perfaware/part3: Complete 61 Page faults (aggregate faults in each test)

part3/base.h

@@ -19,9 +19,10 @@
 #define MIN(a, b) ((a) < (b) ? (a) : (b))
 #define MAX(a, b) ((a) > (b) ? (a) : (b))
 
-typedef float f32;
+typedef float    f32;
-typedef double f64;
+typedef double   f64;
-typedef uint8_t u8;
+typedef bool     b32;
+typedef uint8_t  u8;
 typedef uint16_t u16;
 typedef uint32_t u32;
 typedef uint64_t u64;

part3/listing_0065_haversine_formula.cpp

@@ -0,0 +1,53 @@
+/* ========================================================================
+
+   (C) Copyright 2023 by Molly Rocket, Inc., All Rights Reserved.
+   
+   This software is provided 'as-is', without any express or implied
+   warranty. In no event will the authors be held liable for any damages
+   arising from the use of this software.
+   
+   Please see https://computerenhance.com for more information
+   
+   ======================================================================== */
+
+/* ========================================================================
+   LISTING 65
+   ======================================================================== */
+
+static f64 Square(f64 A)
+{
+    f64 Result = (A*A);
+    return Result;
+}
+
+static f64 RadiansFromDegrees(f64 Degrees)
+{
+    f64 Result = 0.01745329251994329577 * Degrees;
+    return Result;
+}
+
+// NOTE(casey): EarthRadius is generally expected to be 6372.8
+static f64 ReferenceHaversine(f64 X0, f64 Y0, f64 X1, f64 Y1, f64 EarthRadius)
+{
+    /* NOTE(casey): This is not meant to be a "good" way to calculate the Haversine distance.
+       Instead, it attempts to follow, as closely as possible, the formula used in the real-world
+       question on which these homework exercises are loosely based.
+    */
+    
+    f64 lat1 = Y0;
+    f64 lat2 = Y1;
+    f64 lon1 = X0;
+    f64 lon2 = X1;
+    
+    f64 dLat = RadiansFromDegrees(lat2 - lat1);
+    f64 dLon = RadiansFromDegrees(lon2 - lon1);
+    lat1 = RadiansFromDegrees(lat1);
+    lat2 = RadiansFromDegrees(lat2);
+    
+    f64 a = Square(sin(dLat/2.0)) + cos(lat1)*cos(lat2)*Square(sin(dLon/2));
+    f64 c = 2.0*asin(sqrt(a));
+    
+    f64 Result = EarthRadius * c;
+    
+    return Result;
+}

part3/listing_0068_buffer.cpp

@@ -0,0 +1,72 @@
+/* ========================================================================
+
+   (C) Copyright 2023 by Molly Rocket, Inc., All Rights Reserved.
+   
+   This software is provided 'as-is', without any express or implied
+   warranty. In no event will the authors be held liable for any damages
+   arising from the use of this software.
+   
+   Please see https://computerenhance.com for more information
+   
+   ======================================================================== */
+
+/* ========================================================================
+   LISTING 68
+   ======================================================================== */
+
+struct buffer
+{
+    size_t Count;
+    u8 *Data;
+};
+
+#define CONSTANT_STRING(String) {sizeof(String) - 1, (u8 *)(String)}
+
+static b32 IsInBounds(buffer Source, u64 At)
+{
+    b32 Result = (At < Source.Count);
+    return Result;
+}
+
+static b32 AreEqual(buffer A, buffer B)
+{
+    if(A.Count != B.Count)
+    {
+        return false;
+    }
+    
+    for(u64 Index = 0; Index < A.Count; ++Index)
+    {
+        if(A.Data[Index] != B.Data[Index])
+        {
+            return false;
+        }
+    }
+    
+    return true;
+}
+
+static buffer AllocateBuffer(size_t Count)
+{
+    buffer Result = {};
+    Result.Data = (u8 *)malloc(Count);
+    if(Result.Data)
+    {
+        Result.Count = Count;
+    }
+    else
+    {
+        fprintf(stderr, "ERROR: Unable to allocate %llu bytes.\n", Count);
+    }
+    
+    return Result;
+}
+
+static void FreeBuffer(buffer *Buffer)
+{
+    if(Buffer->Data)
+    {
+        free(Buffer->Data);
+    }
+    *Buffer = {};
+}

part3/listing_0094_profiled_lookup_json_parser.cpp

@@ -0,0 +1,516 @@
+/* ========================================================================
+
+   (C) Copyright 2023 by Molly Rocket, Inc., All Rights Reserved.
+   
+   This software is provided 'as-is', without any express or implied
+   warranty. In no event will the authors be held liable for any damages
+   arising from the use of this software.
+   
+   Please see https://computerenhance.com for more information
+   
+   ======================================================================== */
+
+/* ========================================================================
+   LISTING 94
+   ======================================================================== */
+
+enum json_token_type
+{
+    Token_end_of_stream,
+    Token_error,
+    
+    Token_open_brace,
+    Token_open_bracket,
+    Token_close_brace,
+    Token_close_bracket,
+    Token_comma,
+    Token_colon,
+    Token_string_literal,
+    Token_number,
+    Token_true,
+    Token_false,
+    Token_null,
+    
+    Token_count,
+};
+
+struct json_token
+{
+    json_token_type Type;
+    buffer Value;
+};
+
+struct json_element
+{
+    buffer Label;
+    buffer Value;
+    json_element *FirstSubElement;
+    
+    json_element *NextSibling;
+};
+
+struct json_parser
+{
+    buffer Source;
+    u64 At;
+    b32 HadError;
+};
+
+static b32 IsJSONDigit(buffer Source, u64 At)
+{
+    b32 Result = false;
+    if(IsInBounds(Source, At))
+    {
+        u8 Val = Source.Data[At];
+        Result = ((Val >= '0') && (Val <= '9'));
+    }
+    
+    return Result;
+}
+
+static b32 IsJSONWhitespace(buffer Source, u64 At)
+{
+    b32 Result = false;
+    if(IsInBounds(Source, At))
+    {
+        u8 Val = Source.Data[At];
+        Result = ((Val == ' ') || (Val == '\t') || (Val == '\n') || (Val == '\r'));
+    }
+    
+    return Result;
+}
+
+static b32 IsParsing(json_parser *Parser)
+{
+    b32 Result = !Parser->HadError && IsInBounds(Parser->Source, Parser->At);
+    return Result;
+}
+
+static void Error(json_parser *Parser, json_token Token, char const *Message)
+{
+    Parser->HadError = true;
+    fprintf(stderr, "ERROR: \"%.*s\" - %s\n", (u32)Token.Value.Count, (char *)Token.Value.Data, Message);
+}
+
+static void ParseKeyword(buffer Source, u64 *At, buffer KeywordRemaining, json_token_type Type, json_token *Result)
+{
+    if((Source.Count - *At) >= KeywordRemaining.Count)
+    {
+        buffer Check = Source;
+        Check.Data += *At;
+        Check.Count = KeywordRemaining.Count;
+        if(AreEqual(Check, KeywordRemaining))
+        {
+            Result->Type = Type;
+            Result->Value.Count += KeywordRemaining.Count;
+            *At += KeywordRemaining.Count;
+        }
+    }
+}
+
+static json_token GetJSONToken(json_parser *Parser)
+{
+    json_token Result = {};
+    
+    buffer Source = Parser->Source;
+    u64 At = Parser->At;
+    
+    while(IsJSONWhitespace(Source, At))
+    {
+        ++At;
+    }
+    
+    if(IsInBounds(Source, At))
+    {
+        Result.Type = Token_error;
+        Result.Value.Count = 1;
+        Result.Value.Data = Source.Data + At;
+        u8 Val = Source.Data[At++];
+        switch(Val)
+        {
+            case '{': {Result.Type = Token_open_brace;} break;
+            case '[': {Result.Type = Token_open_bracket;} break;
+            case '}': {Result.Type = Token_close_brace;} break;
+            case ']': {Result.Type = Token_close_bracket;} break;
+            case ',': {Result.Type = Token_comma;} break;
+            case ':': {Result.Type = Token_colon;} break;
+
+            case 'f':
+            {
+                ParseKeyword(Source, &At, CONSTANT_STRING("alse"), Token_false, &Result);
+            } break;
+            
+            case 'n':
+            {
+                ParseKeyword(Source, &At, CONSTANT_STRING("ull"), Token_null, &Result);
+            } break;
+            
+            case 't':
+            {
+                ParseKeyword(Source, &At, CONSTANT_STRING("rue"), Token_true, &Result);
+            } break;
+            
+            case '"':
+            {
+                Result.Type = Token_string_literal;
+                
+                u64 StringStart = At;
+                
+                while(IsInBounds(Source, At) && (Source.Data[At] != '"'))
+                {
+                    if(IsInBounds(Source, (At + 1)) &&
+                       (Source.Data[At] == '\\') &&
+                       (Source.Data[At + 1] == '"'))
+                    {
+                        // NOTE(casey): Skip escaped quotation marks
+                        ++At;
+                    }
+                    
+                    ++At;
+                }
+                
+                Result.Value.Data = Source.Data + StringStart;
+                Result.Value.Count = At - StringStart;
+                if(IsInBounds(Source, At))
+                {
+                    ++At;
+                }
+            } break;
+
+            case '-':
+            case '0':
+            case '1':
+            case '2':
+            case '3':
+            case '4':
+            case '5':
+            case '6':
+            case '7':
+            case '8':
+            case '9':
+            {
+                u64 Start = At - 1;
+                Result.Type = Token_number;
+
+                // NOTE(casey): Move past a leading negative sign if one exists
+                if((Val == '-') && IsInBounds(Source, At))
+                {
+                    Val = Source.Data[At++];
+                }
+                
+                // NOTE(casey): If the leading digit wasn't 0, parse any digits before the decimal point
+                if(Val != '0')
+                {
+                    while(IsJSONDigit(Source, At))
+                    {
+                        ++At;
+                    }
+                }
+                
+                // NOTE(casey): If there is a decimal point, parse any digits after the decimal point
+                if(IsInBounds(Source, At) && (Source.Data[At] == '.'))
+                {
+                    ++At;
+                    while(IsJSONDigit(Source, At))
+                    {
+                        ++At;
+                    }
+                }
+                
+                // NOTE(casey): If it's in scientific notation, parse any digits after the "e"
+                if(IsInBounds(Source, At) && ((Source.Data[At] == 'e') || (Source.Data[At] == 'E')))
+                {
+                    ++At;
+                    
+                    if(IsInBounds(Source, At) && ((Source.Data[At] == '+') || (Source.Data[At] == '-')))
+                    {
+                        ++At;
+                    }
+                    
+                    while(IsJSONDigit(Source, At))
+                    {
+                        ++At;
+                    }
+                }
+                
+                Result.Value.Count = At - Start;
+            } break;
+            
+            default:
+            {
+            } break;
+        }
+    }
+    
+    Parser->At = At;
+    
+    return Result;
+}
+
+static json_element *ParseJSONList(json_parser *Parser, json_token_type EndType, b32 HasLabels);
+static json_element *ParseJSONElement(json_parser *Parser, buffer Label, json_token Value)
+{
+    b32 Valid = true;
+    
+    json_element *SubElement = 0;
+    if(Value.Type == Token_open_bracket)
+    {
+        SubElement = ParseJSONList(Parser, Token_close_bracket, false);
+    }
+    else if(Value.Type == Token_open_brace)
+    {
+        SubElement = ParseJSONList(Parser, Token_close_brace, true);
+    }
+    else if((Value.Type == Token_string_literal) ||
+            (Value.Type == Token_true) ||
+            (Value.Type == Token_false) ||
+            (Value.Type == Token_null) ||
+            (Value.Type == Token_number))
+    {
+        // NOTE(casey): Nothing to do here, since there is no additional data
+    }
+    else
+    {
+        Valid = false;
+    }
+    
+    json_element *Result = 0;
+    
+    if(Valid)
+    {
+        Result = (json_element *)malloc(sizeof(json_element));
+        Result->Label = Label;
+        Result->Value = Value.Value;
+        Result->FirstSubElement = SubElement;
+        Result->NextSibling = 0;
+    }
+    
+    return Result;
+}
+
+static json_element *ParseJSONList(json_parser *Parser, json_token_type EndType, b32 HasLabels)
+{
+    json_element *FirstElement = {};
+    json_element *LastElement = {};
+    
+    while(IsParsing(Parser))
+    {
+        buffer Label = {};
+        json_token Value = GetJSONToken(Parser);
+        if(HasLabels)
+        {
+            if(Value.Type == Token_string_literal)
+            {
+                Label = Value.Value;
+                
+                json_token Colon = GetJSONToken(Parser);
+                if(Colon.Type == Token_colon)
+                {
+                    Value = GetJSONToken(Parser);
+                }
+                else
+                {
+                    Error(Parser, Colon, "Expected colon after field name");
+                }
+            }
+            else if(Value.Type != EndType)
+            {
+                Error(Parser, Value, "Unexpected token in JSON");
+            }
+        }
+        
+        json_element *Element = ParseJSONElement(Parser, Label, Value);
+        if(Element)
+        {
+            LastElement = (LastElement ? LastElement->NextSibling : FirstElement) = Element;
+        }
+        else if(Value.Type == EndType)
+        {
+            break;
+        }
+        else
+        {
+            Error(Parser, Value, "Unexpected token in JSON");
+        }
+        
+        json_token Comma = GetJSONToken(Parser);
+        if(Comma.Type == EndType)
+        {
+            break;
+        }
+        else if(Comma.Type != Token_comma)
+        {
+            Error(Parser, Comma, "Unexpected token in JSON");
+        }
+    }
+    
+    return FirstElement;
+}
+
+static json_element *ParseJSON(buffer InputJSON)
+{
+    TimeFunction;
+    
+    json_parser Parser = {};
+    Parser.Source = InputJSON;
+    
+    json_element *Result = ParseJSONElement(&Parser, {}, GetJSONToken(&Parser));
+    return Result;
+}
+
+static void FreeJSON(json_element *Element)
+{
+    while(Element)
+    {
+        json_element *FreeElement = Element;
+        Element = Element->NextSibling;
+    
+        FreeJSON(FreeElement->FirstSubElement);
+        free(FreeElement);
+    }
+}
+
+static json_element *LookupElement(json_element *Object, buffer ElementName)
+{
+    json_element *Result = 0;
+    
+    if(Object)
+    {
+        for(json_element *Search = Object->FirstSubElement; Search; Search = Search->NextSibling)
+        {
+            if(AreEqual(Search->Label, ElementName))
+            {
+                Result = Search;
+                break;
+            }
+        }
+    }
+    
+    return Result;
+}
+
+static f64 ConvertJSONSign(buffer Source, u64 *AtResult)
+{
+    u64 At = *AtResult;
+
+    f64 Result = 1.0;
+    if(IsInBounds(Source, At) && (Source.Data[At] == '-'))
+    {
+        Result = -1.0;
+        ++At;
+    }
+    
+    *AtResult = At;
+
+    return Result;
+}
+
+static f64 ConvertJSONNumber(buffer Source, u64 *AtResult)
+{
+    u64 At = *AtResult;
+    
+    f64 Result = 0.0;
+    while(IsInBounds(Source, At))
+    {
+        u8 Char = Source.Data[At] - (u8)'0';
+        if(Char < 10)
+        {
+            Result = 10.0*Result + (f64)Char;
+            ++At;
+        }
+        else
+        {
+            break;
+        }
+    }
+    
+    *AtResult = At;
+    
+    return Result;
+}
+
+static f64 ConvertElementToF64(json_element *Object, buffer ElementName)
+{
+    f64 Result = 0.0;
+    
+    json_element *Element = LookupElement(Object, ElementName);
+    if(Element)
+    {
+        buffer Source = Element->Value;
+        u64 At = 0;
+        
+        f64 Sign = ConvertJSONSign(Source, &At);
+        f64 Number = ConvertJSONNumber(Source, &At);
+        
+        if(IsInBounds(Source, At) && (Source.Data[At] == '.'))
+        {
+            ++At;
+            f64 C = 1.0 / 10.0;
+            while(IsInBounds(Source, At))
+            {
+                u8 Char = Source.Data[At] - (u8)'0';
+                if(Char < 10)
+                {
+                    Number = Number + C*(f64)Char;
+                    C *= 1.0 / 10.0;
+                    ++At;
+                }
+                else
+                {
+                    break;
+                }
+            }
+        }
+        
+        if(IsInBounds(Source, At) && ((Source.Data[At] == 'e') || (Source.Data[At] == 'E')))
+        {
+            ++At;
+            if(IsInBounds(Source, At) && (Source.Data[At] == '+'))
+            {
+                ++At;
+            }
+
+            f64 ExponentSign = ConvertJSONSign(Source, &At);
+            f64 Exponent = ExponentSign*ConvertJSONNumber(Source, &At);
+            Number *= pow(10.0, Exponent);
+        }
+        
+        Result = Sign*Number;
+    }
+    
+    return Result;
+}
+
+static u64 ParseHaversinePairs(buffer InputJSON, u64 MaxPairCount, haversine_pair *Pairs)
+{
+    TimeFunction;
+    
+    u64 PairCount = 0;
+    
+    json_element *JSON = ParseJSON(InputJSON);
+    
+    json_element *PairsArray = LookupElement(JSON, CONSTANT_STRING("pairs"));
+    if(PairsArray)
+    {
+        TimeBlock("Lookup and Convert");
+        
+        for(json_element *Element = PairsArray->FirstSubElement;
+            Element && (PairCount < MaxPairCount);
+            Element = Element->NextSibling)
+        {
+            haversine_pair *Pair = Pairs + PairCount++;
+            
+            Pair->X0 = ConvertElementToF64(Element, CONSTANT_STRING("x0"));
+            Pair->Y0 = ConvertElementToF64(Element, CONSTANT_STRING("y0"));
+            Pair->X1 = ConvertElementToF64(Element, CONSTANT_STRING("x1"));
+            Pair->Y1 = ConvertElementToF64(Element, CONSTANT_STRING("y1"));
+        }
+    }
+   
+    {
+        TimeBlock("FreeJSON");
+        FreeJSON(JSON);
+    }
+    
+    return PairCount;
+}

