zhangyiss/mlx
1
0
Fork 0
mirror of https://github.com/ml-explore/mlx.git synced 2025-09-18 01:50:16 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

docs update

Browse Source
This commit is contained in:
Awni Hannun
2024-08-23 12:14:53 -07:00
committed by CircleCI Docs
parent d8f7b8cda6
commit 27e07888b0
697 changed files with 15867 additions and 8594 deletions
Download Patch File Download Diff File Expand all files Collapse all files

394
docs/build/html/reduce__row_8h.html vendored
View File

@@ -85,34 +85,52 @@ $(function() {
<table class="memberdecls">
<tr class="heading"><td colspan="2"><h2 class="groupheader"><a id="func-members" name="func-members"></a>
Functions</h2></td></tr>
<tr class="memitem:aea2b1ba640fffb71c73550ec747b3148" id="r_aea2b1ba640fffb71c73550ec747b3148"><td class="memTemplParams" colspan="2">template&lt;typename T , typename U , typename Op &gt; </td></tr>
<tr class="memitem:aea2b1ba640fffb71c73550ec747b3148"><td class="memTemplItemLeft" align="right" valign="top">void&#160;</td><td class="memTemplItemRight" valign="bottom"><a class="el" href="#aea2b1ba640fffb71c73550ec747b3148">row_reduce_general_small</a> (const device T *in, device U *out, const constant size_t &amp;reduction_size, const constant size_t &amp;out_size, const constant size_t &amp;non_row_reductions, const constant int *shape, const constant size_t *strides, const constant int &amp;ndim, uint lid)</td></tr>
<tr class="separator:aea2b1ba640fffb71c73550ec747b3148"><td class="memSeparator" colspan="2">&#160;</td></tr>
<tr class="memitem:ad77d444bcfa9e6afcc1ffc1dcb919f1c" id="r_ad77d444bcfa9e6afcc1ffc1dcb919f1c"><td class="memTemplParams" colspan="2">template&lt;typename T , typename U , typename Op &gt; </td></tr>
<tr class="memitem:ad77d444bcfa9e6afcc1ffc1dcb919f1c"><td class="memTemplItemLeft" align="right" valign="top">void&#160;</td><td class="memTemplItemRight" valign="bottom"><a class="el" href="#ad77d444bcfa9e6afcc1ffc1dcb919f1c">row_reduce_general_med</a> (const device T *in, device U *out, const constant size_t &amp;reduction_size, const constant size_t &amp;out_size, const constant size_t &amp;non_row_reductions, const constant int *shape, const constant size_t *strides, const constant int &amp;ndim, uint tid, uint simd_lane_id, uint simd_per_group, uint simd_group_id)</td></tr>
<tr class="separator:ad77d444bcfa9e6afcc1ffc1dcb919f1c"><td class="memSeparator" colspan="2">&#160;</td></tr>
<tr class="memitem:a5df83c60214206d584be1587c9b723a5" id="r_a5df83c60214206d584be1587c9b723a5"><td class="memTemplParams" colspan="2">template&lt;typename T , typename U , typename Op , int N_READS = REDUCE_N_READS&gt; </td></tr>
<tr class="memitem:a5df83c60214206d584be1587c9b723a5"><td class="memTemplItemLeft" align="right" valign="top">METAL_FUNC U&#160;</td><td class="memTemplItemRight" valign="bottom"><a class="el" href="#a5df83c60214206d584be1587c9b723a5">per_thread_row_reduce</a> (const device T *in, const constant size_t &amp;reduction_size, const constant size_t &amp;out_size, const constant int *shape, const constant size_t *strides, const constant int &amp;ndim, uint lsize_x, uint lid_x, uint2 tid)</td></tr>
<tr class="separator:a5df83c60214206d584be1587c9b723a5"><td class="memSeparator" colspan="2">&#160;</td></tr>
<tr class="memitem:a917b01e3de3372d276f8dd28f0e9b8f1" id="r_a917b01e3de3372d276f8dd28f0e9b8f1"><td class="memTemplParams" colspan="2">template&lt;typename T , typename U , typename Op , int N_READS = REDUCE_N_READS&gt; </td></tr>
<tr class="memitem:a917b01e3de3372d276f8dd28f0e9b8f1"><td class="memTemplItemLeft" align="right" valign="top">void&#160;</td><td class="memTemplItemRight" valign="bottom"><a class="el" href="#a917b01e3de3372d276f8dd28f0e9b8f1">row_reduce_general</a> (const device T *in, device <a class="el" href="structmlx__atomic.html">mlx_atomic</a>&lt; U &gt; *out, const constant size_t &amp;reduction_size, const constant size_t &amp;out_size, const constant size_t &amp;non_row_reductions, const constant int *shape, const constant size_t *strides, const constant int &amp;ndim, uint3 lid, uint3 lsize, uint3 tid, uint simd_lane_id, uint simd_per_group, uint simd_group_id)</td></tr>
