Format time string using integer (verilator#2940)

Co-authored-by: Wilson Snyder <wsnyder@wsnyder.org>
tklam · May 16, 2021 · 31779b8 · 31779b8
1 parent 80d62ad
commit 31779b8
Show file tree

Hide file tree

Showing 12 changed files with 132 additions and 58 deletions.
diff --git a/include/verilated.cpp b/include/verilated.cpp
@@ -637,26 +637,80 @@ std::string VL_DECIMAL_NW(int width, WDataInP lwp) VL_MT_SAFE {
     return output;
 }
 
-std::string _vl_vsformat_time(char* tmp, double ld, bool left, size_t width) {
-    // Double may lose precision, but sc_time_stamp has similar limit
-    std::string suffix = Verilated::threadContextp()->impp()->timeFormatSuffix();
-    int userUnits = Verilated::threadContextp()->impp()->timeFormatUnits();  // 0..-15
-    int fracDigits = Verilated::threadContextp()->impp()->timeFormatPrecision();  // 0..N
-    int prec = Verilated::threadContextp()->timeprecision();  // 0..-15
-    int shift = prec - userUnits + fracDigits;  // 0..-15
-    double shiftd = vl_time_multiplier(shift);
-    double scaled = ld * shiftd;
-    const double fracDiv = vl_time_multiplier(fracDigits);
-    const double whole = scaled / fracDiv;
+template <typename T>
+std::string _vl_vsformat_time(char* tmp, T ld, int timeunit, bool left, size_t width) {
+    const VerilatedContextImp* const ctxImpp = Verilated::threadContextp()->impp();
+    const std::string suffix = ctxImpp->timeFormatSuffix();
+    const int userUnits = ctxImpp->timeFormatUnits();  // 0..-15
+    const int fracDigits = ctxImpp->timeFormatPrecision();  // 0..N
+    const int shift = -userUnits + fracDigits + timeunit;  // 0..-15
     int digits = 0;
-    if (!fracDigits) {
-        digits = VL_SNPRINTF(tmp, VL_VALUE_STRING_MAX_WIDTH, "%.0f%s", whole, suffix.c_str());
+    if (std::numeric_limits<T>::is_integer) {
+        constexpr int b = 128;
+        constexpr int w = VL_WORDS_I(b);
+        WData tmp0[w], tmp1[w], tmp2[w], tmp3[w];
+
+        WDataInP shifted = VL_EXTEND_WQ(b, 0, tmp0, ld);
+        if (shift < 0) {
+            WDataInP pow10 = VL_EXTEND_WQ(b, 0, tmp1, vl_time_pow10(-shift));
+            shifted = VL_DIV_WWW(b, tmp2, shifted, pow10);
+        } else {
+            WDataInP pow10 = VL_EXTEND_WQ(b, 0, tmp1, vl_time_pow10(shift));
+            shifted = VL_MUL_W(w, tmp2, shifted, pow10);
+        }
+
+        WDataInP fracDigitsPow10 = VL_EXTEND_WQ(b, 0, tmp3, vl_time_pow10(fracDigits));
+        WDataInP integer = VL_DIV_WWW(b, tmp0, shifted, fracDigitsPow10);
+        WDataInP frac = VL_MODDIV_WWW(b, tmp1, shifted, fracDigitsPow10);
+        WDataInP max64Bit
+            = VL_EXTEND_WQ(b, 0, tmp2, std::numeric_limits<vluint64_t>::max());  // breaks shifted
+        if (VL_GT_W(w, integer, max64Bit)) {
+            WDataOutP v = VL_ASSIGN_W(b, tmp3, integer);  // breaks fracDigitsPow10
+            WData zero[w], ten[w];
+            VL_ZERO_W(b, zero);
+            VL_EXTEND_WI(b, 0, ten, 10);
+            char buf[128];  // 128B is obviously long enough to represent 128bit integer in decimal
+            char* ptr = buf + sizeof(buf) - 1;
+            *ptr = '\0';
+            while (VL_GT_W(w, v, zero)) {
+                --ptr;
+                WDataInP mod = VL_MODDIV_WWW(b, tmp2, v, ten);  // breaks max64Bit
+                *ptr = "0123456789"[VL_SET_QW(mod)];
+                WData divided[w];
+                VL_DIV_WWW(b, divided, v, ten);
+                VL_ASSIGN_W(b, v, divided);
+            }
+            if (!fracDigits) {
+                digits = VL_SNPRINTF(tmp, VL_VALUE_STRING_MAX_WIDTH, "%s%s", ptr, suffix.c_str());
+            } else {
+                digits = VL_SNPRINTF(tmp, VL_VALUE_STRING_MAX_WIDTH, "%s.%0*" VL_PRI64 "u%s", ptr,
+                                     fracDigits, VL_SET_QW(frac), suffix.c_str());
+            }
+        } else {
+            const vluint64_t integer64 = VL_SET_QW(integer);
+            if (!fracDigits) {
+                digits = VL_SNPRINTF(tmp, VL_VALUE_STRING_MAX_WIDTH, "%" VL_PRI64 "u%s", integer64,
+                                     suffix.c_str());
+            } else {
+                digits = VL_SNPRINTF(tmp, VL_VALUE_STRING_MAX_WIDTH,
+                                     "%" VL_PRI64 "u.%0*" VL_PRI64 "u%s", integer64, fracDigits,
+                                     VL_SET_QW(frac), suffix.c_str());
+            }
+        }
     } else {
-        digits = VL_SNPRINTF(tmp, VL_VALUE_STRING_MAX_WIDTH, "%.*f%s", fracDigits, whole,
-                             suffix.c_str());
+        double shiftd = vl_time_multiplier(shift);
+        double scaled = ld * shiftd;
+        const double fracDiv = vl_time_multiplier(fracDigits);
+        const double whole = scaled / fracDiv;
+        if (!fracDigits) {
+            digits = VL_SNPRINTF(tmp, VL_VALUE_STRING_MAX_WIDTH, "%.0f%s", whole, suffix.c_str());
+        } else {
+            digits = VL_SNPRINTF(tmp, VL_VALUE_STRING_MAX_WIDTH, "%.*f%s", fracDigits, whole,
+                                 suffix.c_str());
+        }
     }
 