part3/listing_0100_bandwidth_profiler.cpp

@@ -0,0 +1,194 @@
+/* ========================================================================
+
+   (C) Copyright 2023 by Molly Rocket, Inc., All Rights Reserved.
+   
+   This software is provided 'as-is', without any express or implied
+   warranty. In no event will the authors be held liable for any damages
+   arising from the use of this software.
+   
+   Please see https://computerenhance.com for more information
+   
+   ======================================================================== */
+
+/* ========================================================================
+   LISTING 100
+   ======================================================================== */
+
+#include "listing_0074_platform_metrics.cpp"
+
+#ifndef PROFILER
+#define PROFILER 0
+#endif
+
+#ifndef READ_BLOCK_TIMER
+#define READ_BLOCK_TIMER ReadCPUTimer
+#endif
+
+#if PROFILER
+
+struct profile_anchor
+{
+    u64 TSCElapsedExclusive; // NOTE(casey): Does NOT include children
+    u64 TSCElapsedInclusive; // NOTE(casey): DOES include children
+    u64 HitCount;
+    u64 ProcessedByteCount;
+    char const *Label;
+};
+static profile_anchor GlobalProfilerAnchors[4096];
+static u32 GlobalProfilerParent;
+
+struct profile_block
+{
+    profile_block(char const *Label_, u32 AnchorIndex_, u64 ByteCount)
+    {
+        ParentIndex = GlobalProfilerParent;
+        
+        AnchorIndex = AnchorIndex_;
+        Label = Label_;
+
+        profile_anchor *Anchor = GlobalProfilerAnchors + AnchorIndex;
+        OldTSCElapsedInclusive = Anchor->TSCElapsedInclusive;
+        Anchor->ProcessedByteCount += ByteCount;
+        
+        GlobalProfilerParent = AnchorIndex;
+        StartTSC = READ_BLOCK_TIMER();
+    }
+    
+    ~profile_block(void)
+    {
+        u64 Elapsed = READ_BLOCK_TIMER() - StartTSC;
+        GlobalProfilerParent = ParentIndex;
+    
+        profile_anchor *Parent = GlobalProfilerAnchors + ParentIndex;
+        profile_anchor *Anchor = GlobalProfilerAnchors + AnchorIndex;
+        
+        Parent->TSCElapsedExclusive -= Elapsed;
+        Anchor->TSCElapsedExclusive += Elapsed;
+        Anchor->TSCElapsedInclusive = OldTSCElapsedInclusive + Elapsed;
+        ++Anchor->HitCount;
+        
+        /* NOTE(casey): This write happens every time solely because there is no
+           straightforward way in C++ to have the same ease-of-use. In a better programming
+           language, it would be simple to have the anchor points gathered and labeled at compile
+           time, and this repetative write would be eliminated. */
+        Anchor->Label = Label;
+    }
+    
+    char const *Label;
+    u64 OldTSCElapsedInclusive;
+    u64 StartTSC;
+    u32 ParentIndex;
+    u32 AnchorIndex;
+};
+
+#define NameConcat2(A, B) A##B
+#define NameConcat(A, B) NameConcat2(A, B)
+#define TimeBandwidth(Name, ByteCount) profile_block NameConcat(Block, __LINE__)(Name, __COUNTER__ + 1, ByteCount)
+#define ProfilerEndOfCompilationUnit static_assert(__COUNTER__ < ArrayCount(GlobalProfilerAnchors), "Number of profile points exceeds size of profiler::Anchors array")
+
+static void PrintTimeElapsed(u64 TotalTSCElapsed, u64 TimerFreq, profile_anchor *Anchor)
+{
+    f64 Percent = 100.0 * ((f64)Anchor->TSCElapsedExclusive / (f64)TotalTSCElapsed);
+    printf("  %s[%llu]: %llu (%.2f%%", Anchor->Label, Anchor->HitCount, Anchor->TSCElapsedExclusive, Percent);
+    if(Anchor->TSCElapsedInclusive != Anchor->TSCElapsedExclusive)
+    {
+        f64 PercentWithChildren = 100.0 * ((f64)Anchor->TSCElapsedInclusive / (f64)TotalTSCElapsed);
+        printf(", %.2f%% w/children", PercentWithChildren);
+    }
+    printf(")");
+    
+    if(Anchor->ProcessedByteCount)
+    {
+        f64 Megabyte = 1024.0f*1024.0f;
+        f64 Gigabyte = Megabyte*1024.0f;
+        
+        f64 Seconds = (f64)Anchor->TSCElapsedInclusive / (f64)TimerFreq;
+        f64 BytesPerSecond = (f64)Anchor->ProcessedByteCount / Seconds;
+        f64 Megabytes = (f64)Anchor->ProcessedByteCount / (f64)Megabyte;
+        f64 GigabytesPerSecond = BytesPerSecond / Gigabyte;
+        
+        printf("  %.3fmb at %.2fgb/s", Megabytes, GigabytesPerSecond);
+    }
+    
+    printf("\n");
+}
+
+static void PrintAnchorData(u64 TotalCPUElapsed, u64 TimerFreq)
+{
+    for(u32 AnchorIndex = 0; AnchorIndex < ArrayCount(GlobalProfilerAnchors); ++AnchorIndex)
+    {
+        profile_anchor *Anchor = GlobalProfilerAnchors + AnchorIndex;
+        if(Anchor->TSCElapsedInclusive)
+        {
+            PrintTimeElapsed(TotalCPUElapsed, TimerFreq, Anchor);
+        }
+    }
+}
+
+#else
+
+#define TimeBandwidth(...)
+#define PrintAnchorData(...)
+#define ProfilerEndOfCompilationUnit
+
+#endif
+
+struct profiler
+{
+    u64 StartTSC;
+    u64 EndTSC;
+};
+static profiler GlobalProfiler;
+
+#define TimeBlock(Name) TimeBandwidth(Name, 0)
+#define TimeFunction TimeBlock(__func__)
+
+static u64 EstimateBlockTimerFreq(void)
+{
+    (void)&EstimateCPUTimerFreq; // NOTE(casey): This has to be voided here to prevent compilers from warning us that it is not used
+    
+	u64 MillisecondsToWait = 100;
+	u64 OSFreq = GetOSTimerFreq();
+
+	u64 BlockStart = READ_BLOCK_TIMER();
+	u64 OSStart = ReadOSTimer();
+	u64 OSEnd = 0;
+	u64 OSElapsed = 0;
+	u64 OSWaitTime = OSFreq * MillisecondsToWait / 1000;
+	while(OSElapsed < OSWaitTime)
+	{
+		OSEnd = ReadOSTimer();
+		OSElapsed = OSEnd - OSStart;
+	}
+	
+	u64 BlockEnd = READ_BLOCK_TIMER();
+	u64 BlockElapsed = BlockEnd - BlockStart;
+	
+	u64 BlockFreq = 0;
+	if(OSElapsed)
+	{
+		BlockFreq = OSFreq * BlockElapsed / OSElapsed;
+	}
+	
+	return BlockFreq;
+}
+
+static void BeginProfile(void)
+{
+    GlobalProfiler.StartTSC = READ_BLOCK_TIMER();
+}
+
+static void EndAndPrintProfile()
+{
+    GlobalProfiler.EndTSC = READ_BLOCK_TIMER();
+    u64 TimerFreq = EstimateBlockTimerFreq();
+    
+    u64 TotalTSCElapsed = GlobalProfiler.EndTSC - GlobalProfiler.StartTSC;
+    
+    if(TimerFreq)
+    {
+        printf("\nTotal time: %0.4fms (timer freq %llu)\n", 1000.0 * (f64)TotalTSCElapsed / (f64)TimerFreq, TimerFreq);
+    }
+    
+    PrintAnchorData(TotalTSCElapsed, TimerFreq);
+}