<tr class="separator:a917b01e3de3372d276f8dd28f0e9b8f1"><td class="memSeparator" colspan="2">&#160;</td></tr>
<tr class="memitem:a034fdccc83e66871550c2d42399f0ab5" id="r_a034fdccc83e66871550c2d42399f0ab5"><td class="memTemplParams" colspan="2">template&lt;typename T , typename U , typename Op , int N_READS = REDUCE_N_READS&gt; </td></tr>
<tr class="memitem:a034fdccc83e66871550c2d42399f0ab5"><td class="memTemplItemLeft" align="right" valign="top">void&#160;</td><td class="memTemplItemRight" valign="bottom"><a class="el" href="#a034fdccc83e66871550c2d42399f0ab5">row_reduce_general_no_atomics</a> (const device T *in, device U *out, const constant size_t &amp;reduction_size, const constant size_t &amp;out_size, const constant size_t &amp;non_row_reductions, const constant int *shape, const constant size_t *strides, const constant int &amp;ndim, uint3 lid, uint3 lsize, uint3 gsize, uint3 tid, uint simd_lane_id, uint simd_per_group, uint simd_group_id)</td></tr>
<tr class="separator:a034fdccc83e66871550c2d42399f0ab5"><td class="memSeparator" colspan="2">&#160;</td></tr>
<tr class="memitem:a9d5e0049a2276f43702fc6907e74a35f" id="r_a9d5e0049a2276f43702fc6907e74a35f"><td class="memTemplParams" colspan="2">template&lt;typename T , typename U , typename Op , int N_READS = REDUCE_N_READS, int N_WRITES = REDUCE_N_WRITES&gt; </td></tr>
<tr class="memitem:a9d5e0049a2276f43702fc6907e74a35f"><td class="memTemplItemLeft" align="right" valign="top">METAL_FUNC void&#160;</td><td class="memTemplItemRight" valign="bottom"><a class="el" href="#a9d5e0049a2276f43702fc6907e74a35f">per_thread_row_reduce</a> (thread U totals[N_WRITES], const device T *inputs[N_WRITES], int blocks, int extra, uint lsize_x, uint lid_x)</td></tr>
<tr class="memdesc:a9d5e0049a2276f43702fc6907e74a35f"><td class="mdescLeft">&#160;</td><td class="mdescRight">The thread group collaboratively reduces across the rows with bounds checking. <br /></td></tr>
<tr class="separator:a9d5e0049a2276f43702fc6907e74a35f"><td class="memSeparator" colspan="2">&#160;</td></tr>
<tr class="memitem:a045ec34228e77c79ec67d11c39ff097a" id="r_a045ec34228e77c79ec67d11c39ff097a"><td class="memTemplParams" colspan="2">template&lt;typename T , typename U , typename Op , int N_READS = REDUCE_N_READS, int N_WRITES = REDUCE_N_WRITES&gt; </td></tr>
<tr class="memitem:a045ec34228e77c79ec67d11c39ff097a"><td class="memTemplItemLeft" align="right" valign="top">METAL_FUNC void&#160;</td><td class="memTemplItemRight" valign="bottom"><a class="el" href="#a045ec34228e77c79ec67d11c39ff097a">per_thread_row_reduce</a> (thread U totals[N_WRITES], const device T *in, const constant size_t &amp;reduction_size, int blocks, int extra, uint lsize_x, uint lid_x)</td></tr>
<tr class="memdesc:a045ec34228e77c79ec67d11c39ff097a"><td class="mdescLeft">&#160;</td><td class="mdescRight">Consecutive rows in a contiguous array. <br /></td></tr>
<tr class="separator:a045ec34228e77c79ec67d11c39ff097a"><td class="memSeparator" colspan="2">&#160;</td></tr>
<tr class="memitem:a4d00c44e5f4a13be529ff8b664a0a342" id="r_a4d00c44e5f4a13be529ff8b664a0a342"><td class="memTemplParams" colspan="2">template&lt;typename T , typename U , typename Op , int N_READS = REDUCE_N_READS, int N_WRITES = REDUCE_N_WRITES&gt; </td></tr>
<tr class="memitem:a4d00c44e5f4a13be529ff8b664a0a342"><td class="memTemplItemLeft" align="right" valign="top">METAL_FUNC void&#160;</td><td class="memTemplItemRight" valign="bottom"><a class="el" href="#a4d00c44e5f4a13be529ff8b664a0a342">per_thread_row_reduce</a> (thread U totals[N_WRITES], const device T *in, const size_t row_idx, int blocks, int extra, const constant int *shape, const constant size_t *strides, const constant int &amp;ndim, uint lsize_x, uint lid_x)</td></tr>
