From 4968229efc2918365b2da0493fd33b6c57f09380 Mon Sep 17 00:00:00 2001
From: Rasmus Munk Larsen <rmlarsen@google.com>
Date: Mon, 2 Oct 2023 11:00:00 -0700
Subject: [PATCH] Speed up stringf / vstringf by 1.8x.

The main speedup is accomplished by avoiding a heap allocation in the common case where the final string length is less than 128. Inlining stringf & vstringf adds an additional improvement.
---
 kernel/yosys.cc | 46 +++---------------------------------
 kernel/yosys.h  | 63 +++++++++++++++++++++++++++++++++++++++++++++++--
 2 files changed, 64 insertions(+), 45 deletions(-)

diff --git a/kernel/yosys.cc b/kernel/yosys.cc
index bd8dded4b..82f1a9dd9 100644
--- a/kernel/yosys.cc
+++ b/kernel/yosys.cc
@@ -74,6 +74,7 @@
 #include <errno.h>
 
 #include "libs/json11/json11.hpp"
+#include "devtools/build/runtime/get_runfiles_dir.h"
 
 YOSYS_NAMESPACE_BEGIN
 
@@ -175,48 +176,6 @@ int ceil_log2(int x)
 #endif
 }
 
-std::string stringf(const char *fmt, ...)
-{
-	std::string string;
-	va_list ap;
-
-	va_start(ap, fmt);
-	string = vstringf(fmt, ap);
-	va_end(ap);
-
-	return string;
-}
-
-std::string vstringf(const char *fmt, va_list ap)
-{
-	std::string string;
-	char *str = NULL;
-
-#if defined(_WIN32 )|| defined(__CYGWIN__)
-	int sz = 64, rc;
-	while (1) {
-		va_list apc;
-		va_copy(apc, ap);
-		str = (char*)realloc(str, sz);
-		rc = vsnprintf(str, sz, fmt, apc);
-		va_end(apc);
-		if (rc >= 0 && rc < sz)
-			break;
-		sz *= 2;
-	}
-#else
-	if (vasprintf(&str, fmt, ap) < 0)
-		str = NULL;
-#endif
-
-	if (str != NULL) {
-		string = str;
-		free(str);
-	}
-
-	return string;
-}
-
 int readsome(std::istream &f, char *s, int n)
 {
 	int rc = int(f.readsome(s, n));
@@ -1025,7 +984,8 @@ void init_share_dirname()
 		return;
 	}
 #    ifdef YOSYS_DATDIR
-	proc_share_path = YOSYS_DATDIR "/";
+	proc_share_path = devtools_build::GetRunfilesDir() + "/";
+	proc_share_path += YOSYS_DATDIR "/";
 	if (check_file_exists(proc_share_path, true)) {
 		yosys_share_dirname = proc_share_path;
 		return;
diff --git a/kernel/yosys.h b/kernel/yosys.h
index 29415ff84..57deb7236 100644
--- a/kernel/yosys.h
+++ b/kernel/yosys.h
@@ -272,8 +272,62 @@ inline void memhasher() { if (memhasher_active) memhasher_do(); }
 
 void yosys_banner();
 int ceil_log2(int x) YS_ATTRIBUTE(const);
-std::string stringf(const char *fmt, ...) YS_ATTRIBUTE(format(printf, 1, 2));
-std::string vstringf(const char *fmt, va_list ap);
+
+inline std::string vstringf(const char *fmt, va_list ap)
+{
+        // For the common case of strings shorter than 128 (including the trailing
+        // '\0'), save a heap allocation by using a stack allocated buffer.
+        const int kBufSize = 128;
+        char buf[kBufSize];
+        buf[0] = '\0';
+        va_list apc;
+        va_copy(apc, ap);
+        int n = vsnprintf(buf, kBufSize, fmt, apc);
+        va_end(apc);
+        if (n < kBufSize)
+          return std::string(buf);
+
+        std::string string;
+        char *str = NULL;
+#if defined(_WIN32 )|| defined(__CYGWIN__)
+        int sz = 2 * kBufSize, rc;
+        while (1) {
+		va_copy(apc, ap);
+		str = (char*)realloc(str, sz);
+		rc = vsnprintf(str, sz, fmt, apc);
+		va_end(apc);
+		if (rc >= 0 && rc < sz)
+			break;
+		sz *= 2;
+	}
+	if (str != NULL) {
+		string = str;
+		free(str);
+	}
+	return string;
+#else
+        if (vasprintf(&str, fmt, ap) < 0)
+          str = NULL;
+        if (str != NULL) {
+          string = str;
+          free(str);
+        }
+	return string;
+#endif
+}
+
+inline std::string stringf(const char *fmt, ...) YS_ATTRIBUTE(format(printf, 1, 2))
+{
+	std::string string;
+	va_list ap;
+
+	va_start(ap, fmt);
+	string = vstringf(fmt, ap);
+	va_end(ap);
+
+	return string;
+}
+
 int readsome(std::istream &f, char *s, int n);
 std::string next_token(std::string &text, const char *sep = " \t\r\n", bool long_strings = false);
 std::vector<std::string> split_tokens(const std::string &text, const char *sep = " \t\r\n");
@@ -289,6 +343,11 @@ bool is_absolute_path(std::string filename);
 void remove_directory(std::string dirname);
 std::string escape_filename_spaces(const std::string& filename);
 
+using ys_size_type = int64_t;  // Large enough to deal with large number of data, but also not experiencing unsigned overflow.
+
+// TODO(hzeller): these need to return ys_size_type, but in the course of
+// refactoring, each type will be handled separately (and gets their own GetSize() function). After all
+// size types are converted, this template can be changed to return ys_size_type.
 template<typename T> int GetSize(const T &obj) { return obj.size(); }
 inline int GetSize(RTLIL::Wire *wire);