part3/listing_0101_read_bandwidth_main.cpp

@@ -0,0 +1,186 @@
+/* ========================================================================
+
+   (C) Copyright 2023 by Molly Rocket, Inc., All Rights Reserved.
+   
+   This software is provided 'as-is', without any express or implied
+   warranty. In no event will the authors be held liable for any damages
+   arising from the use of this software.
+   
+   Please see https://computerenhance.com for more information
+   
+   ======================================================================== */
+
+/* ========================================================================
+   LISTING 101
+   ======================================================================== */
+
+/* NOTE(casey): _CRT_SECURE_NO_WARNINGS is here because otherwise we cannot
+   call fopen(). If we replace fopen() with fopen_s() to avoid the warning,
+   then the code doesn't compile on Linux anymore, since fopen_s() does not
+   exist there.
+   
+   What exactly the CRT maintainers were thinking when they made this choice,
+   I have no idea. */
+#define _CRT_SECURE_NO_WARNINGS
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <stdint.h>
+#include <math.h>
+#include <sys/stat.h>
+
+typedef uint8_t u8;
+typedef uint32_t u32;
+typedef uint64_t u64;
+
+typedef int32_t b32;
+
+typedef float f32;
+typedef double f64;
+
+#define ArrayCount(Array) (sizeof(Array)/sizeof((Array)[0]))
+
+struct haversine_pair
+{
+    f64 X0, Y0;
+    f64 X1, Y1;
+};
+
+#define PROFILER 1
+#include "listing_0100_bandwidth_profiler.cpp"
+#include "listing_0065_haversine_formula.cpp"
+#include "listing_0068_buffer.cpp"
+#include "listing_0094_profiled_lookup_json_parser.cpp"
+
+static buffer ReadEntireFile(char *FileName)
+{
+    TimeFunction;
+    
+    buffer Result = {};
+        
+    FILE *File = fopen(FileName, "rb");
+    if(File)
+    {
+#if _WIN32
+        struct __stat64 Stat;
+        _stat64(FileName, &Stat);
+#else
+        struct stat Stat;
+        stat(FileName, &Stat);
+#endif
+        
+        Result = AllocateBuffer(Stat.st_size);
+        if(Result.Data)
+        {
+            TimeBandwidth("fread", Result.Count);
+            if(fread(Result.Data, Result.Count, 1, File) != 1)
+            {
+                fprintf(stderr, "ERROR: Unable to read \"%s\".\n", FileName);
+                FreeBuffer(&Result);
+            }
+        }
+        
+        fclose(File);
+    }
+    else
+    {
+        fprintf(stderr, "ERROR: Unable to open \"%s\".\n", FileName);
+    }
+    
+    return Result;
+}
+
+static f64 SumHaversineDistances(u64 PairCount, haversine_pair *Pairs)
+{
+    TimeBandwidth(__func__, PairCount*sizeof(haversine_pair));
+    
+    f64 Sum = 0;
+    
+    f64 SumCoef = 1 / (f64)PairCount;
+    for(u64 PairIndex = 0; PairIndex < PairCount; ++PairIndex)
+    {
+        haversine_pair Pair = Pairs[PairIndex];
+        f64 EarthRadius = 6372.8;
+        f64 Dist = ReferenceHaversine(Pair.X0, Pair.Y0, Pair.X1, Pair.Y1, EarthRadius);
+        Sum += SumCoef*Dist;
+    }
+    
+    return Sum;
+}
+
+int main(int ArgCount, char **Args)
+{
+    BeginProfile();
+	
+    int Result = 1;
+    
+    if((ArgCount == 2) || (ArgCount == 3))
+    {
+        buffer InputJSON = ReadEntireFile(Args[1]);
+        
+        u32 MinimumJSONPairEncoding = 6*4;
+        u64 MaxPairCount = InputJSON.Count / MinimumJSONPairEncoding;
+        if(MaxPairCount)
+        {
+            buffer ParsedValues = AllocateBuffer(MaxPairCount * sizeof(haversine_pair));
+            if(ParsedValues.Count)
+            {
+                haversine_pair *Pairs = (haversine_pair *)ParsedValues.Data;
+				
+                u64 PairCount = ParseHaversinePairs(InputJSON, MaxPairCount, Pairs);
+                f64 Sum = SumHaversineDistances(PairCount, Pairs);
+                
+				Result = 0;
+
+                fprintf(stdout, "Input size: %llu\n", InputJSON.Count);
+                fprintf(stdout, "Pair count: %llu\n", PairCount);
+                fprintf(stdout, "Haversine sum: %.16f\n", Sum);
+                
+                if(ArgCount == 3)
+                {
+                    buffer AnswersF64 = ReadEntireFile(Args[2]);
+                    if(AnswersF64.Count >= sizeof(f64))
+                    {
+                        f64 *AnswerValues = (f64 *)AnswersF64.Data;
+                        
+                        fprintf(stdout, "\nValidation:\n");
+                        
+                        u64 RefAnswerCount = (AnswersF64.Count - sizeof(f64)) / sizeof(f64);
+                        if(PairCount != RefAnswerCount)
+                        {
+                            fprintf(stdout, "FAILED - pair count doesn't match %llu.\n", RefAnswerCount);
+                        }
+                        
+                        f64 RefSum = AnswerValues[RefAnswerCount];
+                        fprintf(stdout, "Reference sum: %.16f\n", RefSum);
+                        fprintf(stdout, "Difference: %.16f\n", Sum - RefSum);
+                        
+                        fprintf(stdout, "\n");
+                    }
+                }
+            }
+            
+            FreeBuffer(&ParsedValues);
+        }
+        else
+        {
+            fprintf(stderr, "ERROR: Malformed input JSON\n");
+        }
+
+        FreeBuffer(&InputJSON);
+    }
+    else
+    {
+        fprintf(stderr, "Usage: %s [haversine_input.json]\n", Args[0]);
+        fprintf(stderr, "       %s [haversine_input.json] [answers.f64]\n", Args[0]);
+    }
+
+    if(Result == 0)
+	{
+        EndAndPrintProfile();
+	}
+		
+    return Result;
+}
+
+ProfilerEndOfCompilationUnit;

part3/listing_0102_read_overhead_test.cpp

@@ -0,0 +1,151 @@
+/* ========================================================================
+
+   (C) Copyright 2023 by Molly Rocket, Inc., All Rights Reserved.
+   
+   This software is provided 'as-is', without any express or implied
+   warranty. In no event will the authors be held liable for any damages
+   arising from the use of this software.
+   
+   Please see https://computerenhance.com for more information
+   
+   ======================================================================== */
+
+/* ========================================================================
+   LISTING 102
+   ======================================================================== */
+
+#include <windows.h>
+#include <fcntl.h>
+#include <io.h>
+
+struct read_parameters
+{
+    buffer Dest;
+    char const *FileName;
+};
+
+typedef void read_overhead_test_func(repetition_tester *Tester, read_parameters *Params);
+
+static void ReadViaFRead(repetition_tester *Tester, read_parameters *Params)
+{
+    while(IsTesting(Tester))
+    {
+        FILE *File = fopen(Params->FileName, "rb");
+        if(File)
+        {
+            buffer DestBuffer = Params->Dest;
+            
+            BeginTime(Tester);
+            size_t Result = fread(DestBuffer.Data, DestBuffer.Count, 1, File);
+            EndTime(Tester);
+            
+            if(Result == 1)
+            {
+                CountBytes(Tester, DestBuffer.Count);
+            }
+            else
+            {
+                Error(Tester, "fread failed");
+            }
+            
+            fclose(File);
+        }
+        else
+        {
+            Error(Tester, "fopen failed");
+        }
+    }
+}
+
+static void ReadViaRead(repetition_tester *Tester, read_parameters *Params)
+{
+    while(IsTesting(Tester))
+    {
+        int File = _open(Params->FileName, _O_BINARY|_O_RDONLY);
+        if(File != -1)
+        {
+            buffer DestBuffer = Params->Dest;
+        
+            u8 *Dest = DestBuffer.Data;
+            u64 SizeRemaining = DestBuffer.Count;
+            while(SizeRemaining)
+            {
+                u32 ReadSize = INT_MAX;
+                if((u64)ReadSize > SizeRemaining)
+                {
+                    ReadSize = (u32)SizeRemaining;
+                }
+
+                BeginTime(Tester);
+                int Result = _read(File, Dest, ReadSize);
+                EndTime(Tester);
+
+                if(Result == (int)ReadSize)
+                {
+                    CountBytes(Tester, ReadSize);
+                }
+                else
+                {
+                    Error(Tester, "_read failed");
+                    break;
+                }
+                
+                SizeRemaining -= ReadSize;
+                Dest += ReadSize;
+            }
+            
+            _close(File);
+        }
+        else
+        {
+            Error(Tester, "_open failed");
+        }
+    }
+}
+
+static void ReadViaReadFile(repetition_tester *Tester, read_parameters *Params)
+{
+    while(IsTesting(Tester))
+    {
+        HANDLE File = CreateFileA(Params->FileName, GENERIC_READ, FILE_SHARE_READ|FILE_SHARE_WRITE, 0,
+                                  OPEN_EXISTING, FILE_ATTRIBUTE_NORMAL, 0);
+        if(File != INVALID_HANDLE_VALUE)
+        {
+            buffer DestBuffer = Params->Dest;
+            
+            u64 SizeRemaining = Params->Dest.Count;
+            u8 *Dest = (u8 *)DestBuffer.Data;
+            while(SizeRemaining)
+            {
+                u32 ReadSize = (u32)-1;
+                if((u64)ReadSize > SizeRemaining)
+                {
+                    ReadSize = (u32)SizeRemaining;
+                }
+                
+                DWORD BytesRead = 0;
+                BeginTime(Tester);
+                BOOL Result = ReadFile(File, Dest, ReadSize, &BytesRead, 0);
+                EndTime(Tester);
+                
+                if(Result && (BytesRead == ReadSize))
+                {
+                    CountBytes(Tester, ReadSize);
+                }
+                else
+                {
+                    Error(Tester, "ReadFile failed");
+                }
+                
+                SizeRemaining -= ReadSize;
+                Dest += ReadSize;
+            }
+            
+            CloseHandle(File);
+        }
+        else
+        {
+            Error(Tester, "CreateFileA failed");
+        }
+    }
+}