<tr class="memdesc:a4d00c44e5f4a13be529ff8b664a0a342"><td class="mdescLeft">&#160;</td><td class="mdescRight">Consecutive rows in an arbitrarily ordered array. <br /></td></tr>
<tr class="separator:a4d00c44e5f4a13be529ff8b664a0a342"><td class="memSeparator" colspan="2">&#160;</td></tr>
<tr class="memitem:aa146bb611069fd2892f03714fd1cc3cf" id="r_aa146bb611069fd2892f03714fd1cc3cf"><td class="memTemplParams" colspan="2">template&lt;typename T , typename U , typename Op , int N_READS = REDUCE_N_READS, int N_WRITES = REDUCE_N_WRITES&gt; </td></tr>
<tr class="memitem:aa146bb611069fd2892f03714fd1cc3cf"><td class="memTemplItemLeft" align="right" valign="top">METAL_FUNC void&#160;</td><td class="memTemplItemRight" valign="bottom"><a class="el" href="#aa146bb611069fd2892f03714fd1cc3cf">threadgroup_reduce</a> (thread U totals[N_WRITES], threadgroup U *shared_vals, uint3 lid, uint simd_lane_id, uint simd_per_group, uint simd_group_id)</td></tr>
<tr class="memdesc:aa146bb611069fd2892f03714fd1cc3cf"><td class="mdescLeft">&#160;</td><td class="mdescRight">Reduce within the threadgroup. <br /></td></tr>
<tr class="separator:aa146bb611069fd2892f03714fd1cc3cf"><td class="memSeparator" colspan="2">&#160;</td></tr>
<tr class="memitem:afd80a25fa84e6cc884dcc8698859ade1" id="r_afd80a25fa84e6cc884dcc8698859ade1"><td class="memTemplParams" colspan="2">template&lt;typename T , typename U , typename Op , int N_READS = REDUCE_N_READS&gt; </td></tr>
<tr class="memitem:afd80a25fa84e6cc884dcc8698859ade1"><td class="memTemplItemLeft" align="right" valign="top">METAL_FUNC void&#160;</td><td class="memTemplItemRight" valign="bottom"><a class="el" href="#afd80a25fa84e6cc884dcc8698859ade1">thread_reduce</a> (thread U &amp;total, const device T *row, int blocks, int extra)</td></tr>
<tr class="separator:afd80a25fa84e6cc884dcc8698859ade1"><td class="memSeparator" colspan="2">&#160;</td></tr>
<tr class="memitem:a27e75312086e31f6bd1bbf4b366679da" id="r_a27e75312086e31f6bd1bbf4b366679da"><td class="memTemplParams" colspan="2">template&lt;typename T , typename U , typename Op , int NDIMS = 0, int N_READS = REDUCE_N_READS&gt; </td></tr>
<tr class="memitem:a27e75312086e31f6bd1bbf4b366679da"><td class="memTemplItemLeft" align="right" valign="top">void&#160;</td><td class="memTemplItemRight" valign="bottom"><a class="el" href="#a27e75312086e31f6bd1bbf4b366679da">row_reduce_small</a> (const device T *in, device U *out, const constant size_t &amp;row_size, const constant size_t &amp;non_row_reductions, const constant int *shape, const constant size_t *strides, const constant int &amp;ndim, const constant int *reduce_shape, const constant size_t *reduce_strides, const constant int &amp;reduce_ndim, uint simd_lane_id, uint3 gid, uint3 gsize, uint3 tid, uint3 tsize)</td></tr>
<tr class="separator:a27e75312086e31f6bd1bbf4b366679da"><td class="memSeparator" colspan="2">&#160;</td></tr>
<tr class="memitem:ac01d30987668930c8b38900e47b8308b" id="r_ac01d30987668930c8b38900e47b8308b"><td class="memTemplParams" colspan="2">template&lt;typename T , typename U , typename Op , int N_READS = REDUCE_N_READS, int N_WRITES = REDUCE_N_WRITES&gt; </td></tr>
<tr class="memitem:ac01d30987668930c8b38900e47b8308b"><td class="memTemplItemLeft" align="right" valign="top">void&#160;</td><td class="memTemplItemRight" valign="bottom"><a class="el" href="#ac01d30987668930c8b38900e47b8308b">row_reduce_simple</a> (const device T *in, device U *out, const constant size_t &amp;reduction_size, const constant size_t &amp;out_size, uint3 gid, uint3 gsize, uint3 lid, uint3 lsize, uint simd_lane_id, uint simd_per_group, uint simd_group_id)</td></tr>
