Add P3319R0 iota_v<simd> paper

ChangeLog:

	* P3319_iota/Makefile: New file.
	* P3319_iota/changelog.tex: New file.
	* P3319_iota/main.tex: New file.
	* P3319_iota/strawpolls.tex: New file.

P3319_iota/Makefile

@@ -0,0 +1,133 @@
+NAME := $(shell grep 'newcommand.wgDocumentNumber' main.tex|cut -f2 -d'{'|cut -f1 -d'}')
+DRAFT := $(NAME)_draft
+TEXINPUTS := ..:.:
+PDFLATEX := TEXINPUTS=$(TEXINPUTS) flock . lualatex --shell-escape --halt-on-error --file-line-error --interaction nonstopmode
+BIBER := biber --input-directory .. --output-directory .
+MAKEINDEX := makeindex
+SHELL := /bin/bash
+EXTRA_DEPS := $(wildcard *.sty ../*.sty) ._dummy.tex
+CREATEMD5 := md5sum `find . -regex '.*\.\(aux\|bbl\|toc\|tex\|ind\)'` > ._aux-md5sums
+CHECKMD5  := md5sum --quiet -c ._aux-md5sums
+UPLOADNAME := D$(shell echo $(NAME)|cut -c2-5).pdf
+
+all: final
+
+._dummy.tex:
+	touch $@
+
+help:
+	@echo "all"
+	@echo "devel"
+	@echo "draft-loop"
+	@echo "final-loop"
+	@echo "final-figure<number>-loop"
+	@echo "draft"
+	@echo "final"
+	@echo "clean"
+	@echo "autocommit"
+
+TEX_INPUTS   := $(shell egrep '^[^%]*\\in(put|clude){.*?}' main.tex|sed 's@^.*{\(.*\)}.*$$@\1.tex@')
+TEX_INPUTS2  := $(shell test -n "$(TEX_INPUTS)" && egrep '^[^%]*\\in(put|clude){.*?}' $(TEX_INPUTS) ._dummy.tex|sed -e 's@^.*{\([^.]*\)}.*$$@\1.tex@' -e 's@^.*{\(.*\)}.*$$@\1@') $(TEX_INPUTS)
+TEX_INPUTS   := $(shell test -n "$(TEX_INPUTS2)" && egrep '^[^%]*\\in(put|clude){.*?}' $(TEX_INPUTS2) ._dummy.tex|sed -e 's@^.*{\([^.]*\)}.*$$@\1.tex@' -e 's@^.*{\(.*\)}.*$$@\1@') $(TEX_INPUTS2) $(TEX_INPUTS) main.tex
+TEX_LISTINGS := $(shell test -n "$(TEX_INPUTS)" && awk -F '{|}' -- '/lstinputlisting\[[^\]]*$$/ { getline; while($$0 !~ /\] *{.*\..*}$$/) getline; sub(/^.*\]/, ""); print } /lstinputlisting\[.*\] *{.*}/ { print }' $(TEX_INPUTS) |sed 's=.*\(\[[^]]*\]\)\?{\([^}]*\)}.*=\2='| sort -u)
+TEX_PLOTDATA := $(shell grep '\\addplot.* file' $(TEX_INPUTS)|sed 's=^.*\\addplot.* file *{\(.*\)}.*$$=\1=')
+GRAPHICS := $(shell grep 'includegraphics' main.tex $(TEX_INPUTS)|grep -v newcommand|sed -e 's=.*\\includegraphics[^{]*{\([^}]*\)}.*=\1=' |sort -u|while read f; do for d in $(subst :, ,$(TEXINPUTS)); do test -f $$d/$$f && echo $$d/$$f && break; done; done)
+ALL_DEPS := $(TEX_INPUTS) $(TEX_LISTINGS) $(TEX_PLOTDATA) $(GRAPHICS) $(EXTRA_DEPS)
+
+devel: final-loop
+
+draft: $(DRAFT).pdf
+final: $(NAME).pdf
+mobile: $(NAME)_mobile.pdf
+
+upload: $(NAME).pdf
+	@echo -n "Uploading $< ."
+	@while ! timeout 120s scp -B $< lxpool.gsi.de:web-docs/$(UPLOADNAME); do echo -n .; done
+	@echo .
+
+loop-internal:
+	@screenTitle() { case "$$TERM" in screen*|tmux*) printf '\ek%s\e\\' "$$*";; *) printf '\e]1;%s:q\a' "$$*";; esac }; \
+	while true; do \
+	  while true; do \