-    int needmore = width - digits;
+    const int needmore = width - digits;
     std::string padding;
     if (needmore > 0) padding.append(needmore, ' ');  // Pad with spaces
     return left ? (tmp + padding) : (padding + tmp);
@@ -751,7 +805,8 @@ void _vl_vsformat(std::string& output, const char* formatp, va_list ap) VL_MT_SA
                 if (lbits) {}  // UNUSED - always 64
                 if (fmt == '^') {  // Realtime
                     if (!widthSet) width = Verilated::threadContextp()->impp()->timeFormatWidth();
-                    output += _vl_vsformat_time(t_tmp, d, left, width);
+                    const int timeunit = va_arg(ap, int);
+                    output += _vl_vsformat_time(t_tmp, d, timeunit, left, width);
                 } else {
                     std::string fmts(pctp, pos - pctp + 1);
                     VL_SNPRINTF(t_tmp, VL_VALUE_STRING_MAX_WIDTH, fmts.c_str(), d);
@@ -850,7 +905,8 @@ void _vl_vsformat(std::string& output, const char* formatp, va_list ap) VL_MT_SA
                 }
                 case 't': {  // Time
                     if (!widthSet) width = Verilated::threadContextp()->impp()->timeFormatWidth();
-                    output += _vl_vsformat_time(t_tmp, static_cast<double>(ld), left, width);
+                    const int timeunit = va_arg(ap, int);
+                    output += _vl_vsformat_time(t_tmp, ld, timeunit, left, width);
                     break;
                 }
                 case 'b':
@@ -2153,6 +2209,30 @@ double vl_time_multiplier(int scale) VL_PURE {
         return pow10[scale];
     }
 }
+vluint64_t vl_time_pow10(int n) {
+    static const vluint64_t pow10[20] = {
+        1ULL,
+        10ULL,
+        100ULL,
+        1000ULL,
+        10000ULL,
+        100000ULL,
+        1000000ULL,
+        10000000ULL,
+        100000000ULL,
+        1000000000ULL,
+        10000000000ULL,
+        100000000000ULL,
+        1000000000000ULL,
+        10000000000000ULL,
+        100000000000000ULL,
+        1000000000000000ULL,
+        10000000000000000ULL,
+        100000000000000000ULL,
+        1000000000000000000ULL,
+    };
+    return pow10[n];
+}
 
 void VL_PRINTTIMESCALE(const char* namep, const char* timeunitp,
                        const VerilatedContext* contextp) VL_MT_SAFE {

diff --git a/include/verilated.h b/include/verilated.h
@@ -1187,6 +1187,8 @@ inline vluint64_t VerilatedContext::time() const VL_MT_SAFE {
 
