Merge pull request #26 from Decompollaborate/develop

1.6.0

.gitattributes

@@ -1,2 +1,33 @@
 # Auto detect text files and perform LF normalization
 * text=auto
+
+src/instructions/RabbitizerInstruction/instrOpercandCallbacks_array.table.h linguist-generated=true
+
+src/instructions/InstrCategory_Names_array.table.h linguist-generated=true
+src/instructions/InstrDescriptor_Descriptors_array.table.h linguist-generated=true
+src/instructions/InstrId_Names_array.table.h linguist-generated=true
+src/instructions/RegisterDescriptor_Descriptors_arrays.table.h linguist-generated=true
+src/instructions/Registers_Names_arrays.table.h linguist-generated=true
+
+include/common/Abi_enum.table.h linguist-generated=true
+
+include/instructions/AccessType_enum.table.h linguist-generated=true
+include/instructions/InstrCategory_enum.table.h linguist-generated=true
+include/instructions/InstrId_enum.table.h linguist-generated=true
+include/instructions/InstrSuffix_enum.table.h linguist-generated=true
+include/instructions/OperandType_enum.table.h linguist-generated=true
+include/instructions/OperandType_function_declarations.table.h linguist-generated=true
+include/instructions/Registers_enums.table.h linguist-generated=true
+
+cplusplus/include/instructions/AccessType_enum_class.table.h linguist-generated=true
+cplusplus/include/instructions/OperandType_enum_class.table.h linguist-generated=true
+cplusplus/include/instructions/Registers_enum_classes.table.h linguist-generated=true
+cplusplus/include/instructions/UniqueId_enum_class.table.h linguist-generated=true
+
+rust/src/abi_enum.rs linguist-generated=true
+rust/src/access_type_enum.rs linguist-generated=true
+rust/src/instr_category_enum.rs linguist-generated=true
+rust/src/instr_id_enum.rs linguist-generated=true
+rust/src/instr_suffix_enum.rs linguist-generated=true
+rust/src/operand_type_enum.rs linguist-generated=true
+rust/src/registers_enum.rs linguist-generated=true

Cargo.toml

@@ -4,7 +4,7 @@
 [package]
 name = "rabbitizer"
 # Version should be synced with include/common/RabbitizerVersion.h
-version = "1.5.11"
+version = "1.6.0"
 edition = "2021"
 authors = ["Anghelo Carvajal <[email protected]>"]
 description = "MIPS instruction decoder"

include/common/RabbitizerConfig.h

@@ -49,6 +49,11 @@ typedef struct RabbitizerConfig_ToolchainTweaks {
      * So we replace break instrutions for SN64 with the exact word that the assembler generates when expanding div
      */
     bool sn64DivFix;
+    /**
+     * Enables various tweaks to allow building matching with GNU as which
+     * break original compiler behavior and what's specified in the manuals.
+     */
+    bool gnuMode;
 } RabbitizerConfig_ToolchainTweaks;
 
 typedef struct RabbitizerConfig_Misc {

include/common/RabbitizerVersion.h

@@ -13,8 +13,8 @@ extern "C" {
 
 // Header version
 #define RAB_VERSION_MAJOR 1
-#define RAB_VERSION_MINOR 5
-#define RAB_VERSION_PATCH 11
+#define RAB_VERSION_MINOR 6
+#define RAB_VERSION_PATCH 0
 
 #define RAB_VERSION_STR RAB_STRINGIFY(RAB_VERSION_MAJOR) "." RAB_STRINGIFY(RAB_VERSION_MINOR) "." RAB_STRINGIFY(RAB_VERSION_PATCH)
 

include/instructions/instr_id/r5900/r5900_cop1_fpu_s.inc

@@ -103,19 +103,6 @@ RABBITIZER_DEF_INSTR_ID_ALTNAME(
     .readsFt=true
 ) // floating point MINimum
 
