From 78ce7e567fce86d611353c5a9194833a54a6fbe0 Mon Sep 17 00:00:00 2001
From: Marius Hillenbrand <mhillen@linux.ibm.com>
Date: Tue, 19 Oct 2021 18:09:52 +0200
Subject: [PATCH] Fix encoding/decoding of string literals for big-endian
 systems

Per SPIR-V spec, a string literal's UTF-8 octets are encoded packed into
words with little-endian convention. Explicitly perform that encoding
instead of assuming that the host system is little-endian.

Note that this change requires corresponding fixes in SPIRV-Tools.

Fixes #202
---
 SPIRV/SPVRemapper.cpp | 18 +++++++++++------
 SPIRV/disassemble.cpp | 47 ++++++++++++++++++++++++++++---------------
 SPIRV/spvIR.h         | 22 +++++++++-----------
 3 files changed, 52 insertions(+), 35 deletions(-)

diff --git a/SPIRV/SPVRemapper.cpp b/SPIRV/SPVRemapper.cpp
index 56d6d5d4a..fdfbeb90c 100644
--- a/SPIRV/SPVRemapper.cpp
+++ b/SPIRV/SPVRemapper.cpp
@@ -297,15 +297,21 @@ namespace spv {
     std::string spirvbin_t::literalString(unsigned word) const
     {
         std::string literal;
+        const spirword_t * pos = spv.data() + word;

         literal.reserve(16);

-        const char* bytes = reinterpret_cast<const char*>(spv.data() + word);
-
-        while (bytes && *bytes)
-            literal += *bytes++;
-
-        return literal;
+        do {
+            spirword_t word = *pos;
+            for (int i = 0; i < 4; i++) {
+                char c = word & 0xff;
+                if (c == '\0')
+                    return literal;
+                literal += c;
+                word >>= 8;
+            }
+            pos++;
+        } while (true);
     }

     void spirvbin_t::applyMap()
diff --git a/SPIRV/disassemble.cpp b/SPIRV/disassemble.cpp
index 73c988c5b..74dd60540 100644
--- a/SPIRV/disassemble.cpp
+++ b/SPIRV/disassemble.cpp
@@ -43,6 +43,7 @@
 #include <stack>
 #include <sstream>
 #include <cstring>
+#include <utility>

 #include "disassemble.h"
 #include "doc.h"
@@ -100,6 +101,7 @@ class SpirvStream {
     void outputMask(OperandClass operandClass, unsigned mask);
     void disassembleImmediates(int numOperands);
     void disassembleIds(int numOperands);
+    std::pair<int, std::string> decodeString();
     int disassembleString();
     void disassembleInstruction(Id resultId, Id typeId, Op opCode, int numOperands);

@@ -290,31 +292,44 @@ void SpirvStream::disassembleIds(int numOperands)
     }
 }

-// return the number of operands consumed by the string
-int SpirvStream::disassembleString()
+// decode string from words at current position (non-consuming)
+std::pair<int, std::string> SpirvStream::decodeString()
 {
-    int startWord = word;
-
-    out << " \"";
-
-    const char* wordString;
+    std::string res;
+    int wordPos = word;
+    char c;
     bool done = false;
+
     do {
-        unsigned int content = stream[word];
-        wordString = (const char*)&content;
+        unsigned int content = stream[wordPos];
         for (int charCount = 0; charCount < 4; ++charCount) {
-            if (*wordString == 0) {
+            c = content & 0xff;
+            content >>= 8;
+            if (c == '\0') {
                 done = true;
                 break;
             }
-            out << *(wordString++);
+            res += c;
         }
-        ++word;
-    } while (! done);
+        ++wordPos;
+    } while(! done);
+
+    return std::make_pair(wordPos - word, res);
+}
+
+// return the number of operands consumed by the string
+int SpirvStream::disassembleString()
+{
+    out << " \"";

+    std::pair<int, std::string> decoderes = decodeString();
+
+    out << decoderes.second;
     out << "\"";

-    return word - startWord;
+    word += decoderes.first;
+
+    return decoderes.first;
 }

 void SpirvStream::disassembleInstruction(Id resultId, Id /*typeId*/, Op opCode, int numOperands)
@@ -331,7 +346,7 @@ void SpirvStream::disassembleInstruction(Id resultId, Id /*typeId*/, Op opCode,
             nextNestedControl = 0;
         }
     } else if (opCode == OpExtInstImport) {
-        idDescriptor[resultId] = (const char*)(&stream[word]);
+        idDescriptor[resultId] = decodeString().second;
     }
     else {
         if (resultId != 0 && idDescriptor[resultId].size() == 0) {
@@ -428,7 +443,7 @@ void SpirvStream::disassembleInstruction(Id resultId, Id /*typeId*/, Op opCode,
             --numOperands;
             // Get names for printing "(XXX)" for readability, *after* this id
             if (opCode == OpName)
-                idDescriptor[stream[word - 1]] = (const char*)(&stream[word]);
+                idDescriptor[stream[word - 1]] = decodeString().second;
             break;
         case OperandVariableIds:
             disassembleIds(numOperands);
diff --git a/SPIRV/spvIR.h b/SPIRV/spvIR.h
index 486e80d00..5249a5ba7 100644
--- a/SPIRV/spvIR.h
+++ b/SPIRV/spvIR.h
@@ -111,27 +111,23 @@ class Instruction {

     void addStringOperand(const char* str)
     {
-        unsigned int word;
-        char* wordString = (char*)&word;
-        char* wordPtr = wordString;
-        int charCount = 0;
+        unsigned int word = 0;
+        unsigned int shiftAmount = 0;
         char c;
+
         do {
             c = *(str++);
-            *(wordPtr++) = c;
-            ++charCount;
-            if (charCount == 4) {
+            word |= ((unsigned int)c) << shiftAmount;
+            shiftAmount += 8;
+            if (shiftAmount == 32) {
                 addImmediateOperand(word);
-                wordPtr = wordString;
-                charCount = 0;
+                word = 0;
+                shiftAmount = 0;
             }
         } while (c != 0);

         // deal with partial last word
-        if (charCount > 0) {
-            // pad with 0s
-            for (; charCount < 4; ++charCount)
-                *(wordPtr++) = 0;
+        if (shiftAmount > 0) {
             addImmediateOperand(word);
         }
     }