 // Return time precision as multiplier of time units
 double vl_time_multiplier(int scale) VL_PURE;
+// Return power of 10. e.g. returns 100 if n==2
+vluint64_t vl_time_pow10(int n) VL_PURE;
 
 #ifdef VL_DEBUG
 /// Evaluate statement if Verilated::debug() enabled

diff --git a/src/V3EmitC.cpp b/src/V3EmitC.cpp
@@ -2282,6 +2282,19 @@ void EmitCStmts::displayArg(AstNode* dispp, AstNode** elistp, bool isScan, const
             emitDispState.pushArg(' ', nullptr, cvtToStr(argp->widthMin()));
         }
         emitDispState.pushArg(fmtLetter, argp, "");
+        if (fmtLetter == 't' || fmtLetter == '^') {
+            AstSFormatF* fmtp = nullptr;
+            if (AstDisplay* nodep = VN_CAST(dispp, Display))
+                fmtp = nodep->fmtp();
+            else if (AstSFormat* nodep = VN_CAST(dispp, SFormat))
+                fmtp = nodep->fmtp();
+            else
+                fmtp = VN_CAST(dispp, SFormatF);
+            UASSERT_OBJ(fmtp, dispp,
+                        "Use of %t must be under AstDisplay, AstSFormat, or AstSFormatF");
+            UASSERT_OBJ(!fmtp->timeunit().isNone(), fmtp, "timenunit must be set");
+            emitDispState.pushArg(' ', nullptr, cvtToStr((int)fmtp->timeunit().powerOfTen()));
+        }
     } else {
         emitDispState.pushArg(fmtLetter, nullptr, "");
     }

diff --git a/src/V3Width.cpp b/src/V3Width.cpp
@@ -3864,27 +3864,6 @@ class WidthVisitor final : public AstNVisitor {
                         if (nodep->timeunit().isNone()) {
                             nodep->v3fatalSrc("display %t has no time units");
                         }
-                        double scale = nodep->timeunit().multiplier()
-                                       / v3Global.rootp()->timeprecision().multiplier();
-                        if (scale != 1.0) {
-                            AstNode* newp;
-                            AstNRelinker relinkHandle;
-                            argp->unlinkFrBack(&relinkHandle);
-                            if (argp->isDouble()) {  // Convert it
-                                ch = '^';
-                                newp = new AstMulD(
-                                    argp->fileline(),
-                                    new AstConst(argp->fileline(), AstConst::RealDouble(), scale),
-                                    argp);
-                            } else {
-                                newp = new AstMul(argp->fileline(),
-                                                  new AstConst(argp->fileline(),
-                                                               AstConst::Unsized64(),
-                                                               std::llround(scale)),
-                                                  argp);
-                            }
-                            relinkHandle.relink(newp);
-                        }
                         argp = nextp;
                     }
                     break;

diff --git a/test_regress/t/t_time_vpi_100s10ms.out b/test_regress/t/t_time_vpi_100s10ms.out
@@ -3,11 +3,11 @@ Time scale of t is 100s / 10ms
 [100000000] time%0d=10000  123%0t=1230000
   dig%0t=0 dig%0d=0
   rdig%0t=543 rdig%0f=0.054321
-  acc%0t=11177671081359484928 acc%0d=12345678901234567890
+  acc%0t=123456789012345678900000 acc%0d=12345678901234567890
 [1000000000000000.000000ns] time%0d=10000  123%0t=12300000000000.000000ns
   dig%0t=0.000000ns dig%0d=0
   rdig%0t=5432109876.543210ns rdig%0f=0.054321
-  acc%0t=111776710813594856498135040.000000ns acc%0d=12345678901234567890
+  acc%0t=1234567890123456789000000000000.000000ns acc%0d=12345678901234567890
 [1000000000000000.000000ns] stime%0t=1000000000000000.000000ns  stime%0d=10000  stime%0f=10000.000000
 [1000000000000000.000000ns] rtime%0t=1000000000000000.000000ns  rtime%0d=10000  rtime%0f=10000.000000
 global vpiSimTime = 0,100000000  vpiScaledRealTime = 1e+08