-// Due to the R5900's FPU being non properly complaint the instruction cvt.w.s always behaves as trunc.w.s because is because EE can only do round-to-zero.
-// Assemblers like GAS workaround this issue by decoding cvt.w.s as trunc.w.s, so we mimic that behaviour to allow assembling with GAS.
-// Here's some reading about the binutils rationale:
-// https://sourceware.org/legacy-ml/binutils/2012-11/msg00360.html
-// https://sourceware.org/pipermail/binutils/2013-January/079863.html
-RABBITIZER_DEF_INSTR_ID_ALTNAME(
-    r5900, 0x24, trunc_w_s, trunc.w.s,
-    .operands={RAB_OPERAND_cpu_fd, RAB_OPERAND_cpu_fs},
-    .isFloat=true,
-    .modifiesFd=true,
-    .readsFs=true
-) // Floating-Point Truncate to Word Fixed-Point
-
 RABBITIZER_DEF_INSTR_ID_ALTNAME(
     r5900, 0x34, c_lt_s, c.lt.s,
     .operands={RAB_OPERAND_cpu_fs, RAB_OPERAND_cpu_ft},

pyproject.toml

@@ -4,7 +4,7 @@
 [project]
 name = "rabbitizer"
 # Version should be synced with include/common/RabbitizerVersion.h
-version = "1.5.11"
+version = "1.6.0"
 description = "MIPS instruction decoder"
 # license = "MIT"
 readme = "README.md"

rabbitizer/Config.pyi

@@ -35,8 +35,9 @@ class _RabbitizerConfig:
     pseudos_pseudoNegu: bool = True
     pseudos_pseudoBal: bool = True
 
-    toolchainTweaks_sn64DivFix: bool = False
     toolchainTweaks_treatJAsUnconditionalBranch: bool = False
+    toolchainTweaks_sn64DivFix: bool = False
+    toolchainTweaks_gnuMode: bool = True
 
     misc_opcodeLJust: int = 11
     misc_unknownInstrComment: bool = True

rabbitizer/rabbitizer_global_config.c

@@ -118,6 +118,7 @@ DEF_MEMBER_GET_SET_BOOL(pseudos, pseudoBal)
 
 DEF_MEMBER_GET_SET_BOOL(toolchainTweaks, treatJAsUnconditionalBranch)
 DEF_MEMBER_GET_SET_BOOL(toolchainTweaks, sn64DivFix)
+DEF_MEMBER_GET_SET_BOOL(toolchainTweaks, gnuMode)
 
 DEF_MEMBER_GET_SET_INT(misc, opcodeLJust, false, 0, 0)
 DEF_MEMBER_GET_SET_BOOL(misc, unknownInstrComment)
@@ -144,6 +145,7 @@ static PyGetSetDef rabbitizer_global_config_GetSets[] = {
 
     MEMBER_GET_SET(toolchainTweaks, treatJAsUnconditionalBranch, "", NULL),
     MEMBER_GET_SET(toolchainTweaks, sn64DivFix, "", NULL),
+    MEMBER_GET_SET(toolchainTweaks, gnuMode, "", NULL),
 
     MEMBER_GET_SET(misc, opcodeLJust, "", NULL),
     MEMBER_GET_SET(misc, unknownInstrComment, "", NULL),

rust/src/config.rs

@@ -35,6 +35,7 @@ pub struct PseudoInstr {
 pub struct ToolchainTweaks {
     pub treat_j_as_unconditional_branch: bool,
     pub sn64_div_fix: bool,
+    pub gnu_mode: bool,
 }
 
 #[repr(C)]

src/common/RabbitizerConfig.c

@@ -40,6 +40,7 @@ RabbitizerConfig RabbitizerConfig_Cfg = {
     .toolchainTweaks = {
         .treatJAsUnconditionalBranch = true,
         .sn64DivFix = false,
+        .gnuMode = true,
     },
     .misc = {
         .opcodeLJust = 7+4,

src/instructions/RabbitizerInstruction/RabbitizerInstruction_Disassemble.c

@@ -95,6 +95,38 @@ bool RabbitizerInstruction_mustDisasmAsData(const RabbitizerInstruction *self) {
         }
     }
 