part3/listing_0103_repetition_tester.cpp

@@ -0,0 +1,211 @@
+/* ========================================================================
+
+   (C) Copyright 2023 by Molly Rocket, Inc., All Rights Reserved.
+   
+   This software is provided 'as-is', without any express or implied
+   warranty. In no event will the authors be held liable for any damages
+   arising from the use of this software.
+   
+   Please see https://computerenhance.com for more information
+   
+   ======================================================================== */
+
+/* ========================================================================
+   LISTING 103
+   ======================================================================== */
+
+enum test_mode : u32
+{
+    TestMode_Uninitialized,
+    TestMode_Testing,
+    TestMode_Completed,
+    TestMode_Error,
+};
+
+struct repetition_test_results
+{
+    u64 TestCount;
+    u64 TotalTime;
+    u64 MaxTime;
+    u64 MinTime;
+};
+
+struct repetition_tester
+{
+    u64 TargetProcessedByteCount;
+    u64 CPUTimerFreq;
+    u64 TryForTime;
+    u64 TestsStartedAt;
+    
+    test_mode Mode;
+    b32 PrintNewMinimums;
+    u32 OpenBlockCount;
+    u32 CloseBlockCount;
+    u64 TimeAccumulatedOnThisTest;
+    u64 BytesAccumulatedOnThisTest;
+
+    repetition_test_results Results;
+};
+
+static f64 SecondsFromCPUTime(f64 CPUTime, u64 CPUTimerFreq)
+{
+    f64 Result = 0.0;
+    if(CPUTimerFreq)
+    {
+        Result = (CPUTime / (f64)CPUTimerFreq);
+    }
+    
+    return Result;
+}
+ 
+static void PrintTime(char const *Label, f64 CPUTime, u64 CPUTimerFreq, u64 ByteCount)
+{
+    printf("%s: %.0f", Label, CPUTime);
+    if(CPUTimerFreq)
+    {
+        f64 Seconds = SecondsFromCPUTime(CPUTime, CPUTimerFreq);
+        printf(" (%fms)", 1000.0f*Seconds);
+    
+        if(ByteCount)
+        {
+            f64 Gigabyte = (1024.0f * 1024.0f * 1024.0f);
+            f64 BestBandwidth = ByteCount / (Gigabyte * Seconds);
+            printf(" %fgb/s", BestBandwidth);
+        }
+    }
+}
+
+static void PrintTime(char const *Label, u64 CPUTime, u64 CPUTimerFreq, u64 ByteCount)
+{
+    PrintTime(Label, (f64)CPUTime, CPUTimerFreq, ByteCount);
+}
+
+static void PrintResults(repetition_test_results Results, u64 CPUTimerFreq, u64 ByteCount)
+{
+    PrintTime("Min", (f64)Results.MinTime, CPUTimerFreq, ByteCount);
+    printf("\n");
+    
+    PrintTime("Max", (f64)Results.MaxTime, CPUTimerFreq, ByteCount);
+    printf("\n");
+    
+    if(Results.TestCount)
+    {
+        PrintTime("Avg", (f64)Results.TotalTime / (f64)Results.TestCount, CPUTimerFreq, ByteCount);
+        printf("\n");
+    }
+}
+
+static void Error(repetition_tester *Tester, char const *Message)
+{
+    Tester->Mode = TestMode_Error;
+    fprintf(stderr, "ERROR: %s\n", Message);
+}
+
+static void NewTestWave(repetition_tester *Tester, u64 TargetProcessedByteCount, u64 CPUTimerFreq, u32 SecondsToTry = 10)
+{
+    if(Tester->Mode == TestMode_Uninitialized)
+    {
+        Tester->Mode = TestMode_Testing;
+        Tester->TargetProcessedByteCount = TargetProcessedByteCount;
+        Tester->CPUTimerFreq = CPUTimerFreq;
+        Tester->PrintNewMinimums = true;
+        Tester->Results.MinTime = (u64)-1;
+    }
+    else if(Tester->Mode == TestMode_Completed)
+    {
+        Tester->Mode = TestMode_Testing;
+        
+        if(Tester->TargetProcessedByteCount != TargetProcessedByteCount)
+        {
+            Error(Tester, "TargetProcessedByteCount changed");
+        }
+        
+        if(Tester->CPUTimerFreq != CPUTimerFreq)
+        {
+            Error(Tester, "CPU frequency changed");
+        }
+    }
+
+    Tester->TryForTime = SecondsToTry*CPUTimerFreq;
+    Tester->TestsStartedAt = ReadCPUTimer();
+}
+
+static void BeginTime(repetition_tester *Tester)
+{
+    ++Tester->OpenBlockCount;
+    Tester->TimeAccumulatedOnThisTest -= ReadCPUTimer();
+}
+
+static void EndTime(repetition_tester *Tester)
+{
+    ++Tester->CloseBlockCount;
+    Tester->TimeAccumulatedOnThisTest += ReadCPUTimer();
+}
+
+static void CountBytes(repetition_tester *Tester, u64 ByteCount)
+{
+    Tester->BytesAccumulatedOnThisTest += ByteCount;
+}
+
+static b32 IsTesting(repetition_tester *Tester)
+{
+    if(Tester->Mode == TestMode_Testing)
+    {
+        u64 CurrentTime = ReadCPUTimer();
+        
+        if(Tester->OpenBlockCount) // NOTE(casey): We don't count tests that had no timing blocks - we assume they took some other path
+        {
+            if(Tester->OpenBlockCount != Tester->CloseBlockCount)
+            {
+                Error(Tester, "Unbalanced BeginTime/EndTime");
+            }
+            
+            if(Tester->BytesAccumulatedOnThisTest != Tester->TargetProcessedByteCount)
+            {
+                Error(Tester, "Processed byte count mismatch");
+            }
+    
+            if(Tester->Mode == TestMode_Testing)
+            {
+                repetition_test_results *Results = &Tester->Results;
+                u64 ElapsedTime = Tester->TimeAccumulatedOnThisTest;
+                Results->TestCount += 1;
+                Results->TotalTime += ElapsedTime;
+                if(Results->MaxTime < ElapsedTime)
+                {
+                    Results->MaxTime = ElapsedTime;
+                }
+                
+                if(Results->MinTime > ElapsedTime)
+                {
+                    Results->MinTime = ElapsedTime;
+                    
+                    // NOTE(casey): Whenever we get a new minimum time, we reset the clock to the full trial time
+                    Tester->TestsStartedAt = CurrentTime;
+                    
+                    if(Tester->PrintNewMinimums)
+                    {
+                        PrintTime("Min", Results->MinTime, Tester->CPUTimerFreq, Tester->BytesAccumulatedOnThisTest);
+                        printf("               \r");
+                    }
+                }
+                
+                Tester->OpenBlockCount = 0;
+                Tester->CloseBlockCount = 0;
+                Tester->TimeAccumulatedOnThisTest = 0;
+                Tester->BytesAccumulatedOnThisTest = 0;
+            }
+        }
+        
+        if((CurrentTime - Tester->TestsStartedAt) > Tester->TryForTime)
+        {
+            Tester->Mode = TestMode_Completed;
+            
+            printf("                                                          \r");
+            PrintResults(Tester->Results, Tester->CPUTimerFreq, Tester->TargetProcessedByteCount);
+        }
+    }
+    
+    b32 Result = (Tester->Mode == TestMode_Testing);
+    return Result;
+}

part3/listing_0104_read_overhead_main.cpp

@@ -0,0 +1,116 @@
+/* ========================================================================
+
+   (C) Copyright 2023 by Molly Rocket, Inc., All Rights Reserved.
+   
+   This software is provided 'as-is', without any express or implied
+   warranty. In no event will the authors be held liable for any damages
+   arising from the use of this software.
+   
+   Please see https://computerenhance.com for more information
+   
+   ======================================================================== */
+
+/* ========================================================================
+   LISTING 104
+   ======================================================================== */
+
+/* NOTE(casey): _CRT_SECURE_NO_WARNINGS is here because otherwise we cannot
+   call fopen(). If we replace fopen() with fopen_s() to avoid the warning,
+   then the code doesn't compile on Linux anymore, since fopen_s() does not
+   exist there.
+   
+   What exactly the CRT maintainers were thinking when they made this choice,
+   I have no idea. */
+#define _CRT_SECURE_NO_WARNINGS
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <stdint.h>
+#include <math.h>
+#include <sys/stat.h>
+
+typedef uint8_t u8;
+typedef uint32_t u32;
+typedef uint64_t u64;
+
+typedef int32_t b32;
+
+typedef float f32;
+typedef double f64;
+
+#define ArrayCount(Array) (sizeof(Array)/sizeof((Array)[0]))
+
+#include "listing_0068_buffer.cpp"
+#include "listing_0074_platform_metrics.cpp"
+#include "listing_0103_repetition_tester.cpp"
+#include "listing_0102_read_overhead_test.cpp"
+
+struct test_function
+{
+    char const *Name;
+    read_overhead_test_func *Func;
+};
+test_function TestFunctions[] =
+{
+    {"fread", ReadViaFRead},
+    {"_read", ReadViaRead},
+    {"ReadFile", ReadViaReadFile},
+};
+
+int main(int ArgCount, char **Args)
+{
+    // NOTE(casey): Since we do not use these functions in this particular build, we reference their pointers
+    // here to prevent the compiler from complaining about "unused functions".
+    (void)&IsInBounds;
+    (void)&AreEqual;
+    
+    u64 CPUTimerFreq = EstimateCPUTimerFreq();
+    
+    if(ArgCount == 2)
+    {
+        char *FileName = Args[1];
+#if _WIN32
+        struct __stat64 Stat;
+        _stat64(FileName, &Stat);
+#else
+        struct stat Stat;
+        stat(FileName, &Stat);
+#endif
+        
+        read_parameters Params = {};
+        Params.Dest = AllocateBuffer(Stat.st_size);
+        Params.FileName = FileName;
+    
+        if(Params.Dest.Count > 0)
+        {
+            repetition_tester Testers[ArrayCount(TestFunctions)] = {};
+            
+            for(;;)
+            {
+                for(u32 FuncIndex = 0; FuncIndex < ArrayCount(TestFunctions); ++FuncIndex)
+                {
+                    repetition_tester *Tester = Testers + FuncIndex;
+                    test_function TestFunc = TestFunctions[FuncIndex];
+                    
+                    printf("\n--- %s ---\n", TestFunc.Name);
+                    NewTestWave(Tester, Params.Dest.Count, CPUTimerFreq);
+                    TestFunc.Func(Tester, &Params);
+                }
+            }
+            
+            // NOTE(casey): We would normally call this here, but we can't because the compiler will complain about "unreachable code".
+            // So instead we just reference the pointer to prevent the compiler complaining about unused function :(
+            (void)&FreeBuffer;
+        }
+        else
+        {
+            fprintf(stderr, "ERROR: Test data size must be non-zero\n");
+        }
+    }
+    else
+    {
+        fprintf(stderr, "Usage: %s [existing filename]\n", Args[0]);
+    }
+		
+    return 0;
+}