diff --git a/test_regress/t/t_time_vpi_10ms10ns.out b/test_regress/t/t_time_vpi_10ms10ns.out
@@ -3,11 +3,11 @@ Time scale of t is 10ms / 10ns
 [60000000] time%0d=60  123%0t=123000000
   dig%0t=543000000 dig%0d=543
   rdig%0t=543210988 rdig%0f=543.210988
-  acc%0t=10962463713375475712 acc%0d=12345678901234567890
+  acc%0t=12345678901234567890000000 acc%0d=12345678901234567890
 [600000000.000000ns] time%0d=60  123%0t=1230000000.000000ns
   dig%0t=5430000000.000000ns dig%0d=543
   rdig%0t=5432109876.543210ns rdig%0f=543.210988
-  acc%0t=109624637133754744832.000000ns acc%0d=12345678901234567890
+  acc%0t=123456789012345678900000000.000000ns acc%0d=12345678901234567890
 [600000000.000000ns] stime%0t=600000000.000000ns  stime%0d=60  stime%0f=60.000000
 [600000000.000000ns] rtime%0t=600000000.000000ns  rtime%0d=60  rtime%0f=60.000000
 global vpiSimTime = 0,60000000  vpiScaledRealTime = 6e+07

diff --git a/test_regress/t/t_time_vpi_1fs1fs.out b/test_regress/t/t_time_vpi_1fs1fs.out
@@ -3,11 +3,11 @@ Time scale of t is 1fs / 1fs
 [60] time%0d=60  123%0t=123
   dig%0t=5432109876543210 dig%0d=5432109876543210
   rdig%0t=5432109876543210 rdig%0f=5432109876543210.000000
-  acc%0t=12345678901234567168 acc%0d=12345678901234567890
+  acc%0t=12345678901234567890 acc%0d=12345678901234567890
 [0.000060ns] time%0d=60  123%0t=0.000123ns
   dig%0t=5432109876.543210ns dig%0d=5432109876543210
   rdig%0t=5432109876.543210ns rdig%0f=5432109876543210.000000
-  acc%0t=12345678901234.566406ns acc%0d=12345678901234567890
+  acc%0t=12345678901234.567890ns acc%0d=12345678901234567890
 [0.000060ns] stime%0t=0.000060ns  stime%0d=60  stime%0f=60.000000
 [0.000060ns] rtime%0t=0.000060ns  rtime%0d=60  rtime%0f=60.000000
 global vpiSimTime = 0,60  vpiScaledRealTime = 60

diff --git a/test_regress/t/t_time_vpi_1ms10ns.out b/test_regress/t/t_time_vpi_1ms10ns.out
@@ -3,11 +3,11 @@ Time scale of t is 1ms / 10ns
 [6000000] time%0d=60  123%0t=12300000
   dig%0t=543200000 dig%0d=5432
   rdig%0t=543210988 rdig%0f=5432.109877
-  acc%0t=1096246371337547520 acc%0d=12345678901234567890
+  acc%0t=1234567890123456789000000 acc%0d=12345678901234567890
 [60000000.000000ns] time%0d=60  123%0t=123000000.000000ns
   dig%0t=5432000000.000000ns dig%0d=5432
   rdig%0t=5432109876.543210ns rdig%0f=5432.109877
-  acc%0t=10962463713375473664.000000ns acc%0d=12345678901234567890
+  acc%0t=12345678901234567890000000.000000ns acc%0d=12345678901234567890
 [60000000.000000ns] stime%0t=60000000.000000ns  stime%0d=60  stime%0f=60.000000
 [60000000.000000ns] rtime%0t=60000000.000000ns  rtime%0d=60  rtime%0f=60.000000
 global vpiSimTime = 0,6000000  vpiScaledRealTime = 6e+06

diff --git a/test_regress/t/t_time_vpi_1ns1ns.out b/test_regress/t/t_time_vpi_1ns1ns.out
@@ -3,11 +3,11 @@ Time scale of t is 1ns / 1ns
 [60] time%0d=60  123%0t=123
   dig%0t=5432109877 dig%0d=5432109877
   rdig%0t=5432109877 rdig%0f=5432109876.543210
-  acc%0t=12345678901234567168 acc%0d=12345678901234567890
+  acc%0t=12345678901234567890 acc%0d=12345678901234567890
 [60.000000ns] time%0d=60  123%0t=123.000000ns
   dig%0t=5432109877.000000ns dig%0d=5432109877
   rdig%0t=5432109876.543210ns rdig%0f=5432109876.543210
-  acc%0t=12345678901234565120.000000ns acc%0d=12345678901234567890
+  acc%0t=12345678901234567890.000000ns acc%0d=12345678901234567890
 [60.000000ns] stime%0t=60.000000ns  stime%0d=60  stime%0f=60.000000
 [60.000000ns] rtime%0t=60.000000ns  rtime%0d=60  rtime%0f=60.000000
 global vpiSimTime = 0,60  vpiScaledRealTime = 60

