diff options
Diffstat (limited to 'print.html')
| -rw-r--r-- | print.html | 110 |
1 files changed, 101 insertions, 9 deletions
@@ -25,9 +25,9 @@ <link rel="stylesheet" href="fonts/fonts.css"> <!-- Highlight.js Stylesheets --> - <link rel="stylesheet" href="highlight.css"> - <link rel="stylesheet" href="tomorrow-night.css"> - <link rel="stylesheet" href="ayu-highlight.css"> + <link rel="stylesheet" id="highlight-css" href="highlight.css"> + <link rel="stylesheet" id="tomorrow-night-css" href="tomorrow-night.css"> + <link rel="stylesheet" id="ayu-highlight-css" href="ayu-highlight.css"> <!-- Custom theme stylesheets --> @@ -2828,6 +2828,7 @@ renice -n 5 -p PID <li><a href="trace_profile/./oprofile.html">OProfile</a></li> <li><a href="trace_profile/./callgrind.html">callgrind</a></li> <li><a href="trace_profile/./valgrind.html">valgrind</a></li> +<li><a href="trace_profile/./vtune.html">vtune</a></li> </ul> <div style="break-before: page; page-break-before: always;"></div><h1 id="usrbintime1"><a class="header" href="#usrbintime1">/usr/bin/time(1)</a></h1> <pre><code class="language-markdown"># statistics of process run @@ -3222,6 +3223,59 @@ int main() { --gen-suppressions=yes Generate suppressions file from the run. --suppressions=FILE Load suppressions file. </code></pre> +<div style="break-before: page; page-break-before: always;"></div><h1 id="vtune1"><a class="header" href="#vtune1"><a href="https://www.intel.com/content/www/us/en/docs/vtune-profiler/user-guide">vtune(1)</a></a></h1> +<p>Vtune offers different analysis. Run <code>vtune -collect help</code> to list the +availale analysis.</p> +<h2 id="profiling"><a class="header" href="#profiling">Profiling</a></h2> +<p>The following shows some common flows with the <code>hotspot</code> analsysis +as an example.</p> +<pre><code># Launch and profile process. +vtune -collect hotspots [opts] -- target [args] + +# Attach and profile running process. +vtune -collect hotspots [opts] -target-pid <pid> +</code></pre> +<p>Some common options are the following.</p> +<pre><code>-r <dir> output directory for the profile +-no-follow-child dont attach to to child processes (default is to follow) +-start-paused start with paused profiling +</code></pre> +<h2 id="analyze"><a class="header" href="#analyze">Analyze</a></h2> +<pre><code>vtune-gui <dir> +</code></pre> +<h2 id="programmatically-control-sampling"><a class="header" href="#programmatically-control-sampling">Programmatically control sampling</a></h2> +<p>Vtune offers an API to <em>resume</em> and <em>pause</em> the profile collection from within +the profilee itself. This can be helpful if either only a certain phase should +be profiled or some phase should be skipped.</p> +<p>The following gives an example where only one phase in the program is profiled. +The program makes calls to the vtune API to resume and pause the collection, +while vtune is invoked with <code>-start-paused</code> to pause profiling initially.</p> +<pre><code class="language-c">#include <ittnotify.h> + +void init(); +void compute(); +void shutdown(); + +int main() { + init(); + + __itt_resume(); + compute(); + __itt_pause(); + + shutdown(); + return 0; +} +</code></pre> +<p>The makefile gives an example how to build and profile the application.</p> +<pre><code class="language-makefile">VTUNE ?= /opt/intel/oneapi/vtune/latest + +main: main.c + gcc -o $@ $^ -I$(VTUNE)/include -L$(VTUNE)/lib64 -littnotify + +vtune: main + $(VTUNE)/bin64/vtune -collect hotspots -start-paused -- ./main +</code></pre> <div style="break-before: page; page-break-before: always;"></div><h1 id="debug-1"><a class="header" href="#debug-1">Debug</a></h1> <ul> <li><a href="debug/./gdb.html">gdb</a></li> @@ -4867,6 +4921,16 @@ LD_PRELOAD=./libmtrace.so <binary> </code></pre> <div style="break-before: page; page-break-before: always;"></div><h1 id="gcc1"><a class="header" href="#gcc1">gcc(1)</a></h1> <h2 id="cli-2"><a class="header" href="#cli-2">CLI</a></h2> +<ul> +<li><code>-v</code> verbose, outputs exact compiler/linker invocations made by the gcc driver</li> +<li><code>-###</code> dry-run, outputting exact compiler/linker invocations</li> +<li><code>-print-multi-lib</code> print available multilib configurations</li> +<li><code>--help=<class></code> print description of cmdline options for given class, eg +<code>warnings</code>, <code>optimizers</code>, <code>target</code>, <code>c</code>, <code>c++</code></li> +<li><code>-Wl,<opt></code> additional option passed to the linker invocation (can +be specified multiple times)</li> +<li><code>-Wl,--trace</code> trace each file the linker touches</li> +</ul> <h3 id="preprocessing"><a class="header" href="#preprocessing">Preprocessing</a></h3> <p>While debugging can be helpful to just pre-process files.