<tr class="separator:ac01d30987668930c8b38900e47b8308b"><td class="memSeparator" colspan="2">&#160;</td></tr>
<tr class="memitem:ad98332d74a6824aa7499df3e2f2246ae" id="r_ad98332d74a6824aa7499df3e2f2246ae"><td class="memTemplParams" colspan="2">template&lt;typename T , typename U , typename Op , int NDIMS = 0, int N_READS = REDUCE_N_READS&gt; </td></tr>
<tr class="memitem:ad98332d74a6824aa7499df3e2f2246ae"><td class="memTemplItemLeft" align="right" valign="top">void&#160;</td><td class="memTemplItemRight" valign="bottom"><a class="el" href="#ad98332d74a6824aa7499df3e2f2246ae">row_reduce_looped</a> (const device T *in, device U *out, const constant size_t &amp;row_size, const constant size_t &amp;non_row_reductions, const constant int *shape, const constant size_t *strides, const constant int &amp;ndim, const constant int *reduce_shape, const constant size_t *reduce_strides, const constant int &amp;reduce_ndim, uint3 gid, uint3 gsize, uint3 lid, uint3 lsize, uint simd_lane_id, uint simd_per_group, uint simd_group_id)</td></tr>
<tr class="separator:ad98332d74a6824aa7499df3e2f2246ae"><td class="memSeparator" colspan="2">&#160;</td></tr>
</table>
<h2 class="groupheader">Function Documentation</h2>
<a id="a5df83c60214206d584be1587c9b723a5" name="a5df83c60214206d584be1587c9b723a5"></a>
<h2 class="memtitle"><span class="permalink"><a href="#a5df83c60214206d584be1587c9b723a5">&#9670;&#160;</a></span>per_thread_row_reduce()</h2>
<a id="a045ec34228e77c79ec67d11c39ff097a" name="a045ec34228e77c79ec67d11c39ff097a"></a>
<h2 class="memtitle"><span class="permalink"><a href="#a045ec34228e77c79ec67d11c39ff097a">&#9670;&#160;</a></span>per_thread_row_reduce() <span class="overload">[1/3]</span></h2>
<div class="memitem">
<div class="memproto">
<div class="memtemplate">
template&lt;typename T , typename U , typename Op , int N_READS = REDUCE_N_READS&gt; </div>
template&lt;typename T , typename U , typename Op , int N_READS = REDUCE_N_READS, int N_WRITES = REDUCE_N_WRITES&gt; </div>
<table class="memname">
<tr>
<td class="memname">METAL_FUNC U per_thread_row_reduce </td>
<td class="memname">METAL_FUNC void per_thread_row_reduce </td>
<td>(</td>
<td class="paramtype">thread U</td> <td class="paramname"><span class="paramname"><em>totals</em>[N_WRITES], </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const device T *</td> <td class="paramname"><span class="paramname"><em>in</em>, </span></td>
</tr>
<tr>
@@ -123,7 +141,62 @@ template&lt;typename T , typename U , typename Op , int N_READS = REDUCE_N_READS
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const constant size_t &amp;</td> <td class="paramname"><span class="paramname"><em>out_size</em>, </span></td>
<td class="paramtype">int</td> <td class="paramname"><span class="paramname"><em>blocks</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">int</td> <td class="paramname"><span class="paramname"><em>extra</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">uint</td> <td class="paramname"><span class="paramname"><em>lsize_x</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">uint</td> <td class="paramname"><span class="paramname"><em>lid_x</em></span>&#160;)</td>
</tr>
</table>
</div><div class="memdoc">
<p>Consecutive rows in a contiguous array. </p>
</div>
</div>
<a id="a4d00c44e5f4a13be529ff8b664a0a342" name="a4d00c44e5f4a13be529ff8b664a0a342"></a>
<h2 class="memtitle"><span class="permalink"><a href="#a4d00c44e5f4a13be529ff8b664a0a342">&#9670;&#160;</a></span>per_thread_row_reduce() <span class="overload">[2/3]</span></h2>
<div class="memitem">
<div class="memproto">
<div class="memtemplate">
template&lt;typename T , typename U , typename Op , int N_READS = REDUCE_N_READS, int N_WRITES = REDUCE_N_WRITES&gt; </div>
<table class="memname">
<tr>
<td class="memname">METAL_FUNC void per_thread_row_reduce </td>
