zzi" title="BFVDOT" type="instruction">
|
|
<docvars>
|
|
<docvar key="instr-class" value="mortlach2" />
|
|
<docvar key="isa" value="A64" />
|
|
<docvar key="mnemonic" value="BFVDOT" />
|
|
</docvars>
|
|
<heading>BFVDOT</heading>
|
|
<desc>
|
|
<brief>Multi-vector BFloat16 floating-point vertical dot-product by indexed element</brief>
|
|
<description>
|
|
<para>The instruction operates on two ZA single-vector groups.</para>
|
|
<para>The instruction computes the vertical dot product of the corresponding BF16 elements held in the two first source vectors with pair of BF16 values held in the indexed 32-bit element of the second source vector. The single-precision dot product results are destructively added to the corresponding single-precision elements of the two ZA single-vector groups.</para>
|
|
<para>The BF16 pairs within the second source vector are specified using an immediate index which selects the same BF16 pair position within each 128-bit vector segment. The element index range is from 0 to 3.</para>
|
|
<para>The vector numbers forming the single-vector group within each half of the ZA array are selected by the sum of the vector select register and immediate offset, modulo half the number of ZA array vectors.</para>
|
|
<para>The <arm-defined-word>vector group</arm-defined-word> symbol <syntax>VGx2</syntax> indicates that the ZA operand consists of two ZA single-vector groups. The <arm-defined-word>vector group</arm-defined-word> symbol is preferred for disassembly, but optional in assembler source code.</para>
|
|
<para>This instruction follows SME2 ZA-targeting BFloat16 numerical behaviors.</para>
|
|
<para>This instruction is unpredicated.</para>
|
|
</description>
|
|
<status>Green</status>
|
|
<predicated>False</predicated>
|
|
<sm_policy>SM_1_only</sm_policy>
|
|
</desc>
|
|
<alias_list howmany="0"></alias_list>
|
|
<classes>
|
|
<iclass name="SME2" oneof="1" id="iclass_mortlach2" no_encodings="1" isa="A64">
|
|
<docvars>
|
|
<docvar key="instr-class" value="mortlach2" />
|
|
<docvar key="isa" value="A64" />
|
|
<docvar key="mnemonic" value="BFVDOT" />
|
|
</docvars>
|
|
<iclassintro count="1"></iclassintro>
|
|
<arch_variants>
|
|
<arch_variant name="FEAT_SME2" feature="FEAT_SME2" />
|
|
</arch_variants>
|
|
<regdiagram form="32" psname="BFVDOT-ZA.ZZi-2xi">
|
|
<box hibit="31" width="12" settings="12">
|
|
<c>1</c>
|
|
<c>1</c>
|
|
<c>0</c>
|
|
<c>0</c>
|
|
<c>0</c>
|
|
<c>0</c>
|
|
<c>0</c>
|
|
<c>1</c>
|
|
<c>0</c>
|
|
<c>1</c>
|
|
<c>0</c>
|
|
<c>1</c>
|
|
</box>
|
|
<box hibit="19" width="4" name="Zm" usename="1">
|
|
<c colspan="4"></c>
|
|
</box>
|
|
<box hibit="15" settings="1">
|
|
<c>0</c>
|
|
</box>
|
|
<box hibit="14" width="2" name="Rv" usename="1">
|
|
<c colspan="2"></c>
|
|
</box>
|
|
<box hibit="12" name="op" settings="1">
|
|
<c>0</c>
|
|
</box>
|
|
<box hibit="11" width="2" name="i2" usename="1">
|
|
<c colspan="2"></c>
|
|
</box>
|
|
<box hibit="9" width="4" name="Zn" usename="1">
|
|
<c colspan="4"></c>
|
|
</box>
|
|
<box hibit="5" width="3" name="opc2" settings="3">
|
|
<c>0</c>
|
|
<c>1</c>
|
|
<c>1</c>
|
|
</box>
|
|
<box hibit="2" width="3" name="off3" usename="1">
|
|
<c colspan="3"></c>
|
|
</box>
|
|
</regdiagram>
|
|
<encoding name="bfvdot_za_zzi_2xi" oneofinclass="1" oneof="1" label="">
|
|
<docvars>
|
|
<docvar key="instr-class" value="mortlach2" />
|
|
<docvar key="isa" value="A64" />
|
|
<docvar key="mnemonic" value="BFVDOT" />
|
|
</docvars>
|
|
