Re: [PATCH] wwwdocs: gcc-14: Add RISC-V changes

[Palmer Dabbelt <palmer@dabbelt.com>]

On Wed, 10 Apr 2024 00:58:00 PDT (-0700), kito.cheng@sifive.com wrote:
> ---
>  htdocs/gcc-14/changes.html | 155 ++++++++++++++++++++++++++++++++++++-
>  1 file changed, 154 insertions(+), 1 deletion(-)
>
> diff --git a/htdocs/gcc-14/changes.html b/htdocs/gcc-14/changes.html
> index 2d8968cf..6cbb2e8f 100644
> --- a/htdocs/gcc-14/changes.html
> +++ b/htdocs/gcc-14/changes.html
> @@ -739,7 +739,160 @@ __asm (".global __flmap_lock"  "\n\t"
>
>  <!-- <h3 id="s390">S/390, System z, IBM z Systems</h3> -->
>
> -<!-- <h3 id="riscv">RISC-V</h3> -->
> +<h3 id="riscv">RISC-V</h3>
> +<ul>
> +  <li>The SLP and loop vectorizer is now enabled for RISC-V when the vector

I think "are now enabled"?

> +      extension is enabled, thanks to Ju-Zhe Zhong from
> +      <a href='https://rivai-ic.com.cn/'>RiVAI</a>,
> +      Pan Li from <a href='https://www.intel.com/'>Intel</a>, and Robin Dapp
> +      from <a href='https://www.ventanamicro.com/'>Ventana Micro</a> for
> +      contributing most of the implementation!</li>
> +  <li>The <code>-mrvv-max-lmul=</code> option has been introduced for
> +      performance tuning of the loop vectorizer. The default value is
> +      <code>-mrvv-max-lmul=m1</code>, which limits the maximum LMUL to 1.
> +      The <code>-mrvv-max-lmul=dynamic</code> setting can dynamically select
> +      the maximum LMUL value based on register pressure.</li>
> +  <li>Atomic code generation has been improved and is now in conformance with
> +      the latest psABI specification, thanks to Patrick O'Neill from
> +      <a href='https://www.rivosinc.com/'>Rivos</a>.</li>
> +  <li>Support for the vector intrinsics as specified in
> +      <a href='https://github.com/riscv-non-isa/rvv-intrinsic-doc/tree/v1.0.x'>
> +      version 1.0 of the RISC-V vector intrinsic specification</a>.</li>
> +  <li>Support for the experimental vector crypto intrinsics as specified in
> +      <a href='https://github.com/riscv-non-isa/rvv-intrinsic-doc/pull/234'>
> +      RISC-V vector intrinsic specification</a>, thanks to Feng Wang et al.
> +      from <a href="https://eswincomputing.com/">ESWIN Computing</a></li>
> +  <li>Support for the T-head vector intrinsics.</li>
> +  <li>Support for the scalar bitmanip and scalar crypto  intrinsics, thanks to
> +      Liao Shihua from <a href="https://plctlab.org/">PLCT</a>.</li>
> +  <li>Support for the large code model via option <code>-mcmodel=large</code>,
> +      thanks to Kuan-Lin Chen from
> +      <a href="https://www.andestech.com/">Andes Technology</a>.</li>
> +  <li>Support for the standard vector calling convention variant, thanks to
> +      Lehua Ding from <a href='https://rivai-ic.com.cn/'>RiVAI</a>.</li>
> +  <li>Supports the <code>target</code> attribute, which allows users to compile
> +      a function with specific extensions.</li>
> +  <li><code>-march=</code> option no longer requires the architecture string
> +      to be in canonical order, with only a few constraints remaining: the
> +      architecture string must start with <code>rv[32|64][i|g|e]</code>, and
> +      must use an underscore as the separator after a multi-letter extension.
> +  </li>
> +  <li><code>-march=help</code> option has been introduced to dump all
> +      supported extensions.</li>
> +  <li>Added experimental support for the <code>-mrvv-vector-bits=zvl</code>
> +      option and the <code>riscv_rvv_vector_bits</code> attribute, which
> +      specify a fixed length for scalable vector types. This option is
> +      optimized for specific vector core implementations; however, the code
> +      generated with this option is NOT portable,

IIUC the code is just optimized for a specific vector length, not any 
specific core.  It's portable to other cores, just not portable to cores 
with different vector lengths.

So I think we can soften the language a bit there, as it's not like 
we're emitting vendor-specific code on this one.