<td>(</td>
<td class="paramtype">thread U</td> <td class="paramname"><span class="paramname"><em>totals</em>[N_WRITES], </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const device T *</td> <td class="paramname"><span class="paramname"><em>in</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const size_t</td> <td class="paramname"><span class="paramname"><em>row_idx</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">int</td> <td class="paramname"><span class="paramname"><em>blocks</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">int</td> <td class="paramname"><span class="paramname"><em>extra</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
@@ -148,45 +221,83 @@ template&lt;typename T , typename U , typename Op , int N_READS = REDUCE_N_READS
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">uint</td> <td class="paramname"><span class="paramname"><em>lid_x</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">uint2</td> <td class="paramname"><span class="paramname"><em>tid</em></span>&#160;)</td>
<td class="paramtype">uint</td> <td class="paramname"><span class="paramname"><em>lid_x</em></span>&#160;)</td>
</tr>
</table>
</div><div class="memdoc">
<p>Consecutive rows in an arbitrarily ordered array. </p>
</div>
</div>
<a id="a917b01e3de3372d276f8dd28f0e9b8f1" name="a917b01e3de3372d276f8dd28f0e9b8f1"></a>
<h2 class="memtitle"><span class="permalink"><a href="#a917b01e3de3372d276f8dd28f0e9b8f1">&#9670;&#160;</a></span>row_reduce_general()</h2>
<a id="a9d5e0049a2276f43702fc6907e74a35f" name="a9d5e0049a2276f43702fc6907e74a35f"></a>
<h2 class="memtitle"><span class="permalink"><a href="#a9d5e0049a2276f43702fc6907e74a35f">&#9670;&#160;</a></span>per_thread_row_reduce() <span class="overload">[3/3]</span></h2>
<div class="memitem">
<div class="memproto">
<div class="memtemplate">
template&lt;typename T , typename U , typename Op , int N_READS = REDUCE_N_READS&gt; </div>
template&lt;typename T , typename U , typename Op , int N_READS = REDUCE_N_READS, int N_WRITES = REDUCE_N_WRITES&gt; </div>
<table class="memname">
<tr>
<td class="memname">void row_reduce_general </td>
<td class="memname">METAL_FUNC void per_thread_row_reduce </td>
<td>(</td>
<td class="paramtype">thread U</td> <td class="paramname"><span class="paramname"><em>totals</em>[N_WRITES], </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const device T *</td> <td class="paramname"><span class="paramname"><em>inputs</em>[N_WRITES], </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">int</td> <td class="paramname"><span class="paramname"><em>blocks</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">int</td> <td class="paramname"><span class="paramname"><em>extra</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">uint</td> <td class="paramname"><span class="paramname"><em>lsize_x</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">uint</td> <td class="paramname"><span class="paramname"><em>lid_x</em></span>&#160;)</td>
</tr>
</table>
</div><div class="memdoc">
<p>The thread group collaboratively reduces across the rows with bounds checking. </p>
<p>In the end each thread holds a part of the reduction. </p>
</div>
</div>
<a id="ad98332d74a6824aa7499df3e2f2246ae" name="ad98332d74a6824aa7499df3e2f2246ae"></a>
<h2 class="memtitle"><span class="permalink"><a href="#ad98332d74a6824aa7499df3e2f2246ae">&#9670;&#160;</a></span>row_reduce_looped()</h2>
<div class="memitem">
<div class="memproto">
<div class="memtemplate">
template&lt;typename T , typename U , typename Op , int NDIMS = 0, int N_READS = REDUCE_N_READS&gt; </div>
<table class="memname">
<tr>
<td class="memname">void row_reduce_looped </td>
<td>(</td>