<asmtemplate><text>BFVDOT ZA.S[</text><a link="sa_wv" hover="32-bit vector select register W8-W11 (field "Rv")"><Wv></a><text>, </text><a link="sa_offs" hover="Vector select offset [0-7] (field "off3")"><offs></a><a>{, VGx2}</a><text>], </text><text>{</text><text> </text><a link="sa_zn1" hover="First scalable vector register of a multi-vector sequence (field Zn)"><Zn1></a><text>.H-</text><a link="sa_zn2" hover="Second scalable vector register of a multi-vector sequence (field Zn)"><Zn2></a><text>.H </text><text>}</text><text>, </text><a link="sa_zm" hover="Second source scalable vector register Z0-Z15 (field "Zm")"><Zm></a><text>.H[</text><a link="sa_index" hover="Element index [0-3] (field "i2")"><index></a><text>]</text></asmtemplate>
|
|
</encoding>
|
|
<ps_section howmany="1">
|
|
<ps name="BFVDOT-ZA.ZZi-2xi" mylink="BFVDOT-ZA.ZZi-2xi" enclabels="" sections="1" secttype="noheading">
|
|
<pstext mayhavelinks="1" section="Decode" rep_section="decode">if !<a link="impl-aarch64.HaveSME2.0" file="shared_pseudocode.xml" hover="function: boolean HaveSME2()">HaveSME2</a>() then UNDEFINED;
|
|
integer v = <a link="impl-shared.UInt.1" file="shared_pseudocode.xml" hover="function: integer UInt(bits(N) x)">UInt</a>('010':Rv);
|
|
integer n = <a link="impl-shared.UInt.1" file="shared_pseudocode.xml" hover="function: integer UInt(bits(N) x)">UInt</a>(Zn:'0');
|
|
integer m = <a link="impl-shared.UInt.1" file="shared_pseudocode.xml" hover="function: integer UInt(bits(N) x)">UInt</a>('0':Zm);
|
|
integer offset = <a link="impl-shared.UInt.1" file="shared_pseudocode.xml" hover="function: integer UInt(bits(N) x)">UInt</a>(off3);
|
|
integer index = <a link="impl-shared.UInt.1" file="shared_pseudocode.xml" hover="function: integer UInt(bits(N) x)">UInt</a>(i2);</pstext>
|
|
</ps>
|
|
</ps_section>
|
|
</iclass>
|
|
</classes>
|
|
<explanations scope="all">
|
|
<explanation enclist="bfvdot_za_zzi_2xi" symboldefcount="1">
|
|
<symbol link="sa_wv"><Wv></symbol>
|
|
<account encodedin="Rv">
|
|
<intro>
|
|
<para>Is the 32-bit name of the vector select register W8-W11, encoded in the "Rv" field.</para>
|
|
</intro>
|
|
</account>
|
|
</explanation>
|
|
<explanation enclist="bfvdot_za_zzi_2xi" symboldefcount="1">
|
|
<symbol link="sa_offs"><offs></symbol>
|
|
<account encodedin="off3">
|
|
<intro>
|
|
<para>Is the vector select offset, in the range 0 to 7, encoded in the "off3" field.</para>
|
|
</intro>
|
|
</account>
|
|
</explanation>
|
|
<explanation enclist="bfvdot_za_zzi_2xi" symboldefcount="1">
|
|
<symbol link="sa_zn1"><Zn1></symbol>
|
|
<account encodedin="Zn">
|
|
<intro>
|
|
<para>Is the name of the first scalable vector register of a multi-vector sequence, encoded as "Zn" times 2.</para>
|
|
</intro>
|
|
</account>
|
|
</explanation>
|
|
<explanation enclist="bfvdot_za_zzi_2xi" symboldefcount="1">
|
|
<symbol link="sa_zn2"><Zn2></symbol>
|
|
<account encodedin="Zn">
|
|
<intro>
|
|
<para>Is the name of the second scalable vector register of a multi-vector sequence, encoded as "Zn" times 2 plus 1.</para>
|
|
</intro>
|
|
</account>
|
|
</explanation>
|
|
<explanation enclist="bfvdot_za_zzi_2xi" symboldefcount="1">
|
|
<symbol link="sa_zm"><Zm></symbol>
|
|
<account encodedin="Zm">
|
|
<intro>
|
|
<para>Is the name of the second source scalable vector register Z0-Z15, encoded in the "Zm" field.</para>
|
|
</intro>
|
|
</account>
|
|
</explanation>
|
|
<explanation enclist="bfvdot_za_zzi_2xi" symboldefcount="1">
|
|
<symbol link="sa_index"><index></symbol>
|
|
<account encodedin="i2">
|
|
<intro>
|
|
<para>Is the element index, in the range 0 to 3, encoded in the "i2" field.</para>
|
|