</p> <pre><code class="language-bash">gcc -E [-dM] ... @@ -4874,10 +4938,6 @@ LD_PRELOAD=./libmtrace.so <binary> <ul> <li><code>-E</code> run only preprocessor</li> <li><code>-dM</code> list only <code>#define</code> statements</li> -<li><code>-###</code> dry-run, outputting exact compiler/linker invocations</li> -<li><code>-print-multi-lib</code> print available multilib configurations</li> -<li><code>--help=<class></code> print description of cmdline options for given class, eg -<code>warnings</code>, <code>optimizers</code>, <code>target</code>, <code>c</code>, <code>c++</code></li> </ul> <h3 id="target-options"><a class="header" href="#target-options">Target options</a></h3> <pre><code class="language-bash"># List all target options with their description. @@ -4895,6 +4955,19 @@ gcc --help=optimizers # Prepend --help with `-Q` to print wheter options are enabled or disabled # instead showing their description. </code></pre> +<h3 id="sanitizer"><a class="header" href="#sanitizer">Sanitizer</a></h3> +<pre><code class="language-bash"># Enable address sanitizer, a memory error checker (out of bounds, use after free, ..). +gcc -fsanitize=address ... + +# Enable leak sanitizer, a memory leak detector. +gcc -fsanitize=leak + +# Enable undefined behavior sanitizer, detects various UBs (integer overflow, ..). +gcc -fsanitize=undefined ... + +# Enable thread sanitizer, a data race detector. +gcc -fsanitize=thread +</code></pre> <h2 id="builtins"><a class="header" href="#builtins"><a href="https://gcc.gnu.org/onlinedocs/gcc/Other-Builtins.html">Builtins</a></a></h2> <h3 id="__builtin_expectexpr-cond"><a class="header" href="#__builtin_expectexpr-cond"><code>__builtin_expect(expr, cond)</code></a></h3> <p>Give the compiler a hint which branch is hot, so it can lay out the code @@ -8231,6 +8304,9 @@ popfd // pop flags (4byte) from stack <pre><code class="language-x86asm">rdmsr // Read MSR register, effectively does EDX:EAX <- MSR[ECX] wrmsr // Write MSR register, effectively does MSR[ECX] <- EDX:EAX </code></pre> +<blockquote> +<p>See <a href="https://github.com/johannst/mini-kvm-rs/blob/main/guest/guest64-msr.S">guest64-msr.S</a> as an example.</p> +</blockquote> <h2 id="size-directives"><a class="header" href="#size-directives">Size directives</a></h2> <p>Explicitly specify size of the operation.</p> <pre><code class="language-x86asm">mov byte ptr [rax], 0xff // save 1 byte(s) at [rax] @@ -8316,6 +8392,13 @@ core changes frequency. This is the architectural behavior moving forward.</p> implemented TSC ticks with a constant frequency.</p> <pre><code class="language-sh">grep constant_tsc /proc/cpuinfo </code></pre> +<h2 id="cpu--hw-features---cpuid"><a class="header" href="#cpu--hw-features---cpuid">Cpu & hw features - <code>cpuid</code></a></h2> +<pre><code class="language-x86asm">cpuid // in: eax leaf; ecx sub-leaf + // out: eax, ebx, ecx, edx (interpreting depends on leaf) +</code></pre> +<p>This instruction is used to query for availability of certain +instructions or hardware details like cache sizes and son on.</p> +<p>An example how to read cpuid leafs is show in <a href="arch/x86/cpuid/cpuid.c">cpuid.c</a>.</p> <h2 id="sysv-x86_64-abi"><a class="header" href="#sysv-x86_64-abi"><a href="https://gitlab.com/x86-psABIs/x86-64-ABI">SysV x86_64 ABI</a></a></h2> <h3 id="passing-arguments-to-functions"><a class="header" href="#passing-arguments-to-functions">Passing arguments to functions</a></h3> <ul> @@ -8666,8 +8749,8 @@ itself.</p> <pre><code># Segment Selector (cs, ds, es, ss, fs, gs). [15:3] I Descriptor Index - [2:1] TI Table Indicator (0=GTD | 1=LDT) - [0] RPL Requested Privilege Level + [2] TI Table Indicator (0=GTD | 1=LDT) + [1:0] RPL Requested Privilege Level # Segment Descriptor (2 x 4 byte words). @@ -8693,6 +8776,15 @@ itself.</p> [47:16] Base address of GDT table. [15:0] Length of GDT table. </code></pre> +<blockquote> +<p>In 64-bit mode the <code>{cs, ds, es, ss}</code> segment register have no +effect, segmentation is effectively disabled. The <code>{gs, fs}</code> segment +register however can still be used for segmented memory access in +64-bit with paging enabled. Segmentation takes place before VA -> PA +address translation.</p> +<p>The example in <a href="arch/x86/seg/seg.c">seg.c</a> shows how to set the <code>gs</code> base +address and to relative accesses.</p> +</blockquote> <h2 id="references-8"><a class="header" href="#references-8">References</a></h2> <ul> <li><a href="https://gitlab.com/x86-psABIs/x86-64-ABI">SystemV AMD64 ABI</a></li> |