<td class="paramtype">const device T *</td> <td class="paramname"><span class="paramname"><em>in</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">device <a class="el" href="structmlx__atomic.html">mlx_atomic</a>&lt; U &gt; *</td> <td class="paramname"><span class="paramname"><em>out</em>, </span></td>
<td class="paramtype">device U *</td> <td class="paramname"><span class="paramname"><em>out</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const constant size_t &amp;</td> <td class="paramname"><span class="paramname"><em>reduction_size</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const constant size_t &amp;</td> <td class="paramname"><span class="paramname"><em>out_size</em>, </span></td>
<td class="paramtype">const constant size_t &amp;</td> <td class="paramname"><span class="paramname"><em>row_size</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
@@ -208,6 +319,31 @@ template&lt;typename T , typename U , typename Op , int N_READS = REDUCE_N_READS
<td></td>
<td class="paramtype">const constant int &amp;</td> <td class="paramname"><span class="paramname"><em>ndim</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const constant int *</td> <td class="paramname"><span class="paramname"><em>reduce_shape</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const constant size_t *</td> <td class="paramname"><span class="paramname"><em>reduce_strides</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const constant int &amp;</td> <td class="paramname"><span class="paramname"><em>reduce_ndim</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">uint3</td> <td class="paramname"><span class="paramname"><em>gid</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">uint3</td> <td class="paramname"><span class="paramname"><em>gsize</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
@@ -218,11 +354,6 @@ template&lt;typename T , typename U , typename Op , int N_READS = REDUCE_N_READS
<td></td>
<td class="paramtype">uint3</td> <td class="paramname"><span class="paramname"><em>lsize</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">uint3</td> <td class="paramname"><span class="paramname"><em>tid</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
@@ -243,16 +374,16 @@ template&lt;typename T , typename U , typename Op , int N_READS = REDUCE_N_READS
</div>
</div>
<a id="ad77d444bcfa9e6afcc1ffc1dcb919f1c" name="ad77d444bcfa9e6afcc1ffc1dcb919f1c"></a>
<h2 class="memtitle"><span class="permalink"><a href="#ad77d444bcfa9e6afcc1ffc1dcb919f1c">&#9670;&#160;</a></span>row_reduce_general_med()</h2>
<a id="ac01d30987668930c8b38900e47b8308b" name="ac01d30987668930c8b38900e47b8308b"></a>
<h2 class="memtitle"><span class="permalink"><a href="#ac01d30987668930c8b38900e47b8308b">&#9670;&#160;</a></span>row_reduce_simple()</h2>
<div class="memitem">
<div class="memproto">
<div class="memtemplate">
template&lt;typename T , typename U , typename Op &gt; </div>
template&lt;typename T , typename U , typename Op , int N_READS = REDUCE_N_READS, int N_WRITES = REDUCE_N_WRITES&gt; </div>
<table class="memname">
<tr>
<td class="memname">void row_reduce_general_med </td>
<td class="memname">void row_reduce_simple </td>
<td>(</td>
<td class="paramtype">const device T *</td> <td class="paramname"><span class="paramname"><em>in</em>, </span></td>
</tr>
@@ -274,27 +405,22 @@ template&lt;typename T , typename U , typename Op &gt; </div>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const constant size_t &amp;</td> <td class="paramname"><span class="paramname"><em>non_row_reductions</em>, </span></td>
<td class="paramtype">uint3</td> <td class="paramname"><span class="paramname"><em>gid</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const constant int *</td> <td class="paramname"><span class="paramname"><em>shape</em>, </span></td>
<td class="paramtype">uint3</td> <td class="paramname"><span class="paramname"><em>gsize</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const constant size_t *</td> <td class="paramname"><span class="paramname"><em>strides</em>, </span></td>