+    if (RabbitizerConfig_Cfg.toolchainTweaks.gnuMode) {
+        switch (self->uniqueId) {
+            case RABBITIZER_INSTR_ID_cpu_trunc_w_s:
+            case RABBITIZER_INSTR_ID_cpu_cvt_w_s:
+                if (self->category == RABBITIZER_INSTRCAT_R5900) {
+                    /**
+                     * Due to the R5900's FPU being non properly complaint, the instruction cvt.w.s always behaves as
+                     * trunc.w.s because EE can only do round-to-zero.
+                     *
+                     * Assemblers like GAS workaround this issue by decoding cvt.w.s as trunc.w.s, but other assemblers
+                     * just use trunc.w.s and cvt.w.s as-is.
+                     *
+                     * Here's some reading about the binutils rationale:
+                     * - https://sourceware.org/legacy-ml/binutils/2012-11/msg00360.html
+                     * - https://sourceware.org/pipermail/binutils/2013-January/079863.html
+                     *
+                     * Because of this, building with GAS with the -march=r5900 flag produces:
+                     * - trunc.w.s is built as the cvt.w.s instruction.
+                     * - cvt.w.s errors complaining as not being supported by the processor.
+                     *
+                     * To ensure the produced disassembly will still match when built with GAS, we decode this two
+                     * instructions as .word
+                     */
+                    return true;
+                }
+                break;
+
+            default:
+                break;
+        }
+    }
+
     if (!RabbitizerInstruction_isValid(self)) {
         return true;
     }

src/instructions/RabbitizerInstruction/RabbitizerInstruction_ProcessUniqueId.c

@@ -101,13 +101,15 @@ void RabbitizerInstruction_processUniqueId_Special(RabbitizerInstruction *self)
 
     switch (self->uniqueId) {
         case RABBITIZER_INSTR_ID_cpu_div:
-            if (RabbitizerConfig_Cfg.toolchainTweaks.sn64DivFix && !self->inHandwrittenFunction) {
+            if ((!RabbitizerConfig_Cfg.toolchainTweaks.gnuMode) ||
+                (RabbitizerConfig_Cfg.toolchainTweaks.sn64DivFix && !self->inHandwrittenFunction)) {
                 self->descriptor = &RabbitizerInstrDescriptor_Descriptors[RABBITIZER_INSTR_ID_cpu_sn64_div];
             }
             break;
 
         case RABBITIZER_INSTR_ID_cpu_divu:
-            if (RabbitizerConfig_Cfg.toolchainTweaks.sn64DivFix && !self->inHandwrittenFunction) {
+            if ((!RabbitizerConfig_Cfg.toolchainTweaks.gnuMode) ||
+                (RabbitizerConfig_Cfg.toolchainTweaks.sn64DivFix && !self->inHandwrittenFunction)) {
                 self->descriptor = &RabbitizerInstrDescriptor_Descriptors[RABBITIZER_INSTR_ID_cpu_sn64_divu];
             }
             break;

src/instructions/RabbitizerInstructionR5900/RabbitizerInstructionR5900_OperandType.c

@@ -15,7 +15,7 @@ size_t RabbitizerOperandType_process_r5900_I(UNUSED const RabbitizerInstruction
                                              UNUSED const char *immOverride, UNUSED size_t immOverrideLength) {
     size_t totalSize = 0;
 
-    RABUTILS_BUFFER_CPY(dst, totalSize, "$I");
+    RABUTILS_BUFFER_CPY(dst, totalSize, "I");
 
     return totalSize;
 }
@@ -24,7 +24,7 @@ size_t RabbitizerOperandType_process_r5900_Q(UNUSED const RabbitizerInstruction
                                              UNUSED const char *immOverride, UNUSED size_t immOverrideLength) {
     size_t totalSize = 0;
 
-    RABUTILS_BUFFER_CPY(dst, totalSize, "$Q");
+    RABUTILS_BUFFER_CPY(dst, totalSize, "Q");
 
     return totalSize;
 }
@@ -33,7 +33,7 @@ size_t RabbitizerOperandType_process_r5900_R(UNUSED const RabbitizerInstruction
                                              UNUSED const char *immOverride, UNUSED size_t immOverrideLength) {
     size_t totalSize = 0;
 
-    RABUTILS_BUFFER_CPY(dst, totalSize, "$R");
+    RABUTILS_BUFFER_CPY(dst, totalSize, "R");
 
     return totalSize;
 }
@@ -42,7 +42,7 @@ size_t RabbitizerOperandType_process_r5900_ACC(UNUSED const RabbitizerInstructio
                                                UNUSED const char *immOverride, UNUSED size_t immOverrideLength) {
     size_t totalSize = 0;
 