+	    screenTitle 'LaTeX waiting: $(loopinternaltarget)'; \
+	    if ! $(MAKE) -q $(loopinternaltarget).pdf>/dev/null; then \
+	      screenTitle 'LaTeX build: $(loopinternaltarget)'; \
+	      ls -lt $(ALL_DEPS) | head -n1 > ._stamp; \
+	      nice ionice -c idle $(MAKE) $(loopinternaltarget).pdf || break; \
+	    fi; \
+	    sleep 1s; \
+	  done; \
+	  screenTitle 'failed'; \
+	  kdialog --passivepopup "LaTeX ($(NAME)) failed" 1; \
+	  echo -e '\a'; \
+	  ls -lt $(ALL_DEPS) $(MORE_DEPS) | head -n1 > ._stamp.new; \
+	  while diff -q ._stamp ._stamp.new; do \
+	    sleep 1s; \
+	    ls -lt $(ALL_DEPS) $(MORE_DEPS) | head -n1 > ._stamp.new; \
+	  done; \
+	done
+
+draft-loop:
+	$(MAKE) loopinternaltarget=$(DRAFT) MORE_DEPS=../draft.tex loop-internal
+
+final-loop:
+	$(MAKE) loopinternaltarget=$(NAME) MORE_DEPS=../final.tex loop-internal
+
+%-loop:
+	$(MAKE) loopinternaltarget=$(NAME)_$* MORE_DEPS=../$*.tex loop-internal
+
+final-figure%-loop:
+	@while true; do $(MAKE) -q final-figure$*.pdf || $(MAKE) final-figure$*.pdf && sleep 1s; done
+
+clean:
+	rm -f $(NAME).pdf $(DRAFT).pdf *.aux *.bbl *.blg *.brf *.lof *.log *.lol *.lot *.out *.toc */*.aux *.auxlock ._aux-md5sums *.bcf *.run.xml ._stamp ._stamp.new
+
+ALL_DEPS=$(TEX_INPUTS) $(TEX_LISTINGS) $(EXTRA_DEPS) $(TEX_PLOTDATA)
+
+$(DRAFT).pdf: ../draft.tex $(ALL_DEPS)
+	$(PDFLATEX) ../draft
+	@test -s draft.pdf && mv draft.pdf $@
+	$(BIBER) draft || true
+
+$(NAME).pdf: ../final.tex $(ALL_DEPS)
+	@$(CREATEMD5)
+	$(PDFLATEX) ../final
+	@test -s final.pdf && mv final.pdf $@
+	$(BIBER) final || true
+	@while ! $(CHECKMD5); do \
+		$(CREATEMD5); \
+		$(PDFLATEX) ../final; \
+		test -s final.pdf && mv final.pdf $@; \
+	done
+	test "$(USER)" = "mkretz" && timeout 20s scp -B $@ lxpool.gsi.de:web-docs/$(UPLOADNAME) &
+
+$(NAME)_%.pdf: ../%.tex $(ALL_DEPS)
+	@$(CREATEMD5)
+	$(PDFLATEX) ../$*
+	@test -s $*.pdf && mv $*.pdf $@
+	test -f $*.bcf && $(BIBER) $* || true
+	test -f $*.ind && $(MAKEINDEX) $* || true
+	while ! $(CHECKMD5); do \
+		$(CREATEMD5); \
+		$(PDFLATEX) ../$*; \
+		test -s $*.pdf && mv $*.pdf $@; \
+		test -f $*.ind && $(MAKEINDEX) $*; \
+	done
+
+final-figure%.pdf: final.tex $(ALL_DEPS)
+	$(PDFLATEX) --halt-on-error --interaction=nonstopmode --jobname "final-figure$*" "\def\tikzexternalrealjob{final}\input{final}"
+
+autocommit:
+	@screenTitle() { echo $$TERM|grep -q screen && printf '\ek%s\e\\' "$$*"; }; \
+	screenTitle 'autocommit' \
+	while true; do \
+	  git commit -am "auto: `git status --porcelain|grep -v '^??'|cut -c4-|paste -s -d' '`" && \
+	  screenTitle 'committed' && \
+	  git push && \
+	  screenTitle 'committed & pushed'; \
+	  sleep 5s; \
+	  screenTitle 'autocommit'; \
+	  sleep 55s; \
+	done
+
+.PHONY: clean