diff --git a/test_regress/t/t_time_vpi_1ps1fs.out b/test_regress/t/t_time_vpi_1ps1fs.out
@@ -2,12 +2,12 @@
 Time scale of t is 1ps / 1fs
 [60000] time%0d=60  123%0t=123000
   dig%0t=5432109876543000 dig%0d=5432109876543
-  rdig%0t=5432109876543209 rdig%0f=5432109876543.209961
-  acc%0t=4807115922877858816 acc%0d=12345678901234567890
+  rdig%0t=5432109876543210 rdig%0f=5432109876543.209961
+  acc%0t=12345678901234567890000 acc%0d=12345678901234567890
 [0.060000ns] time%0d=60  123%0t=0.123000ns
   dig%0t=5432109876.543000ns dig%0d=5432109876543
-  rdig%0t=5432109876.543209ns rdig%0f=5432109876543.209961
-  acc%0t=4807115922877.858398ns acc%0d=12345678901234567890
+  rdig%0t=5432109876.543210ns rdig%0f=5432109876543.209961
+  acc%0t=12345678901234567.890000ns acc%0d=12345678901234567890
 [0.060000ns] stime%0t=0.060000ns  stime%0d=60  stime%0f=60.000000
 [0.060000ns] rtime%0t=0.060000ns  rtime%0d=60  rtime%0f=60.000000
 global vpiSimTime = 0,60000  vpiScaledRealTime = 60000

diff --git a/test_regress/t/t_time_vpi_1s10ns.out b/test_regress/t/t_time_vpi_1s10ns.out
@@ -3,11 +3,11 @@ Time scale of t is 1s / 10ns
 [6000000000] time%0d=60  123%0t=12300000000
   dig%0t=500000000 dig%0d=5
   rdig%0t=543210988 rdig%0f=5.432110
-  acc%0t=7888470988684038144 acc%0d=12345678901234567890
+  acc%0t=1234567890123456789000000000 acc%0d=12345678901234567890
 [60000000000.000000ns] time%0d=60  123%0t=123000000000.000000ns
   dig%0t=5000000000.000000ns dig%0d=5
   rdig%0t=5432109876.543210ns rdig%0f=5.432110
-  acc%0t=78884709886840389632.000000ns acc%0d=12345678901234567890
+  acc%0t=12345678901234567890000000000.000000ns acc%0d=12345678901234567890
 [60000000000.000000ns] stime%0t=60000000000.000000ns  stime%0d=60  stime%0f=60.000000
 [60000000000.000000ns] rtime%0t=60000000000.000000ns  rtime%0d=60  rtime%0f=60.000000
 global vpiSimTime = 1,1705032704  vpiScaledRealTime = 6e+09

diff --git a/test_regress/t/t_time_vpi_1us1ns.out b/test_regress/t/t_time_vpi_1us1ns.out
@@ -3,11 +3,11 @@ Time scale of t is 1us / 1ns
 [60000] time%0d=60  123%0t=123000
   dig%0t=5432110000 dig%0d=5432110
   rdig%0t=5432109877 rdig%0f=5432109.876543
-  acc%0t=4807115922877858816 acc%0d=12345678901234567890
+  acc%0t=12345678901234567890000 acc%0d=12345678901234567890
 [60000.000000ns] time%0d=60  123%0t=123000.000000ns
   dig%0t=5432110000.000000ns dig%0d=5432110
-  rdig%0t=5432109876.543209ns rdig%0f=5432109.876543
-  acc%0t=4807115922877858816.000000ns acc%0d=12345678901234567890
+  rdig%0t=5432109876.543210ns rdig%0f=5432109.876543
+  acc%0t=12345678901234567890000.000000ns acc%0d=12345678901234567890
 [60000.000000ns] stime%0t=60000.000000ns  stime%0d=60  stime%0f=60.000000
 [60000.000000ns] rtime%0t=60000.000000ns  rtime%0d=60  rtime%0f=60.000000
 global vpiSimTime = 0,60000  vpiScaledRealTime = 60000