<td class="paramtype">uint3</td> <td class="paramname"><span class="paramname"><em>lid</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const constant int &amp;</td> <td class="paramname"><span class="paramname"><em>ndim</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">uint</td> <td class="paramname"><span class="paramname"><em>tid</em>, </span></td>
<td class="paramtype">uint3</td> <td class="paramname"><span class="paramname"><em>lsize</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
@@ -316,16 +442,16 @@ template&lt;typename T , typename U , typename Op &gt; </div>
</div>
</div>
<a id="a034fdccc83e66871550c2d42399f0ab5" name="a034fdccc83e66871550c2d42399f0ab5"></a>
<h2 class="memtitle"><span class="permalink"><a href="#a034fdccc83e66871550c2d42399f0ab5">&#9670;&#160;</a></span>row_reduce_general_no_atomics()</h2>
<a id="a27e75312086e31f6bd1bbf4b366679da" name="a27e75312086e31f6bd1bbf4b366679da"></a>
<h2 class="memtitle"><span class="permalink"><a href="#a27e75312086e31f6bd1bbf4b366679da">&#9670;&#160;</a></span>row_reduce_small()</h2>
<div class="memitem">
<div class="memproto">
<div class="memtemplate">
template&lt;typename T , typename U , typename Op , int N_READS = REDUCE_N_READS&gt; </div>
template&lt;typename T , typename U , typename Op , int NDIMS = 0, int N_READS = REDUCE_N_READS&gt; </div>
<table class="memname">
<tr>
<td class="memname">void row_reduce_general_no_atomics </td>
<td class="memname">void row_reduce_small </td>
<td>(</td>
<td class="paramtype">const device T *</td> <td class="paramname"><span class="paramname"><em>in</em>, </span></td>
</tr>
@@ -337,12 +463,7 @@ template&lt;typename T , typename U , typename Op , int N_READS = REDUCE_N_READS
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const constant size_t &amp;</td> <td class="paramname"><span class="paramname"><em>reduction_size</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const constant size_t &amp;</td> <td class="paramname"><span class="paramname"><em>out_size</em>, </span></td>
<td class="paramtype">const constant size_t &amp;</td> <td class="paramname"><span class="paramname"><em>row_size</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
@@ -367,12 +488,27 @@ template&lt;typename T , typename U , typename Op , int N_READS = REDUCE_N_READS
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">uint3</td> <td class="paramname"><span class="paramname"><em>lid</em>, </span></td>
<td class="paramtype">const constant int *</td> <td class="paramname"><span class="paramname"><em>reduce_shape</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">uint3</td> <td class="paramname"><span class="paramname"><em>lsize</em>, </span></td>
<td class="paramtype">const constant size_t *</td> <td class="paramname"><span class="paramname"><em>reduce_strides</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const constant int &amp;</td> <td class="paramname"><span class="paramname"><em>reduce_ndim</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">uint</td> <td class="paramname"><span class="paramname"><em>simd_lane_id</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">uint3</td> <td class="paramname"><span class="paramname"><em>gid</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
@@ -384,6 +520,72 @@ template&lt;typename T , typename U , typename Op , int N_READS = REDUCE_N_READS
<td></td>
<td class="paramtype">uint3</td> <td class="paramname"><span class="paramname"><em>tid</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">uint3</td> <td class="paramname"><span class="paramname"><em>tsize</em></span>&#160;)</td>
</tr>
</table>
</div><div class="memdoc">
</div>
</div>
<a id="afd80a25fa84e6cc884dcc8698859ade1" name="afd80a25fa84e6cc884dcc8698859ade1"></a>