> +      thanks to Pan Li from <a href="https://www.intel.com/">Intel</a>.
> +  </li>
> +  <li>Support for TLS descriptors has been introduced, which can be enabled by
> +      the <code>-mtls-dialect=desc</code> option. The default behavior can be
> +      configured with <code>--with-tls=[trad|desc]</code>.</li>
> +  <li>Support for the TLS descriptors, this can be enabled by
> +      <code>-mtls-dialect=desc</code> and the default behavior can be configure
> +      by <code>--with-tls=[trad|desc], thanks to Tatsuyuki Ishi from
> +      <a href="https://bluewhale.systems/">Blue Whale Systems</a></code>

Maybe should call out that this will require the next glibc release to 
function correctly?

> +  </li>
> +  <li>Support for the following standard extensions has been added:
> +    <ul>
> +      <li>Vector crypto extensions:
> +	<ul>
> +	  <li>Zvbb</li>
> +	  <li>Zvkb</li>
> +	  <li>Zvbc</li>
> +	  <li>Zvkg</li>
> +	  <li>Zvkned</li>
> +	  <li>Zvkhna</li>
> +	  <li>Zvkhnb</li>
> +	  <li>Zvksed</li>
> +	  <li>Zvksh</li>
> +	  <li>Zvkn</li>
> +	  <li>Zvknc</li>
> +	  <li>Zvkng</li>
> +	  <li>Zvks</li>
> +	  <li>Zvksc</li>
> +	  <li>Zvksg</li>
> +	  <li>Zvkt</li>
> +	</ul>
> +      </li>
> +      <li>Code size reduction extensions:
> +	<ul>
> +	  <li>Zca</li>
> +	  <li>Zcb</li>
> +	  <li>Zce</li>
> +	  <li>Zcf</li>
> +	  <li>Zcd</li>
> +	  <li>Zcmp</li>
> +	  <li>Zcmt</li>
> +	</ul>
> +      </li>
> +      <li>Zicond</li>
> +      <li>Zfa</li>
> +      <li>Ztso</li>
> +      <li>Zvfbfmin</li>
> +      <li>Zvfhmin</li>
> +      <li>Zvfh</li>
> +      <li>Za64rs</li>
> +      <li>Za128rs</li>
> +      <li>Ziccif</li>
> +      <li>Ziccrse</li>
> +      <li>Ziccamoa</li>
> +      <li>Zicclsm</li>
> +      <li>Zic64b</li>
> +      <li>Smaia</li>
> +      <li>Smepmp</li>
> +      <li>Smstateen</li>
> +      <li>Ssaia</li>
> +      <li>Sscofpmf</li>
> +      <li>Ssstateen</li>
> +      <li>Sstc</li>
> +      <li>Svinval</li>
> +      <li>Svnapot</li>
> +      <li>Svpbmt</li>
> +    </ul>
> +  </li>
> +  <li>Support for the following vendor extensions has been added:
> +    <ul>
> +      <li>T-Head:
> +	<ul>
> +	  <li>XTheadVector</li>
> +	</ul>
> +      </li>
> +      <li>CORE-V:
> +	<ul>
> +	  <li>XCVmac</li>
> +	  <li>XCValu</li>
> +	  <li>XCVelw</li>
> +	  <li>XCVsimd</li>
> +	  <li>XCVbi</li>
> +	</ul>
> +      </li>
> +      <li>Ventana Micro:
> +	<ul>
> +	  <li>XVentanaCondops</li>
> +	</ul>
> +      </li>
> +    </ul>
> +  </li>
> +  <li>The following new CPUs are supported through the <code>-mcpu</code>
> +      option (GCC identifiers in parentheses).
> +    <ul>
> +      <li>SiFive's X280 (<code>sifive-x280</code>).</li>
> +      <li>SiFive's P450 (<code>sifive-p450</code>).</li>
> +      <li>SiFive's P670 (<code>sifive-p670</code>).</li>
> +    </ul>
> +  </li>
> +  <li>The following new CPUs are supported through the <code>-mtune</code>
> +      option (GCC identifiers in parentheses).
> +    <ul>
> +      <li>Generic out-of-order core (<code>generic-ooo</code>).</li>
> +      <li>SiFive's P400 series (<code>sifive-p400-series</code>).</li>
> +      <li>SiFive's P600 series (<code>sifive-p600-series</code>).</li>
> +      <li>XiangShan's Nanhu microarchitecture (<code>xiangshan-nanhu</code>).</li>
> +    </ul>
> +  </li>
> +</ul>
>
>  <!-- <h3 id="rx">RX</h3> -->

Thanks for doing this.  This all pretty minor wording stuff, so

Reviewed-by: Palmer Dabbelt <palmer@rivosinc.com>
Acked-by: Palmer Dabbelt <palmer@rivosinc.com>

Maybe next year we'll remember to ask submitters for these ;)