part3/listing_0105_read_overhead_reordered_main.cpp

@@ -0,0 +1,116 @@
+/* ========================================================================
+
+   (C) Copyright 2023 by Molly Rocket, Inc., All Rights Reserved.
+   
+   This software is provided 'as-is', without any express or implied
+   warranty. In no event will the authors be held liable for any damages
+   arising from the use of this software.
+   
+   Please see https://computerenhance.com for more information
+   
+   ======================================================================== */
+
+/* ========================================================================
+   LISTING 105
+   ======================================================================== */
+
+/* NOTE(casey): _CRT_SECURE_NO_WARNINGS is here because otherwise we cannot
+   call fopen(). If we replace fopen() with fopen_s() to avoid the warning,
+   then the code doesn't compile on Linux anymore, since fopen_s() does not
+   exist there.
+   
+   What exactly the CRT maintainers were thinking when they made this choice,
+   I have no idea. */
+#define _CRT_SECURE_NO_WARNINGS
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <stdint.h>
+#include <math.h>
+#include <sys/stat.h>
+
+typedef uint8_t u8;
+typedef uint32_t u32;
+typedef uint64_t u64;
+
+typedef int32_t b32;
+
+typedef float f32;
+typedef double f64;
+
+#define ArrayCount(Array) (sizeof(Array)/sizeof((Array)[0]))
+
+#include "listing_0068_buffer.cpp"
+#include "listing_0074_platform_metrics.cpp"
+#include "listing_0103_repetition_tester.cpp"
+#include "listing_0102_read_overhead_test.cpp"
+
+struct test_function
+{
+    char const *Name;
+    read_overhead_test_func *Func;
+};
+test_function TestFunctions[] =
+{
+    {"ReadFile", ReadViaReadFile},
+    {"_read", ReadViaRead},
+    {"fread", ReadViaFRead},
+};
+
+int main(int ArgCount, char **Args)
+{
+    // NOTE(casey): Since we do not use these functions in this particular build, we reference their pointers
+    // here to prevent the compiler from complaining about "unused functions".
+    (void)&IsInBounds;
+    (void)&AreEqual;
+    
+    u64 CPUTimerFreq = EstimateCPUTimerFreq();
+    
+    if(ArgCount == 2)
+    {
+        char *FileName = Args[1];
+#if _WIN32
+        struct __stat64 Stat;
+        _stat64(FileName, &Stat);
+#else
+        struct stat Stat;
+        stat(FileName, &Stat);
+#endif
+        
+        read_parameters Params = {};
+        Params.Dest = AllocateBuffer(Stat.st_size);
+        Params.FileName = FileName;
+    
+        if(Params.Dest.Count > 0)
+        {
+            repetition_tester Testers[ArrayCount(TestFunctions)] = {};
+            
+            for(;;)
+            {
+                for(u32 FuncIndex = 0; FuncIndex < ArrayCount(TestFunctions); ++FuncIndex)
+                {
+                    repetition_tester *Tester = Testers + FuncIndex;
+                    test_function TestFunc = TestFunctions[FuncIndex];
+                    
+                    printf("\n--- %s ---\n", TestFunc.Name);
+                    NewTestWave(Tester, Params.Dest.Count, CPUTimerFreq);
+                    TestFunc.Func(Tester, &Params);
+                }
+            }
+            
+            // NOTE(casey): We would normally call this here, but we can't because the compiler will complain about "unreachable code".
+            // So instead we just reference the pointer to prevent the compiler complaining about unused function :(
+            (void)&FreeBuffer;
+        }
+        else
+        {
+            fprintf(stderr, "ERROR: Test data size must be non-zero\n");
+        }
+    }
+    else
+    {
+        fprintf(stderr, "Usage: %s [existing filename]\n", Args[0]);
+    }
+		
+    return 0;
+}

part3/listing_0106_mallocread_overhead_test.cpp

@@ -0,0 +1,219 @@
+/* ========================================================================
+
+   (C) Copyright 2023 by Molly Rocket, Inc., All Rights Reserved.
+   
+   This software is provided 'as-is', without any express or implied
+   warranty. In no event will the authors be held liable for any damages
+   arising from the use of this software.
+   
+   Please see https://computerenhance.com for more information
+   
+   ======================================================================== */
+
+/* ========================================================================
+   LISTING 106
+   ======================================================================== */
+
+#include <windows.h>
+#include <fcntl.h>
+#include <io.h>
+
+enum allocation_type
+{
+    AllocType_none,
+    AllocType_malloc,
+    
+    AllocType_Count,
+};
+
+struct read_parameters
+{
+    allocation_type AllocType;
+    buffer Dest;
+    char const *FileName;
+};
+
+typedef void read_overhead_test_func(repetition_tester *Tester, read_parameters *Params);
+
+static char const *DescribeAllocationType(allocation_type AllocType)
+{
+    char const *Result;
+    switch(AllocType)
+    {
+        case AllocType_none: {Result = "";} break;
+        case AllocType_malloc: {Result = "malloc";} break;
+        default : {Result = "UNKNOWN";} break;
+    }
+    
+    return Result;
+}
+
+static void HandleAllocation(read_parameters *Params, buffer *Buffer)
+{
+    switch(Params->AllocType)
+    {
+        case AllocType_none:
+        {
+        } break;
+        
+        case AllocType_malloc:
+        {
+            *Buffer = AllocateBuffer(Params->Dest.Count);
+        } break;
+        
+        default:
+        {
+            fprintf(stderr, "ERROR: Unrecognized allocation type");
+        } break;
+    }
+}
+
+static void HandleDeallocation(read_parameters *Params, buffer *Buffer)
+{
+    switch(Params->AllocType)
+    {
+        case AllocType_none:
+        {
+        } break;
+        
+        case AllocType_malloc:
+        {
+            FreeBuffer(Buffer);
+        } break;
+        
+        default:
+        {
+            fprintf(stderr, "ERROR: Unrecognized allocation type");
+        } break;
+    }
+}
+
+static void ReadViaFRead(repetition_tester *Tester, read_parameters *Params)
+{
+    while(IsTesting(Tester))
+    {
+        FILE *File = fopen(Params->FileName, "rb");
+        if(File)
+        {
+            buffer DestBuffer = Params->Dest;
+            HandleAllocation(Params, &DestBuffer);
+            
+            BeginTime(Tester);
+            size_t Result = fread(DestBuffer.Data, DestBuffer.Count, 1, File);
+            EndTime(Tester);
+            
+            if(Result == 1)
+            {
+                CountBytes(Tester, DestBuffer.Count);
+            }
+            else
+            {
+                Error(Tester, "fread failed");
+            }
+            
+            HandleDeallocation(Params, &DestBuffer);
+            fclose(File);
+        }
+        else
+        {
+            Error(Tester, "fopen failed");
+        }
+    }
+}
+
+static void ReadViaRead(repetition_tester *Tester, read_parameters *Params)
+{
+    while(IsTesting(Tester))
+    {
+        int File = _open(Params->FileName, _O_BINARY|_O_RDONLY);
+        if(File != -1)
+        {
+            buffer DestBuffer = Params->Dest;
+            HandleAllocation(Params, &DestBuffer);
+        
+            u8 *Dest = DestBuffer.Data;
+            u64 SizeRemaining = DestBuffer.Count;
+            while(SizeRemaining)
+            {
+                u32 ReadSize = INT_MAX;
+                if((u64)ReadSize > SizeRemaining)
+                {
+                    ReadSize = (u32)SizeRemaining;
+                }
+
+                BeginTime(Tester);
+                int Result = _read(File, Dest, ReadSize);
+                EndTime(Tester);
+
+                if(Result == (int)ReadSize)
+                {
+                    CountBytes(Tester, ReadSize);
+                }
+                else
+                {
+                    Error(Tester, "_read failed");
+                    break;
+                }
+                
+                SizeRemaining -= ReadSize;
+                Dest += ReadSize;
+            }
+            
+            HandleDeallocation(Params, &DestBuffer);
+            _close(File);
+        }
+        else
+        {
+            Error(Tester, "_open failed");
+        }
+    }
+}
+
+static void ReadViaReadFile(repetition_tester *Tester, read_parameters *Params)
+{
+    while(IsTesting(Tester))
+    {
+        HANDLE File = CreateFileA(Params->FileName, GENERIC_READ, FILE_SHARE_READ|FILE_SHARE_WRITE, 0,
+                                  OPEN_EXISTING, FILE_ATTRIBUTE_NORMAL, 0);
+        if(File != INVALID_HANDLE_VALUE)
+        {
+            buffer DestBuffer = Params->Dest;
+            HandleAllocation(Params, &DestBuffer);
+            
+            u64 SizeRemaining = Params->Dest.Count;
+            u8 *Dest = (u8 *)DestBuffer.Data;
+            while(SizeRemaining)
+            {
+                u32 ReadSize = (u32)-1;
+                if((u64)ReadSize > SizeRemaining)
+                {
+                    ReadSize = (u32)SizeRemaining;
+                }
+                
+                DWORD BytesRead = 0;
+                BeginTime(Tester);
+                BOOL Result = ReadFile(File, Dest, ReadSize, &BytesRead, 0);
+                EndTime(Tester);
+                
+                if(Result && (BytesRead == ReadSize))
+                {
+                    CountBytes(Tester, ReadSize);
+                }
+                else
+                {
+                    Error(Tester, "ReadFile failed");
+                }
+                
+                SizeRemaining -= ReadSize;
+                Dest += ReadSize;
+            }
+            
+            HandleDeallocation(Params, &DestBuffer);
+            CloseHandle(File);
+        }
+        else
+        {
+            Error(Tester, "CreateFileA failed");
+        }
+    }
+}

part3/listing_0107_mallocread_overhead_main.cpp

@@ -0,0 +1,124 @@
+/* ========================================================================
+
+   (C) Copyright 2023 by Molly Rocket, Inc., All Rights Reserved.
+   
+   This software is provided 'as-is', without any express or implied
+   warranty. In no event will the authors be held liable for any damages
+   arising from the use of this software.
+   
+   Please see https://computerenhance.com for more information
+   
+   ======================================================================== */
+
+/* ========================================================================
+   LISTING 107
+   ======================================================================== */
+
+/* NOTE(casey): _CRT_SECURE_NO_WARNINGS is here because otherwise we cannot
+   call fopen(). If we replace fopen() with fopen_s() to avoid the warning,
+   then the code doesn't compile on Linux anymore, since fopen_s() does not
+   exist there.
+   
+   What exactly the CRT maintainers were thinking when they made this choice,
+   I have no idea. */
+#define _CRT_SECURE_NO_WARNINGS
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <stdint.h>
+#include <math.h>
+#include <sys/stat.h>
+
+typedef uint8_t u8;
+typedef uint32_t u32;
+typedef uint64_t u64;
+
+typedef int32_t b32;
+
+typedef float f32;
+typedef double f64;
+
+#define ArrayCount(Array) (sizeof(Array)/sizeof((Array)[0]))
+
+#include "listing_0068_buffer.cpp"
+#include "listing_0074_platform_metrics.cpp"
+#include "listing_0103_repetition_tester.cpp"
+#include "listing_0106_mallocread_overhead_test.cpp"
+
+struct test_function
+{
+    char const *Name;
+    read_overhead_test_func *Func;
+};
+test_function TestFunctions[] =
+{
+    {"fread", ReadViaFRead},
+    {"_read", ReadViaRead},
+    {"ReadFile", ReadViaReadFile},
+};
+
+int main(int ArgCount, char **Args)
+{
+    // NOTE(casey): Since we do not use these functions in this particular build, we reference their pointers
+    // here to prevent the compiler from complaining about "unused functions".
+    (void)&IsInBounds;
+    (void)&AreEqual;
+    
+    u64 CPUTimerFreq = EstimateCPUTimerFreq();
+    
+    if(ArgCount == 2)
+    {
+        char *FileName = Args[1];
+#if _WIN32
+        struct __stat64 Stat;
+        _stat64(FileName, &Stat);
+#else
+        struct stat Stat;
+        stat(FileName, &Stat);
+#endif
+        
+        read_parameters Params = {};
+        Params.Dest = AllocateBuffer(Stat.st_size);
+        Params.FileName = FileName;
+    
+        if(Params.Dest.Count > 0)
+        {
+            repetition_tester Testers[ArrayCount(TestFunctions)][AllocType_Count] = {};
+            
+            for(;;)
+            {
+                for(u32 FuncIndex = 0; FuncIndex < ArrayCount(TestFunctions); ++FuncIndex)
+                {
+                    for(u32 AllocType = 0; AllocType < AllocType_Count; ++AllocType)
+                    {
+                        Params.AllocType = (allocation_type)AllocType;
+                        
+                        repetition_tester *Tester = &Testers[FuncIndex][AllocType];
+                        test_function TestFunc = TestFunctions[FuncIndex];
+                        
+                        printf("\n--- %s%s%s ---\n",
+                               DescribeAllocationType(Params.AllocType),
+                               Params.AllocType ? " + " : "",
+                               TestFunc.Name);
+                        NewTestWave(Tester, Params.Dest.Count, CPUTimerFreq);
+                        TestFunc.Func(Tester, &Params);
+                    }
+                }
+            }
+            
+            // NOTE(casey): We would normally call this here, but we can't because the compiler will complain about "unreachable code".
+            // So instead we just reference the pointer to prevent the compiler complaining about unused function :(
+            (void)&FreeBuffer;
+        }
+        else
+        {
+            fprintf(stderr, "ERROR: Test data size must be non-zero\n");
+        }
+    }
+    else
+    {
+        fprintf(stderr, "Usage: %s [existing filename]\n", Args[0]);
+    }
+		
+    return 0;
+}