-    RABUTILS_BUFFER_CPY(dst, totalSize, "$ACC");
+    RABUTILS_BUFFER_CPY(dst, totalSize, "ACC");
 
     return totalSize;
 }

src/instructions/RabbitizerInstructionR5900/RabbitizerInstructionR5900_ProcessUniqueId.c

@@ -2,6 +2,7 @@
 /* SPDX-License-Identifier: MIT */
 
 #include "instructions/RabbitizerInstructionR5900.h"
+#include "common/RabbitizerConfig.h"
 
 #define RABBITIZER_DEF_INSTR_ID(prefix, caseBits, name, ...)    \
     case (caseBits):                                            \

tests/asm/r5900/trunc_w_c.s

@@ -0,0 +1,15 @@
+.set noreorder
+
+.section .text
+
+.global test
+trunc.w.s $f0, $f12
+jr $31
+nop
+
+.global test2
+cvt.w.s $f0, $f12
+jr $31
+nop
+
+

tests/c/instruction_checks/r5900_trunc_cvt.c

@@ -0,0 +1,58 @@
+/* SPDX-FileCopyrightText: © 2023 Decompollaborate */
+/* SPDX-License-Identifier: MIT */
+
+#include "rabbitizer.h"
+
+#include <string.h>
+#include <stdlib.h>
+#include <assert.h>
+
+
+typedef struct TestEntry {
+    uint32_t word;
+    bool gnuMode;
+    const char *expectedStr;
+} TestEntry;
+
+
+const TestEntry entries[] = {
+    { 0x4600600D, true, ".word       0x4600600D                   # trunc.w.s   $f0, $f12 # 00000000" },
+    { 0x46006024, true, ".word       0x46006024                   # cvt.w.s     $f0, $f12 # 00000000" },
+    { 0x4600600D, false, "trunc.w.s   $f0, $f12" },
+    { 0x46006024, false, "cvt.w.s     $f0, $f12" },
+};
+
+int main() {
+    int errorCount = 0;
+    size_t i;
+
+    for (i = 0; i < ARRAY_COUNT(entries); i++) {
+        const TestEntry *entry = &entries[i];
+        RabbitizerConfig_Cfg.toolchainTweaks.gnuMode = entry->gnuMode;
+        RabbitizerInstruction instr;
+        char *buffer;
+        size_t bufferSize;
+
+        RabbitizerInstructionR5900_init(&instr, entry->word, 0x80000000);
+        RabbitizerInstructionR5900_processUniqueId(&instr);
+
+        bufferSize = RabbitizerInstruction_getSizeForBuffer(&instr, 0, 0);
+        buffer = malloc(bufferSize + 1);
+        assert(buffer != NULL);
+
+        RabbitizerInstruction_disassemble(&instr, buffer, NULL, 0, 0);
+
+        if (entry->expectedStr == NULL) {
+            printf("Word '0x%08X' gnuMode '%s' doesn't have a expected str, got '%s'\n", entry->word, entry->gnuMode ? "true" : "false", buffer);
+            errorCount++;
+        } else if (strcmp(buffer, entry->expectedStr) != 0) {
+            fprintf(stderr, "Error on word '0x%08X' gnuMode '%s'. Expected '%s', got '%s'\n", entry->word, entry->gnuMode ? "true" : "false", entry->expectedStr, buffer);
+            errorCount++;
+        }
+
+        free(buffer);
+        RabbitizerInstructionR5900_destroy(&instr);
+    }
+
+    return errorCount;
+}

tests/run_instruction_checks.sh

@@ -1,9 +1,10 @@
 #!/bin/bash
 
-# SPDX-FileCopyrightText: © 2022 Decompollaborate
+# SPDX-FileCopyrightText: © 2022-2023 Decompollaborate
 # SPDX-License-Identifier: MIT
 
 set -e
 
 ./build/tests/c/instruction_checks/jalr.elf
 ./build/tests/c/instruction_checks/plain_disassembly.elf
+./build/tests/c/instruction_checks/r5900_trunc_cvt.elf