P3319_iota/changelog.tex

@@ -0,0 +1,5 @@
+\section{Changelog}
+(placeholder)
+%\begin{revision}
+%  \todo Everything
+%\end{revision}

P3319_iota/main.tex

@@ -0,0 +1,143 @@
+\newcommand\wgTitle{Add an iota object for simd (and more)}
+\newcommand\wgName{Matthias Kretz <[email protected]>}
+\newcommand\wgDocumentNumber{P3319R0}
+\newcommand\wgGroup{LEWG}
+\newcommand\wgTarget{\CC{}26}
+%\newcommand\wgAcknowledgements{Daniel Towner and Ruslan Arutyunyan contributed to this paper via discussions / reviews. Thanks also to Jeff Garland for reviewing.}
+
+\usepackage{mymacros}
+\usepackage{wg21}
+\setcounter{tocdepth}{2} % show sections and subsections in TOC
+\hypersetup{bookmarksdepth=5}
+\usepackage{changelog}
+\usepackage{underscore}
+\usepackage{multirow}
+
+\addbibresource{extra.bib}
+
+\newcommand\simd[1][]{\type{ba\-sic\_simd#1}\xspace}
+\newcommand\simdT{\type{ba\-sic\_simd\MayBreak<\MayBreak{}T>}\xspace}
+\newcommand\valuetype{\type{val\-ue\_type}\xspace}
+\newcommand\referencetype{\type{ref\-er\-ence}\xspace}
+\newcommand\mask[1][]{\type{ba\-sic\_simd\_mask#1}\xspace}
+\newcommand\maskT{\type{ba\-sic\_simd\_mask\MayBreak<\MayBreak{}T>}\xspace}
+\newcommand\wglink[1]{\href{https://wg21.link/#1}{#1}}
+
+\newcommand\nativeabi{\UNSP{native-abi}}
+\newcommand\deducet{\UNSP{deduce-t}}
+\newcommand\simdsizev{\UNSP{simd-size-v}}
+\newcommand\simdsizetype{\UNSP{simd-size-type}}
+\newcommand\simdselect{\UNSP{simd-select-impl}}
+\newcommand\maskelementsize{\UNSP{mask-element-size}}
+\newcommand\integerfrom{\UNSP{integer-from}}
+\newcommand\constexprwrapperlike{\UNSP{constexpr-wrapper-like}}
+
+\renewcommand{\lst}[1]{Listing~\ref{#1}}
+\renewcommand{\sect}[1]{Section~\ref{#1}}
+\renewcommand{\ttref}[1]{Tony~Table~\ref{#1}}
+\renewcommand{\tabref}[1]{Table~\ref{#1}}
+
+\begin{document}
+\selectlanguage{american}
+\begin{wgTitlepage}
+  There is one important constant in SIMD programming: \code{{0, 1, 2, 3,
+  ...}}. In the standard library we have an algorithm called \code{iota} that
+  can initialize a range with such values. For \code{simd} we want to have
+  simple to spell constants that scale with the SIMD width. This paper proposes
+  a simple facility that can be generalized.
+\end{wgTitlepage}
+
+\pagestyle{scrheadings}
+
+\input{changelog}
+\input{strawpolls}
+
+\section{Motivation}
+The 90\% use case for simd generator constructors is a simd with values 0, 1,
+2, 3, \ldots{} potentially with scaling and offset applied.
+However, often it would be more easier and more readable to use an “iota”
+\code{simd} object instead.
+
+\begingroup
+  \smaller[1]
+\begin{tabular}{p{.45\textwidth}|p{.45\textwidth}}
+  generator ctor & iota \\
+  \hline
+  \begin{lstlisting}
+std::simd<int> a([](int i) { return i; };
+
+std::simd<int> b([](int i) { return 2 + 3 * i; };
+  \end{lstlisting}
+  &
+  \begin{lstlisting}
+auto a = std::iota_v<std::simd<int>>;
+
+auto b = 2 + 3 * std::iota_v<std::simd<int>>;
+  \end{lstlisting}
+\end{tabular}
+\endgroup
+
+The minimal definition I propose for \simd can look like this:
+\medskip\begin{lstlisting}[style=Vc]
+  template <class T>
+    inline constexpr T
+    iota_v;
+
+  template <class T>
+    requires(std::is_arithmetic_v<T>)
+    inline constexpr T
+    iota_v<T> = T();
+
+  template <detail::simd_type T>
+    inline constexpr T
+    iota_v<T> = T([](auto i) { return static_cast<typename T::value_type>(i); });
+\end{lstlisting}
+
+
+\section{Generalization}
+If we define a (\code{constexpr}) variable template \std\code{iota_v<T>} where
+\code{T} must be a \simd type, we're simply filling a sequence of values.
+We can create such an object for any type with static extent.
+This is especially interesting for the degenerate case in SIMD-generic
+programming, where \code{T} could \eg be an \code{int}.
+A \std\code{iota_v<int>} is nothing other than an object \code{int} with value
+\code{0}.
+We can easily generalize to \code{iota_v<std::array<T, N>>} and
+\code{iota_v<T[N]>}.
+And the next step then is to allow any type that
+\begin{itemize}
+  \item has a static extent,
+  \item has a \code{value_type} member type,
+  \item can be list-initialized with \code{N} numbers of type
+    \code{value_type}, where \code{N} equals the static extent of the type, and
+  \item where \code{value_type() + 1} is an constant expression and convertible to \code{value_type}.
+\end{itemize}
+
+Consequently you could write
+\medskip\begin{lstlisting}[style=Vc]
+auto x = std::iota_v<float[5]>;
+auto y = std::iota_v<std::array<my_fixed_point, 8>>;
+// ...
+\end{lstlisting}
+
+\section{Relation to list-initialization of \code{simd}}
+If we add a constructor to \simd that enables list-initialization, then many
+users might use that in place of a generator constructor.
+This leads to code that doesn't scale with the vector width anymore.
+Therefore we should provide a simple facility that works better and is more
+portable.
+
+\section{Proposed polls}
+
+\wgPoll{We want an iota facility for \simd}{&&&&}
+
+\wgPoll{The iota facility should be generalized to scalars}{&&&&}
+
+\wgPoll{The iota facility should be generalized to any sequence of static extent}{&&&&}
+
+\section{Wording}\label{sec:wording}
+
+TBD after deciding on the preferred solution.
+
+\end{document}
+% vim: sw=2 sts=2 ai et tw=0

P3319_iota/strawpolls.tex

@@ -0,0 +1,2 @@
+\section{Straw Polls}
+(placeholder)