</intro>
|
|
</account>
|
|
</explanation>
|
|
</explanations>
|
|
<ps_section howmany="1">
|
|
<ps name="BFVDOT-ZA.ZZi-2xi" mylink="execute" enclabels="" sections="1" secttype="Operation">
|
|
<pstext mayhavelinks="1" section="Execute" rep_section="execute"><a link="impl-aarch64.CheckStreamingSVEAndZAEnabled.0" file="shared_pseudocode.xml" hover="function: CheckStreamingSVEAndZAEnabled()">CheckStreamingSVEAndZAEnabled</a>();
|
|
constant integer VL = <a link="impl-aarch64.CurrentVL.read.none" file="shared_pseudocode.xml" hover="accessor: integer CurrentVL">CurrentVL</a>;
|
|
constant integer elements = VL DIV 32;
|
|
integer vectors = VL DIV 8;
|
|
integer vstride = vectors DIV 2;
|
|
integer eltspersegment = 128 DIV 32;
|
|
bits(32) vbase = <a link="impl-aarch64.X.read.2" file="shared_pseudocode.xml" hover="accessor: bits(width) X[integer n, integer width]">X</a>[v, 32];
|
|
integer vec = (<a link="impl-shared.UInt.1" file="shared_pseudocode.xml" hover="function: integer UInt(bits(N) x)">UInt</a>(vbase) + offset) MOD vstride;
|
|
bits(VL) result;
|
|
|
|
for r = 0 to 1
|
|
bits(VL) operand1a = <a link="impl-aarch64.Z.read.2" file="shared_pseudocode.xml" hover="accessor: bits(width) Z[integer n, integer width]">Z</a>[n, VL];
|
|
bits(VL) operand1b = <a link="impl-aarch64.Z.read.2" file="shared_pseudocode.xml" hover="accessor: bits(width) Z[integer n, integer width]">Z</a>[n+1, VL];
|
|
bits(VL) operand2 = <a link="impl-aarch64.Z.read.2" file="shared_pseudocode.xml" hover="accessor: bits(width) Z[integer n, integer width]">Z</a>[m, VL];
|
|
bits(VL) operand3 = <a link="impl-aarch64.ZAvector.read.2" file="shared_pseudocode.xml" hover="accessor: bits(width) ZAvector[integer index, integer width]">ZAvector</a>[vec, VL];
|
|
for e = 0 to elements-1
|
|
integer segmentbase = e - (e MOD eltspersegment);
|
|
integer s = segmentbase + index;
|
|
bits(16) elt1_a = <a link="impl-shared.Elem.read.3" file="shared_pseudocode.xml" hover="accessor: bits(size) Elem[bits(N) vector, integer e, integer size]">Elem</a>[operand1a, 2 * e + r, 16];
|
|
bits(16) elt1_b = <a link="impl-shared.Elem.read.3" file="shared_pseudocode.xml" hover="accessor: bits(size) Elem[bits(N) vector, integer e, integer size]">Elem</a>[operand1b, 2 * e + r, 16];
|
|
bits(16) elt2_a = <a link="impl-shared.Elem.read.3" file="shared_pseudocode.xml" hover="accessor: bits(size) Elem[bits(N) vector, integer e, integer size]">Elem</a>[operand2, 2 * s + 0, 16];
|
|
bits(16) elt2_b = <a link="impl-shared.Elem.read.3" file="shared_pseudocode.xml" hover="accessor: bits(size) Elem[bits(N) vector, integer e, integer size]">Elem</a>[operand2, 2 * s + 1, 16];
|
|
bits(32) sum = <a link="impl-shared.Elem.read.3" file="shared_pseudocode.xml" hover="accessor: bits(size) Elem[bits(N) vector, integer e, integer size]">Elem</a>[operand3, e, 32];
|
|
sum = <a link="impl-shared.BFDotAdd.6" file="shared_pseudocode.xml" hover="function: bits(32) BFDotAdd(bits(32) addend, bits(16) op1_a, bits(16) op1_b, bits(16) op2_a, bits(16) op2_b, FPCRType fpcr_in)">BFDotAdd</a>(sum, elt1_a, elt1_b, elt2_a, elt2_b, FPCR[]);
|
|
<a link="impl-shared.Elem.write.3" file="shared_pseudocode.xml" hover="accessor: Elem[bits(N) &vector, integer e, integer size] = bits(size) value">Elem</a>[result, e, 32] = sum;
|
|
<a link="impl-aarch64.ZAvector.write.2" file="shared_pseudocode.xml" hover="accessor: ZAvector[integer index, integer width] = bits(width) value">ZAvector</a>[vec, VL] = result;
|
|
vec = vec + vstride;</pstext>
|
|
</ps>
|
|
</ps_section>
|
|
</instructionsection>
|