part3/listing_0108_platform_metrics.cpp

@@ -0,0 +1,145 @@
+/* ========================================================================
+
+   (C) Copyright 2023 by Molly Rocket, Inc., All Rights Reserved.
+   
+   This software is provided 'as-is', without any express or implied
+   warranty. In no event will the authors be held liable for any damages
+   arising from the use of this software.
+   
+   Please see https://computerenhance.com for more information
+   
+   ======================================================================== */
+
+/* ========================================================================
+   LISTING 108
+   ======================================================================== */
+
+#if _WIN32
+
+#include <intrin.h>
+#include <windows.h>
+#include <psapi.h>
+
+typedef struct os_metrics
+{
+    b32 Initialized;
+    HANDLE ProcessHandle;
+} os_metrics;
+static os_metrics GlobalMetrics;
+
+static u64 GetOSTimerFreq(void)
+{
+	LARGE_INTEGER Freq;
+	QueryPerformanceFrequency(&Freq);
+	return Freq.QuadPart;
+}
+
+static u64 ReadOSTimer(void)
+{
+	LARGE_INTEGER Value;
+	QueryPerformanceCounter(&Value);
+	return Value.QuadPart;
+}
+
+static u64 ReadOSPageFaultCount(void)
+{
+    PROCESS_MEMORY_COUNTERS_EX MemoryCounters = {};
+    MemoryCounters.cb = sizeof(MemoryCounters);
+    GetProcessMemoryInfo(GlobalMetrics.ProcessHandle, (PROCESS_MEMORY_COUNTERS *)&MemoryCounters, sizeof(MemoryCounters));
+    
+    u64 Result = MemoryCounters.PageFaultCount;
+    return Result;
+}
+
+static void InitializeOSMetrics(void)
+{
+    if(!GlobalMetrics.Initialized)
+    {
+        GlobalMetrics.Initialized = true;
+        GlobalMetrics.ProcessHandle = OpenProcess(PROCESS_QUERY_INFORMATION | PROCESS_VM_READ, FALSE, GetCurrentProcessId());
+    }
+}
+
+#else
+
+#include <x86intrin.h>
+#include <sys/time.h>
+
+static u64 GetOSTimerFreq(void)
+{
+	return 1000000;
+}
+
+static u64 ReadOSTimer(void)
+{
+	// NOTE(casey): The "struct" keyword is not necessary here when compiling in C++,
+	// but just in case anyone is using this file from C, I include it.
+	struct timeval Value;
+	gettimeofday(&Value, 0);
+	
+	u64 Result = GetOSTimerFreq()*(u64)Value.tv_sec + (u64)Value.tv_usec;
+	return Result;
+}
+
+static u64 ReadOSPageFaultCount(void)
+{
+    // NOTE(casey): The course materials are not tested on MacOS/Linux.
+    // This code was contributed to the public github. It may or may not work
+    // for your system.
+    
+    struct rusage Usage = {};
+    getrusage(RUSAGE_SELF, &Usage);
+    
+    // ru_minflt  the number of page faults serviced without any I/O activity.
+    // ru_majflt  the number of page faults serviced that required I/O activity.
+    u64 Result = Usage.ru_minflt + Usage.ru_majflt;
+    
+    return Result;
+}
+
+static void InitializeOSMetrics(void)
+{
+}
+
+#endif
+
+/* NOTE(casey): This does not need to be "inline", it could just be "static"
+   because compilers will inline it anyway. But compilers will warn about
+   static functions that aren't used. So "inline" is just the simplest way
+   to tell them to stop complaining about that. */
+inline u64 ReadCPUTimer(void)
+{
+	// NOTE(casey): If you were on ARM, you would need to replace __rdtsc
+	// with one of their performance counter read instructions, depending
+	// on which ones are available on your platform.
+	
+	return __rdtsc();
+}
+
+static u64 EstimateCPUTimerFreq(void)
+{
+	u64 MillisecondsToWait = 100;
+	u64 OSFreq = GetOSTimerFreq();
+
+	u64 CPUStart = ReadCPUTimer();
+	u64 OSStart = ReadOSTimer();
+	u64 OSEnd = 0;
+	u64 OSElapsed = 0;
+	u64 OSWaitTime = OSFreq * MillisecondsToWait / 1000;
+	while(OSElapsed < OSWaitTime)
+	{
+		OSEnd = ReadOSTimer();
+		OSElapsed = OSEnd - OSStart;
+	}
+	
+	u64 CPUEnd = ReadCPUTimer();
+	u64 CPUElapsed = CPUEnd - CPUStart;
+	
+	u64 CPUFreq = 0;
+	if(OSElapsed)
+	{
+		CPUFreq = OSFreq * CPUElapsed / OSElapsed;
+	}
+	
+	return CPUFreq;
+}

part3/listing_0109_pagefault_repetition_tester.cpp

@@ -0,0 +1,240 @@
+/* ========================================================================
+
+   (C) Copyright 2023 by Molly Rocket, Inc., All Rights Reserved.
+   
+   This software is provided 'as-is', without any express or implied
+   warranty. In no event will the authors be held liable for any damages
+   arising from the use of this software.
+   
+   Please see https://computerenhance.com for more information
+   
+   ======================================================================== */
+
+/* ========================================================================
+   LISTING 109
+   ======================================================================== */
+
+enum test_mode : u32
+{
+    TestMode_Uninitialized,
+    TestMode_Testing,
+    TestMode_Completed,
+    TestMode_Error,
+};
+
+enum repetition_value_type
+{
+    RepValue_TestCount,
+    
+    RepValue_CPUTimer,
+    RepValue_MemPageFaults,
+    RepValue_ByteCount,
+    
+    RepValue_Count,
+};
+
+struct repetition_value
+{
+    u64 E[RepValue_Count];
+};
+
+struct repetition_test_results
+{
+    repetition_value Total;
+    repetition_value Min;
+    repetition_value Max;
+};
+
+struct repetition_tester
+{
+    u64 TargetProcessedByteCount;
+    u64 CPUTimerFreq;
+    u64 TryForTime;
+    u64 TestsStartedAt;
+    
+    test_mode Mode;
+    b32 PrintNewMinimums;
+    u32 OpenBlockCount;
+    u32 CloseBlockCount;
+    
+    repetition_value AccumulatedOnThisTest;
+    repetition_test_results Results;
+};
+
+static f64 SecondsFromCPUTime(f64 CPUTime, u64 CPUTimerFreq)
+{
+    f64 Result = 0.0;
+    if(CPUTimerFreq)
+    {
+        Result = (CPUTime / (f64)CPUTimerFreq);
+    }
+    
+    return Result;
+}
+ 
+static void PrintValue(char const *Label, repetition_value Value, u64 CPUTimerFreq)
+{
+    u64 TestCount = Value.E[RepValue_TestCount];
+    f64 Divisor = TestCount ? (f64)TestCount : 1;
+    
+    f64 E[RepValue_Count];
+    for(u32 EIndex = 0; EIndex < ArrayCount(E); ++EIndex)
+    {
+        E[EIndex] = (f64)Value.E[EIndex] / Divisor;
+    }
+    
+    printf("%s: %.0f", Label, E[RepValue_CPUTimer]);
+    if(CPUTimerFreq)
+    {
+        f64 Seconds = SecondsFromCPUTime(E[RepValue_CPUTimer], CPUTimerFreq);
+        printf(" (%fms)", 1000.0f*Seconds);
+    
+        if(E[RepValue_ByteCount] > 0)
+        {
+            f64 Gigabyte = (1024.0f * 1024.0f * 1024.0f);
+            f64 Bandwidth = E[RepValue_ByteCount] / (Gigabyte * Seconds);
+            printf(" %fgb/s", Bandwidth);
+        }
+    }
+
+    if(E[RepValue_MemPageFaults] > 0)
+    {
+        printf(" PF: %0.4f (%0.4fk/fault)", E[RepValue_MemPageFaults], E[RepValue_ByteCount] / (E[RepValue_MemPageFaults] * 1024.0));
+    }
+}
+
+static void PrintResults(repetition_test_results Results, u64 CPUTimerFreq)
+{
+    PrintValue("Min", Results.Min, CPUTimerFreq);
+    printf("\n");
+    PrintValue("Max", Results.Max, CPUTimerFreq);
+    printf("\n");
+    PrintValue("Avg", Results.Total, CPUTimerFreq);
+    printf("\n");
+}
+
+static void Error(repetition_tester *Tester, char const *Message)
+{
+    Tester->Mode = TestMode_Error;
+    fprintf(stderr, "ERROR: %s\n", Message);
+}
+
+static void NewTestWave(repetition_tester *Tester, u64 TargetProcessedByteCount, u64 CPUTimerFreq, u32 SecondsToTry = 10)
+{
+    if(Tester->Mode == TestMode_Uninitialized)
+    {
+        Tester->Mode = TestMode_Testing;
+        Tester->TargetProcessedByteCount = TargetProcessedByteCount;
+        Tester->CPUTimerFreq = CPUTimerFreq;
+        Tester->PrintNewMinimums = true;
+        Tester->Results.Min.E[RepValue_CPUTimer] = (u64)-1;
+    }
+    else if(Tester->Mode == TestMode_Completed)
+    {
+        Tester->Mode = TestMode_Testing;
+        
+        if(Tester->TargetProcessedByteCount != TargetProcessedByteCount)
+        {
+            Error(Tester, "TargetProcessedByteCount changed");
+        }
+        
+        if(Tester->CPUTimerFreq != CPUTimerFreq)
+        {
+            Error(Tester, "CPU frequency changed");
+        }
+    }
+
+    Tester->TryForTime = SecondsToTry*CPUTimerFreq;
+    Tester->TestsStartedAt = ReadCPUTimer();
+}
+
+static void BeginTime(repetition_tester *Tester)
+{
+    ++Tester->OpenBlockCount;
+    
+    repetition_value *Accum = &Tester->AccumulatedOnThisTest;
+    Accum->E[RepValue_MemPageFaults] -= ReadOSPageFaultCount();
+    Accum->E[RepValue_CPUTimer] -= ReadCPUTimer();
+}
+
+static void EndTime(repetition_tester *Tester)
+{
+    repetition_value *Accum = &Tester->AccumulatedOnThisTest;
+    Accum->E[RepValue_CPUTimer] += ReadCPUTimer();
+    Accum->E[RepValue_MemPageFaults] += ReadOSPageFaultCount();
+
+    ++Tester->CloseBlockCount;
+}
+
+static void CountBytes(repetition_tester *Tester, u64 ByteCount)
+{
+    repetition_value *Accum = &Tester->AccumulatedOnThisTest;
+    Accum->E[RepValue_ByteCount] += ByteCount;
+}
+
+static b32 IsTesting(repetition_tester *Tester)
+{
+    if(Tester->Mode == TestMode_Testing)
+    {
+        repetition_value Accum = Tester->AccumulatedOnThisTest;
+        u64 CurrentTime = ReadCPUTimer();
+        
+        if(Tester->OpenBlockCount) // NOTE(casey): We don't count tests that had no timing blocks - we assume they took some other path
+        {
+            if(Tester->OpenBlockCount != Tester->CloseBlockCount)
+            {
+                Error(Tester, "Unbalanced BeginTime/EndTime");
+            }
+            
+            if(Accum.E[RepValue_ByteCount] != Tester->TargetProcessedByteCount)
+            {
+                Error(Tester, "Processed byte count mismatch");
+            }
+    
+            if(Tester->Mode == TestMode_Testing)
+            {
+                repetition_test_results *Results = &Tester->Results;
+                
+                Accum.E[RepValue_TestCount] = 1;
+                for(u32 EIndex = 0; EIndex < ArrayCount(Accum.E); ++EIndex)
+                {
+                    Results->Total.E[EIndex] += Accum.E[EIndex];
+                }
+                
+                if(Results->Max.E[RepValue_CPUTimer] < Accum.E[RepValue_CPUTimer])
+                {
+                    Results->Max = Accum;
+                }
+                
+                if(Results->Min.E[RepValue_CPUTimer] > Accum.E[RepValue_CPUTimer])
+                {
+                    Results->Min = Accum;
+                    
+                    // NOTE(casey): Whenever we get a new minimum time, we reset the clock to the full trial time
+                    Tester->TestsStartedAt = CurrentTime;
+                    
+                    if(Tester->PrintNewMinimums)
+                    {
+                        PrintValue("Min", Results->Min, Tester->CPUTimerFreq);
+                        printf("                                   \r");
+                    }
+                }
+                
+                Tester->OpenBlockCount = 0;
+                Tester->CloseBlockCount = 0;
+                Tester->AccumulatedOnThisTest = {};
+            }
+        }
+        
+        if((CurrentTime - Tester->TestsStartedAt) > Tester->TryForTime)
+        {
+            Tester->Mode = TestMode_Completed;
+            
+            printf("                                                          \r");
+            PrintResults(Tester->Results, Tester->CPUTimerFreq);
+        }
+    }
+    
+    b32 Result = (Tester->Mode == TestMode_Testing);
+    return Result;
+}