<h2 class="memtitle"><span class="permalink"><a href="#afd80a25fa84e6cc884dcc8698859ade1">&#9670;&#160;</a></span>thread_reduce()</h2>
<div class="memitem">
<div class="memproto">
<div class="memtemplate">
template&lt;typename T , typename U , typename Op , int N_READS = REDUCE_N_READS&gt; </div>
<table class="memname">
<tr>
<td class="memname">METAL_FUNC void thread_reduce </td>
<td>(</td>
<td class="paramtype">thread U &amp;</td> <td class="paramname"><span class="paramname"><em>total</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const device T *</td> <td class="paramname"><span class="paramname"><em>row</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">int</td> <td class="paramname"><span class="paramname"><em>blocks</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">int</td> <td class="paramname"><span class="paramname"><em>extra</em></span>&#160;)</td>
</tr>
</table>
</div><div class="memdoc">
</div>
</div>
<a id="aa146bb611069fd2892f03714fd1cc3cf" name="aa146bb611069fd2892f03714fd1cc3cf"></a>
<h2 class="memtitle"><span class="permalink"><a href="#aa146bb611069fd2892f03714fd1cc3cf">&#9670;&#160;</a></span>threadgroup_reduce()</h2>
<div class="memitem">
<div class="memproto">
<div class="memtemplate">
template&lt;typename T , typename U , typename Op , int N_READS = REDUCE_N_READS, int N_WRITES = REDUCE_N_WRITES&gt; </div>
<table class="memname">
<tr>
<td class="memname">METAL_FUNC void threadgroup_reduce </td>
<td>(</td>
<td class="paramtype">thread U</td> <td class="paramname"><span class="paramname"><em>totals</em>[N_WRITES], </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">threadgroup U *</td> <td class="paramname"><span class="paramname"><em>shared_vals</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">uint3</td> <td class="paramname"><span class="paramname"><em>lid</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
@@ -402,63 +604,7 @@ template&lt;typename T , typename U , typename Op , int N_READS = REDUCE_N_READS
</table>
</div><div class="memdoc">
</div>
</div>
<a id="aea2b1ba640fffb71c73550ec747b3148" name="aea2b1ba640fffb71c73550ec747b3148"></a>
<h2 class="memtitle"><span class="permalink"><a href="#aea2b1ba640fffb71c73550ec747b3148">&#9670;&#160;</a></span>row_reduce_general_small()</h2>
<div class="memitem">
<div class="memproto">
<div class="memtemplate">
template&lt;typename T , typename U , typename Op &gt; </div>
<table class="memname">
<tr>
<td class="memname">void row_reduce_general_small </td>
<td>(</td>
<td class="paramtype">const device T *</td> <td class="paramname"><span class="paramname"><em>in</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">device U *</td> <td class="paramname"><span class="paramname"><em>out</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const constant size_t &amp;</td> <td class="paramname"><span class="paramname"><em>reduction_size</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const constant size_t &amp;</td> <td class="paramname"><span class="paramname"><em>out_size</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const constant size_t &amp;</td> <td class="paramname"><span class="paramname"><em>non_row_reductions</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const constant int *</td> <td class="paramname"><span class="paramname"><em>shape</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const constant size_t *</td> <td class="paramname"><span class="paramname"><em>strides</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const constant int &amp;</td> <td class="paramname"><span class="paramname"><em>ndim</em>, </span></td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">uint</td> <td class="paramname"><span class="paramname"><em>lid</em></span>&#160;)</td>
</tr>
</table>
</div><div class="memdoc">
<p>Reduce within the threadgroup. </p>
</div>
</div>