From 1106e89c84de1f6fef6ef39bc2e84e32cf40d8e1 Mon Sep 17 00:00:00 2001
From: Geza Lore <glore@tenstorrent.com>
Date: Thu, 2 Jul 2026 14:50:55 +0100
Subject: [PATCH] Optimize random initialization

Fetching a thread-local is relatively expensive. Random initializing
wides used to do it once or twice per word, and on short runs for
large designs can be noticeably expensive, so fetch once per variable
instead. Also remove unused VL_RAND_RESET_{Q,W} functions.
---
 include/verilated.cpp     | 47 +++++++++++++++++----------------------
 include/verilated_funcs.h |  4 ----
 2 files changed, 20 insertions(+), 31 deletions(-)

diff --git a/include/verilated.cpp b/include/verilated.cpp
index a5fdc91ac..d981f7c0a 100644
--- a/include/verilated.cpp
+++ b/include/verilated.cpp
@@ -563,9 +563,10 @@ IData VL_URANDOM_SEEDED_II(IData seed) VL_MT_SAFE {
 }
 
 IData VL_SCOPED_RAND_RESET_I(int obits, uint64_t scopeHash, uint64_t salt) VL_MT_UNSAFE {
-    if (Verilated::threadContextp()->randReset() == 0) return 0;
+    const int randReset = Verilated::threadContextp()->randReset();
+    if (randReset == 0) return 0;
     IData data = ~0;
-    if (Verilated::threadContextp()->randReset() != 1) {  // if 2, randomize
+    if (randReset != 1) {  // if 2, randomize
         VlRNG rng{Verilated::threadContextp()->randSeed() ^ scopeHash ^ salt};
         data = rng.rand64();
     }
@@ -574,9 +575,10 @@ IData VL_SCOPED_RAND_RESET_I(int obits, uint64_t scopeHash, uint64_t salt) VL_MT
 }
 
 QData VL_SCOPED_RAND_RESET_Q(int obits, uint64_t scopeHash, uint64_t salt) VL_MT_UNSAFE {
-    if (Verilated::threadContextp()->randReset() == 0) return 0;
+    const int randReset = Verilated::threadContextp()->randReset();
+    if (randReset == 0) return 0;
     QData data = ~0ULL;
-    if (Verilated::threadContextp()->randReset() != 1) {  // if 2, randomize
+    if (randReset != 1) {  // if 2, randomize
         VlRNG rng{Verilated::threadContextp()->randSeed() ^ scopeHash ^ salt};
         data = rng.rand64();
     }
@@ -586,10 +588,17 @@ QData VL_SCOPED_RAND_RESET_Q(int obits, uint64_t scopeHash, uint64_t salt) VL_MT
 
 WDataOutP VL_SCOPED_RAND_RESET_W(int obits, WDataOutP outwp, uint64_t scopeHash,
                                  uint64_t salt) VL_MT_UNSAFE {
-    if (Verilated::threadContextp()->randReset() != 2) { return VL_RAND_RESET_W(obits, outwp); }
-    VlRNG rng{Verilated::threadContextp()->randSeed() ^ scopeHash ^ salt};
-    for (int i = 0; i < VL_WORDS_I(obits) - 1; ++i) outwp[i] = rng.rand64();
-    outwp[VL_WORDS_I(obits) - 1] = rng.rand64() & VL_MASK_E(obits);
+    const int words = VL_WORDS_I(obits);
+    const int randReset = Verilated::threadContextp()->randReset();
+    if (randReset == 0) {
+        VL_MEMSET_ZERO_W(outwp, words);
+    } else if (randReset == 1) {
+        VL_MEMSET_ONES_W(outwp, words);
+    } else {
+        VlRNG rng{Verilated::threadContextp()->randSeed() ^ scopeHash ^ salt};
+        for (int i = 0; i < words; ++i) outwp[i] = rng.rand64();
+    }
+    outwp[words - 1] &= VL_MASK_E(obits);
     return outwp;
 }
 
@@ -614,30 +623,14 @@ WDataOutP VL_SCOPED_RAND_RESET_ASSIGN_W(int obits, WDataOutP outwp, uint64_t sco
 }
 
 IData VL_RAND_RESET_I(int obits) VL_MT_SAFE {
-    if (Verilated::threadContextp()->randReset() == 0) return 0;
+    const int randReset = Verilated::threadContextp()->randReset();
+    if (randReset == 0) return 0;
     IData data = ~0;
-    if (Verilated::threadContextp()->randReset() != 1) {  // if 2, randomize
-        data = VL_RANDOM_I();
-    }
+    if (randReset != 1) data = VL_RANDOM_I();  // if 2, randomize
     data &= VL_MASK_I(obits);
     return data;
 }
 
-QData VL_RAND_RESET_Q(int obits) VL_MT_SAFE {
-    if (Verilated::threadContextp()->randReset() == 0) return 0;
-    QData data = ~0ULL;
-    if (Verilated::threadContextp()->randReset() != 1) {  // if 2, randomize
-        data = VL_RANDOM_Q();
-    }
-    data &= VL_MASK_Q(obits);
-    return data;
-}
-
-WDataOutP VL_RAND_RESET_W(int obits, WDataOutP outwp) VL_MT_SAFE {
-    for (int i = 0; i < VL_WORDS_I(obits) - 1; ++i) outwp[i] = VL_RAND_RESET_I(32);
-    outwp[VL_WORDS_I(obits) - 1] = VL_RAND_RESET_I(32) & VL_MASK_E(obits);
-    return outwp;
-}
 WDataOutP VL_ZERO_RESET_W(int obits, WDataOutP outwp) VL_MT_SAFE {
     // Not inlined to speed up compilation of slowpath code
     return VL_ZERO_W(obits, outwp);
diff --git a/include/verilated_funcs.h b/include/verilated_funcs.h
index 16ed18728..349fe44da 100644
--- a/include/verilated_funcs.h
+++ b/include/verilated_funcs.h
@@ -124,10 +124,6 @@ extern WDataOutP VL_SCOPED_RAND_RESET_ASSIGN_W(int obits, WDataOutP outwp, uint6
 
 /// Random reset a signal of given width (init time only)
 extern IData VL_RAND_RESET_I(int obits) VL_MT_SAFE;
-/// Random reset a signal of given width (init time only)
-extern QData VL_RAND_RESET_Q(int obits) VL_MT_SAFE;
-/// Random reset a signal of given width (init time only)
-extern WDataOutP VL_RAND_RESET_W(int obits, WDataOutP outwp) VL_MT_SAFE;
 
 /// Zero reset a signal (slow - else use VL_ZERO_W)
 extern WDataOutP VL_ZERO_RESET_W(int obits, WDataOutP outwp) VL_MT_SAFE;