part3/listing_0110_pagefault_overhead_test.cpp

@@ -0,0 +1,35 @@
+/* ========================================================================
+
+   (C) Copyright 2023 by Molly Rocket, Inc., All Rights Reserved.
+   
+   This software is provided 'as-is', without any express or implied
+   warranty. In no event will the authors be held liable for any damages
+   arising from the use of this software.
+   
+   Please see https://computerenhance.com for more information
+   
+   ======================================================================== */
+
+/* ========================================================================
+   LISTING 110
+   ======================================================================== */
+
+static void WriteToAllBytes(repetition_tester *Tester, read_parameters *Params)
+{
+    while(IsTesting(Tester))
+    {
+        buffer DestBuffer = Params->Dest;
+        HandleAllocation(Params, &DestBuffer);
+        
+        BeginTime(Tester);
+        for(u64 Index = 0; Index < DestBuffer.Count; ++Index)
+        {
+            DestBuffer.Data[Index] = (u8)Index;
+        }
+        EndTime(Tester);
+        
+        CountBytes(Tester, DestBuffer.Count);
+        
+        HandleDeallocation(Params, &DestBuffer);
+    }
+}

part3/listing_0111_pagefault_overhead_main.cpp

@@ -0,0 +1,127 @@
+/* ========================================================================
+
+   (C) Copyright 2023 by Molly Rocket, Inc., All Rights Reserved.
+   
+   This software is provided 'as-is', without any express or implied
+   warranty. In no event will the authors be held liable for any damages
+   arising from the use of this software.
+   
+   Please see https://computerenhance.com for more information
+   
+   ======================================================================== */
+
+/* ========================================================================
+   LISTING 111
+   ======================================================================== */
+
+/* NOTE(casey): _CRT_SECURE_NO_WARNINGS is here because otherwise we cannot
+   call fopen(). If we replace fopen() with fopen_s() to avoid the warning,
+   then the code doesn't compile on Linux anymore, since fopen_s() does not
+   exist there.
+   
+   What exactly the CRT maintainers were thinking when they made this choice,
+   I have no idea. */
+#define _CRT_SECURE_NO_WARNINGS
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <stdint.h>
+#include <math.h>
+#include <sys/stat.h>
+
+typedef uint8_t u8;
+typedef uint32_t u32;
+typedef uint64_t u64;
+
+typedef int32_t b32;
+
+typedef float f32;
+typedef double f64;
+
+#define ArrayCount(Array) (sizeof(Array)/sizeof((Array)[0]))
+
+#include "listing_0068_buffer.cpp"
+#include "listing_0108_platform_metrics.cpp"
+#include "listing_0109_pagefault_repetition_tester.cpp"
+#include "listing_0106_mallocread_overhead_test.cpp"
+#include "listing_0110_pagefault_overhead_test.cpp"
+
+struct test_function
+{
+    char const *Name;
+    read_overhead_test_func *Func;
+};
+test_function TestFunctions[] =
+{
+    {"WriteToAllBytes", WriteToAllBytes},
+    {"fread", ReadViaFRead},
+    {"_read", ReadViaRead},
+    {"ReadFile", ReadViaReadFile},
+};
+
+int main(int ArgCount, char **Args)
+{
+    // NOTE(casey): Since we do not use these functions in this particular build, we reference their pointers
+    // here to prevent the compiler from complaining about "unused functions".
+    (void)&IsInBounds;
+    (void)&AreEqual;
+
+    InitializeOSMetrics();
+    u64 CPUTimerFreq = EstimateCPUTimerFreq();
+    
+    if(ArgCount == 2)
+    {
+        char *FileName = Args[1];
+#if _WIN32
+        struct __stat64 Stat;
+        _stat64(FileName, &Stat);
+#else
+        struct stat Stat;
+        stat(FileName, &Stat);
+#endif
+        
+        read_parameters Params = {};
+        Params.Dest = AllocateBuffer(Stat.st_size);
+        Params.FileName = FileName;
+    
+        if(Params.Dest.Count > 0)
+        {
+            repetition_tester Testers[ArrayCount(TestFunctions)][AllocType_Count] = {};
+            
+            for(;;)
+            {
+                for(u32 FuncIndex = 0; FuncIndex < ArrayCount(TestFunctions); ++FuncIndex)
+                {
+                    for(u32 AllocType = 0; AllocType < AllocType_Count; ++AllocType)
+                    {
+                        Params.AllocType = (allocation_type)AllocType;
+                        
+                        repetition_tester *Tester = &Testers[FuncIndex][AllocType];
+                        test_function TestFunc = TestFunctions[FuncIndex];
+                        
+                        printf("\n--- %s%s%s ---\n",
+                               DescribeAllocationType(Params.AllocType),
+                               Params.AllocType ? " + " : "",
+                               TestFunc.Name);
+                        NewTestWave(Tester, Params.Dest.Count, CPUTimerFreq);
+                        TestFunc.Func(Tester, &Params);
+                    }
+                }
+            }
+            
+            // NOTE(casey): We would normally call this here, but we can't because the compiler will complain about "unreachable code".
+            // So instead we just reference the pointer to prevent the compiler complaining about unused function :(
+            (void)&FreeBuffer;
+        }
+        else
+        {
+            fprintf(stderr, "ERROR: Test data size must be non-zero\n");
+        }
+    }
+    else
+    {
+        fprintf(stderr, "Usage: %s [existing filename]\n", Args[0]);
+    }
+		
+    return 0;
+}

part3/listing_0112_os_fault_counter_main.cpp

@@ -0,0 +1,89 @@
+/* ========================================================================
+
+   (C) Copyright 2023 by Molly Rocket, Inc., All Rights Reserved.
+   
+   This software is provided 'as-is', without any express or implied
+   warranty. In no event will the authors be held liable for any damages
+   arising from the use of this software.
+   
+   Please see https://computerenhance.com for more information
+   
+   ======================================================================== */
+
+/* ========================================================================
+   LISTING 112
+   ======================================================================== */
+
+/* NOTE(casey): _CRT_SECURE_NO_WARNINGS is here because otherwise we cannot
+   call fopen(). If we replace fopen() with fopen_s() to avoid the warning,
+   then the code doesn't compile on Linux anymore, since fopen_s() does not
+   exist there.
+   
+   What exactly the CRT maintainers were thinking when they made this choice,
+   I have no idea. */
+#define _CRT_SECURE_NO_WARNINGS
+
+#include <stdio.h>
+#include <stdint.h>
+
+typedef uint8_t u8;
+typedef uint32_t u32;
+typedef uint64_t u64;
+
+typedef int32_t b32;
+
+typedef float f32;
+typedef double f64;
+
+#define ArrayCount(Array) (sizeof(Array)/sizeof((Array)[0]))
+
+#include "listing_0108_platform_metrics.cpp"
+
+int main(int ArgCount, char **Args)
+{
+    // NOTE(casey): Since we do not use these functions in this particular build, we reference their pointers
+    // here to prevent the compiler from complaining about "unused functions".
+    (void)&EstimateCPUTimerFreq;
+    
+    InitializeOSMetrics();
+    
+    if(ArgCount == 2)
+    {
+        u64 PageSize = 4096; // NOTE(casey): This may not be the OS page size! It is merely our testing page size.
+        u64 PageCount = atol(Args[1]);
+        u64 TotalSize = PageSize*PageCount;
+        
+        printf("Page Count, Touch Count, Fault Count, Extra Faults\n");
+        
+        for(u64 TouchCount = 0; TouchCount <= PageCount; ++TouchCount)
+        {
+            u64 TouchSize = PageSize*TouchCount;
+            u8 *Data = (u8 *)VirtualAlloc(0, TotalSize, MEM_RESERVE|MEM_COMMIT, PAGE_READWRITE);
+            if(Data)
+            {
+                u64 StartFaultCount = ReadOSPageFaultCount();
+                for(u64 Index = 0; Index < TouchSize; ++Index)
+                {
+                    Data[Index] = (u8)Index;
+                }
+                u64 EndFaultCount = ReadOSPageFaultCount();
+                
+                u64 FaultCount = EndFaultCount - StartFaultCount;
+                
+                printf("%llu, %llu, %llu, %lld\n", PageCount, TouchCount, FaultCount, (FaultCount - TouchCount));
+                
+                VirtualFree(Data, 0, MEM_RELEASE);
+            }
+            else
+            {
+                fprintf(stderr, "ERROR: Unable to allocate memory\n");
+            }
+        }
+    }
+    else
+    {
+        fprintf(stderr, "Usage: %s [# of 4k pages to allocate]\n", Args[0]);
+    }
+		
+    return 0;
+}

part3/listing_0113_os_fault_counter_backward_main.cpp

@@ -0,0 +1,89 @@
+/* ========================================================================
+
+   (C) Copyright 2023 by Molly Rocket, Inc., All Rights Reserved.
+   
+   This software is provided 'as-is', without any express or implied
+   warranty. In no event will the authors be held liable for any damages
+   arising from the use of this software.
+   
+   Please see https://computerenhance.com for more information
+   
+   ======================================================================== */
+
+/* ========================================================================
+   LISTING 113
+   ======================================================================== */
+
+/* NOTE(casey): _CRT_SECURE_NO_WARNINGS is here because otherwise we cannot
+   call fopen(). If we replace fopen() with fopen_s() to avoid the warning,
+   then the code doesn't compile on Linux anymore, since fopen_s() does not
+   exist there.
+   
+   What exactly the CRT maintainers were thinking when they made this choice,
+   I have no idea. */
+#define _CRT_SECURE_NO_WARNINGS
+
+#include <stdio.h>
+#include <stdint.h>
+
+typedef uint8_t u8;
+typedef uint32_t u32;
+typedef uint64_t u64;
+
+typedef int32_t b32;
+
+typedef float f32;
+typedef double f64;
+
+#define ArrayCount(Array) (sizeof(Array)/sizeof((Array)[0]))
+
+#include "listing_0108_platform_metrics.cpp"
+
+int main(int ArgCount, char **Args)
+{
+    // NOTE(casey): Since we do not use these functions in this particular build, we reference their pointers
+    // here to prevent the compiler from complaining about "unused functions".
+    (void)&EstimateCPUTimerFreq;
+    
+    InitializeOSMetrics();
+    
+    if(ArgCount == 2)
+    {
+        u64 PageSize = 4096; // NOTE(casey): This may not be the OS page size! It is merely our testing page size.
+        u64 PageCount = atol(Args[1]);
+        u64 TotalSize = PageSize*PageCount;
+        
+        printf("Page Count, Touch Count, Fault Count, Extra Faults\n");
+        
+        for(u64 TouchCount = 0; TouchCount <= PageCount; ++TouchCount)
+        {
+            u64 TouchSize = PageSize*TouchCount;
+            u8 *Data = (u8 *)VirtualAlloc(0, TotalSize, MEM_RESERVE|MEM_COMMIT, PAGE_READWRITE);
+            if(Data)
+            {
+                u64 StartFaultCount = ReadOSPageFaultCount();
+                for(u64 Index = 0; Index < TouchSize; ++Index)
+                {
+                    Data[TotalSize - 1 - Index] = (u8)Index;
+                }
+                u64 EndFaultCount = ReadOSPageFaultCount();
+                
+                u64 FaultCount = EndFaultCount - StartFaultCount;
+                
+                printf("%llu, %llu, %llu, %lld\n", PageCount, TouchCount, FaultCount, (FaultCount - TouchCount));
+                
+                VirtualFree(Data, 0, MEM_RELEASE);
+            }
+            else
+            {
+                fprintf(stderr, "ERROR: Unable to allocate memory\n");
+            }
+        }
+    }
+    else
+    {
+        fprintf(stderr, "Usage: %s [# of 4k pages to allocate]\n", Args[0]);
+    }
+		
+    return 0;
+}

part3/repetition_tester.c

@@ -7,7 +7,7 @@
 #include <sys/stat.h>
 
 #include "base.h"
-#include "listing_0074_platform_metrics.cpp"
+#include "listing_0108_platform_metrics.cpp"
 #include "base.c"
 #include "repetition_tester.h"
 
@@ -63,18 +63,20 @@ void RepTester_Error(RepTester *tester, Str8 msg)
 void RepTester_BeginTime(RepTester *tester)
 {
     tester->open_block_count++;
-    tester->time_accumulated_on_this_test -= ReadCPUTimer();
+    tester->accumulated_on_this_test.e[RepTesterValueType_CPUTimer] -= ReadCPUTimer();
+    tester->accumulated_on_this_test.e[RepTesterValueType_MemPageFaults] -= ReadOSPageFaultCount();
 }
 
 void RepTester_EndTime(RepTester *tester)
 {
     tester->close_block_count++;
-    tester->time_accumulated_on_this_test += ReadCPUTimer();
+    tester->accumulated_on_this_test.e[RepTesterValueType_CPUTimer] += ReadCPUTimer();
+    tester->accumulated_on_this_test.e[RepTesterValueType_MemPageFaults] += ReadOSPageFaultCount();
 }
 
 void RepTester_CountBytes(RepTester *tester, size_t bytes_read)
 {
-    tester->total_bytes_read += bytes_read;
+    tester->accumulated_on_this_test.e[RepTesterValueType_ByteCount] += bytes_read;
 }
 
 static void PrintTime(Str8 label, f64 cpu_time, u64 cpu_timer_freq, u64 byte_count)
@@ -92,18 +94,42 @@ static void PrintTime(Str8 label, f64 cpu_time, u64 cpu_timer_freq, u64 byte_cou
     }
 }
 
-static void RepTester_PrintResult(RepTesterResults results, u64 cpu_timer_freq, u64 byte_count)
+static void RepTester_PrintValue(Str8 label, RepTesterValue value, u64 cpu_timer_freq)
 {
-    PrintTime(STR8("Min"), (f64)results.min_time, cpu_timer_freq, byte_count);
+    u64 test_count = value.e[RepTesterValueType_TestCount];
-    printf("\n");
+    f64 divisor    = test_count ? (f64)test_count : 1;
 
-    PrintTime(STR8("Max"), (f64)results.max_time, cpu_timer_freq, byte_count);
+    f64 avg_values[ARRAY_UCOUNT(value.e)];
-    printf("\n");
+    for (u64 index = 0; index < ARRAY_UCOUNT(value.e); index++)
+        avg_values[index] = value.e[index] / divisor;
 
-    if (results.test_count) {
+    printf("%.*s: %.0f", STR8_FMT(label), avg_values[RepTesterValueType_CPUTimer]);
-        PrintTime(STR8("Avg"), (f64)results.total_time / (f64)results.test_count, cpu_timer_freq, byte_count);
+    if (cpu_timer_freq) {
-        printf("\n");
+        f64 seconds = avg_values[RepTesterValueType_CPUTimer] / cpu_timer_freq;
+        printf(" (%fms)", 1000.0f*seconds);
+
+        if (avg_values[RepTesterValueType_ByteCount] > 0) {
+            f64 gigabyte  = (1024.0f * 1024.0f * 1024.0f);
+            f64 bandwidth = avg_values[RepTesterValueType_ByteCount] / (gigabyte * seconds);
+            printf(" %fgb/s", bandwidth);
+        }
     }
+
+    if(avg_values[RepTesterValueType_MemPageFaults] > 0) {
+        printf(" PF: %0.4f (%0.4fk/fault)",
+               avg_values[RepTesterValueType_MemPageFaults],
+               avg_values[RepTesterValueType_ByteCount] / (avg_values[RepTesterValueType_MemPageFaults] * 1024.0));
+    }
+}
+
+static void RepTester_PrintResult(RepTesterResults results, u64 cpu_timer_freq)
+{
+    RepTester_PrintValue(STR8("Min"), results.min, cpu_timer_freq);
+    printf("\n");
+    RepTester_PrintValue(STR8("Max"), results.max, cpu_timer_freq);
+    printf("\n");
+    RepTester_PrintValue(STR8("Avg"), results.total, cpu_timer_freq);
+    printf("\n");
 }
 
 bool RepTester_IsTesting(RepTester *tester)
@@ -114,29 +140,34 @@ bool RepTester_IsTesting(RepTester *tester)
             if (tester->open_block_count != tester->close_block_count)
                 RepTester_Error(tester, STR8("Unbalanced begin/end time"));
 
-            if (tester->total_bytes_read != tester->desired_bytes_read)
+            RepTesterValue accum = tester->accumulated_on_this_test;
+            if (accum.e[RepTesterValueType_ByteCount] != tester->desired_bytes_read)
                 RepTester_Error(tester, STR8("Processed byte count mismatch"));
 
             if (tester->mode == RepTesterMode_Testing) {
-                RepTesterResults *results       = &tester->results;
+                RepTesterResults *results = &tester->results;
-                u64               elapsed_time  = tester->time_accumulated_on_this_test;
+                accum.e[RepTesterValueType_TestCount] = 1;
-                results->test_count            += 1;
-                results->total_time            += elapsed_time;
-                results->max_time               = MAX(results->max_time, elapsed_time);
 
-                if (results->min_time > elapsed_time) {
+                for (u64 index = 0; index < ARRAY_UCOUNT(accum.e); index++) {
-                    results->min_time = elapsed_time;
+                    results->total.e[index] += accum.e[index];
+                }
+
+                if (results->max.e[RepTesterValueType_CPUTimer] < accum.e[RepTesterValueType_CPUTimer]) {
+                    results->max = accum;
+                }
+
+                if (results->min.e[RepTesterValueType_CPUTimer] > accum.e[RepTesterValueType_CPUTimer]) {
+                    results->min = accum;
 
                     // NOTE: Reset the trial time when new min time is achieved
                     tester->start_time = current_time;
-                    PrintTime(STR8("Min"), (f64)results->min_time, tester->cpu_timer_freq, tester->desired_bytes_read);
+                    RepTester_PrintValue(STR8("Min"), results->min, tester->cpu_timer_freq);
-                    printf("               \r");
+                    printf("                                                  \r");
                 }
 
-                tester->open_block_count              = 0;
+                tester->open_block_count         = 0;
-                tester->close_block_count             = 0;
+                tester->close_block_count        = 0;
-                tester->time_accumulated_on_this_test = 0;
+                tester->accumulated_on_this_test = (RepTesterValue){};
-                tester->total_bytes_read              = 0;
             }
         }
 
@@ -144,7 +175,7 @@ bool RepTester_IsTesting(RepTester *tester)
             tester->mode = RepTesterMode_Complete;
 
             printf("                                                          \r");
-            RepTester_PrintResult(tester->results, tester->cpu_timer_freq, tester->desired_bytes_read);
+            RepTester_PrintResult(tester->results, tester->cpu_timer_freq);
         }
     }
 
@@ -260,6 +291,20 @@ static void ReadWithReadFile(RepTester *tester, ReadArgs *args)
     }
 }
 
+static void WriteToAllBytes(RepTester *tester, ReadArgs *args)
+{
+    while (RepTester_IsTesting(tester)) {
+        Buffer buffer = args->dest;
+        TestAlloc(args, &buffer);
+        RepTester_BeginTime(tester);
+        for (u64 index = 0; index < buffer.size; index++)
+            buffer.data[index] = (u8)index;
+        RepTester_EndTime(tester);
+        RepTester_CountBytes(tester, buffer.size);
+        TestDealloc(args, &buffer);
+    }
+}
+
 typedef void TestFuncPtr(RepTester *tester, ReadArgs *args);
 typedef struct TestFunction {
     Str8         name;
@@ -292,17 +337,20 @@ int main(int argc, char const **argv)
         return -1;
     }
 
+    InitializeOSMetrics();
+
     TestFunction test_functions[] = {
-        {STR8("fread"),    ReadWithFRead},
+        {STR8("WriteToAllBytes"), WriteToAllBytes},
-        {STR8("_read"),    ReadWithRead},
+        {STR8("fread"),           ReadWithFRead},
-        {STR8("ReadFile"), ReadWithReadFile},
+        {STR8("_read"),           ReadWithRead},
+        {STR8("ReadFile"),        ReadWithReadFile},
     };
 
     RepTester testers[ARRAY_UCOUNT(test_functions)][AllocType_Count] = {};
     u64       cpu_timer_freq                                         = EstimateCPUTimerFreq();
     for (u64 index = 0; index != UINT64_MAX; index++) {
         for (u64 func_index = 0; func_index < ARRAY_UCOUNT(testers); func_index++) {
-            for (u64 alloc_index = AllocType_VirtualAlloc; alloc_index < AllocType_Count; alloc_index++) {
+            for (u64 alloc_index = 0; alloc_index < AllocType_Count; alloc_index++) {
                 RepTester   *tester    = &testers[func_index][alloc_index];
                 TestFunction test_func = test_functions[func_index];
 
@@ -311,10 +359,10 @@ int main(int argc, char const **argv)
                        STR8_FMT(test_func.name));
 
                 if (tester->mode == RepTesterMode_Nil) {
-                    tester->mode               = RepTesterMode_Testing;
+                    tester->mode                                       = RepTesterMode_Testing;
-                    tester->desired_bytes_read = args.dest.size;
+                    tester->desired_bytes_read                         = args.dest.size;
-                    tester->cpu_timer_freq     = cpu_timer_freq;
+                    tester->cpu_timer_freq                             = cpu_timer_freq;
-                    tester->results.min_time   = (u64)-1;
+                    tester->results.min.e[RepTesterValueType_CPUTimer] = (u64)-1;
                 } else if (tester->mode == RepTesterMode_Complete) {
                     tester->mode = RepTesterMode_Testing;
 

part3/repetition_tester.h

@@ -5,11 +5,22 @@ typedef enum RepTesterMode {
     RepTesterMode_Complete,
 } RepTesterMode;
 
+typedef enum RepTesterValueType {
+    RepTesterValueType_TestCount,
+    RepTesterValueType_CPUTimer,
+    RepTesterValueType_MemPageFaults,
+    RepTesterValueType_ByteCount,
+    RepTesterValueType_Count,
+} RepTesterValueType;
+
+typedef struct RepTesterValue {
+    u64 e[RepTesterValueType_Count];
+} RepTesterValue;
+
 typedef struct RepTesterResults {
-    u64 test_count;
+    RepTesterValue total;
-    u64 total_time;
+    RepTesterValue max;
-    u64 max_time;
+    RepTesterValue min;
-    u64 min_time;
 } RepTesterResults;
 
 typedef struct RepTester {
@@ -17,12 +28,12 @@ typedef struct RepTester {
     u32              open_block_count;
     u32              close_block_count;
     u64              cpu_timer_freq;
-    RepTesterResults results;
 
-    size_t           total_bytes_read;
+    RepTesterResults results;
+    RepTesterValue   accumulated_on_this_test;
+
     size_t           desired_bytes_read;
 
-    u64              time_accumulated_on_this_test;
     u64              start_time;
     u64              run_duration;
 } RepTester;