ISUB_R fix for ARM/RISC-V JIT

.github/workflows/c-cpp.yml

@@ -20,7 +20,7 @@ jobs:
           - {arch: armhf, branch: latest-stable}
           - {arch: armv7, branch: latest-stable}
           - {arch: ppc64le, branch: latest-stable}
-          - {arch: riscv64, branch: edge}
+          - {arch: riscv64, branch: latest-stable}
           - {arch: s390x, branch: latest-stable}
 
     steps:
@@ -45,7 +45,7 @@ jobs:
       run: |
         mkdir build
         cd build
-        cmake ..
+        cmake .. -DCMAKE_BUILD_TYPE=Release -DARCH=native
         make -j$(nproc)
 
     - name: Run tests
@@ -61,8 +61,8 @@ jobs:
     strategy:
       matrix:
         config:
-          - {os: ubuntu-20.04, c: gcc-11, cpp: g++-11}
-          - {os: ubuntu-22.04, c: gcc-12, cpp: g++-12}
+          - {os: ubuntu-24.04, c: gcc-14, cpp: g++-14}
+          - {os: ubuntu-24.04-arm, c: gcc-14, cpp: g++-14}
 
     steps:
     - name: Install dependencies
@@ -133,9 +133,7 @@ jobs:
     strategy:
       matrix:
         config:
-          - {arch: x64, os: 2019, vs: Visual Studio 16 2019, msbuild: "C:\\Program Files (x86)\\Microsoft Visual Studio\\2019\\Enterprise\\MSBuild\\Current\\Bin\\amd64\\"}
           - {arch: x64, os: 2022, vs: Visual Studio 17 2022, msbuild: "C:\\Program Files\\Microsoft Visual Studio\\2022\\Enterprise\\Msbuild\\Current\\Bin\\amd64\\"}
-          - {arch: Win32, os: 2019, vs: Visual Studio 16 2019, msbuild: "C:\\Program Files (x86)\\Microsoft Visual Studio\\2019\\Enterprise\\MSBuild\\Current\\Bin\\"}
           - {arch: Win32, os: 2022, vs: Visual Studio 17 2022, msbuild: "C:\\Program Files\\Microsoft Visual Studio\\2022\\Enterprise\\Msbuild\\Current\\Bin\\"}
 
     steps:
@@ -165,23 +163,20 @@ jobs:
 
     strategy:
       matrix:
-        os: [macos-11, macos-12, macos-13]
+        os: [macos-15, macos-15-intel]
 
     steps:
     - name: Checkout repository
       uses: actions/checkout@v3
       with:
         submodules: recursive
 
-    - name: Install dependencies
-      run: HOMEBREW_NO_AUTO_UPDATE=1 brew install cmake
-
     - name: Build RandomX
       run: |
         mkdir build
         cd build
         cmake ..
-        make -j3
+        make -j4
 
     - name: Run tests
       run: |
@@ -197,13 +192,13 @@ jobs:
         os:
           - name: freebsd
             architecture: x86-64
-            version: '13.2'
-            host: ubuntu-22.04
+            version: '13.3'
+            host: ubuntu-latest
 
           - name: freebsd
             architecture: arm64
-            version: '13.2'
-            host: ubuntu-22.04
+            version: '13.3'
+            host: ubuntu-latest
 
     steps:
     - name: Checkout repository
@@ -212,11 +207,13 @@ jobs:
         submodules: recursive
 
     - name: Build RandomX
-      uses: cross-platform-actions/action@v0.19.0
+      uses: cross-platform-actions/action@v0.24.0
       with:
         operating_system: ${{ matrix.os.name }}
         architecture: ${{ matrix.os.architecture }}
         version: ${{ matrix.os.version }}
+        memory: 12G
+        cpu_count: 4
         shell: bash
         run: |
           sudo pkg install -y cmake

src/jit_compiler_a64.cpp

@@ -561,7 +561,16 @@ void JitCompilerA64::h_ISUB_R(Instruction& instr, uint32_t& codePos)
 	}
 	else
 	{
-		emitAddImmediate(dst, dst, -instr.getImm32(), code, k);
+		const uint32_t imm = instr.getImm32();
+
+		if (imm == 0x80000000ul) {
+			constexpr uint32_t tmp_reg = 20;
+			emit32(ARMV8A::MOVZ | tmp_reg | (1u << 21) | (0x8000u << 5), code, k);
+			emit32(ARMV8A::ADD | dst | (dst << 5) | (tmp_reg << 16), code, k);
+		}
+		else {
+			emitAddImmediate(dst, dst, -instr.getImm32(), code, k);
+		}
 	}
 
 	reg_changed_offset[instr.dst] = k;

src/jit_compiler_rv64.cpp

@@ -714,9 +714,16 @@ namespace randomx {
 			state.emit(rvi(rv64::SUB, regR(isn.dst), regR(isn.dst), regR(isn.src)));
 		}
 		else {
-			int32_t imm = unsigned32ToSigned2sCompl(-isn.getImm32()); //convert to add
-			//x{dst} = x{dst} + {-imm}
-			emitImm32(state, imm, regR(isn.dst), regR(isn.dst), Tmp1Reg);
+			const uint32_t uimm = isn.getImm32();
+			if (uimm == 0x80000000ul) {
+				state.emit(rv64::LUI | (0x80000 << 12) | rvrd(Tmp1Reg));
+				state.emit(rvi(rv64::SUB, regR(isn.dst), regR(isn.dst), Tmp1Reg));
+			}
+			else {
+				int32_t imm = unsigned32ToSigned2sCompl(-uimm); //convert to add
+				//x{dst} = x{dst} + {-imm}
+				emitImm32(state, imm, regR(isn.dst), regR(isn.dst), Tmp1Reg);
+			}
 		}
 	}
 

src/tests/tests.cpp

@@ -59,7 +59,7 @@ void runTest(const char* name, bool condition, FUNC f) {
 	std::cout << "[";
 	std::cout.width(2);
 	std::cout << std::right << ++testNo << "] ";
-	std::cout.width(40);
+	std::cout.width(44);
 	std::cout << std::left << name << " ... ";
 	std::cout.flush();
 	if (condition) {
@@ -1001,6 +1001,20 @@ int main() {
 		assert(equalsHex(hash, "c56414121acda1713c2f2a819d8ae38aed7c80c35c2a769298d34f03833cd5f1"));
 	};
 
+	auto test_f = [&] {
+		uint8_t key_u[RANDOMX_HASH_SIZE] = {
+			0x77, 0x97, 0x37, 0x3e, 0xa4, 0x63, 0x31, 0x94, 0x64, 0x0b, 0xf8, 0xd8, 0xc3, 0xb6, 0x67, 0x24, 0xd6, 0xaa, 0x7b, 0xd2, 0xdc, 0x20, 0xe0, 0x09, 0xdf, 0x2f, 0x8f, 0x17, 0x10, 0xab, 0xe8, 0x24
+		};
+		char key[RANDOMX_HASH_SIZE];
+
+		// workaround for picky compilers
+		memcpy(key, key_u, sizeof(key_u));
+
+		char hash[RANDOMX_HASH_SIZE];
+		calcHexHash(key, "1010e1eaf8cf067b37b5f0ee031ab23ed1755e090a3af4415830145853e2be3e1f6821fed84dae58d00e00da5214d6c1f2d0622e0abd51f9373d04e0b0f8e6d6514d90689721c4aac5a9bb0d", &hash);
+		assert(equalsHex(hash, "78af2a1864c42abce36d2e8983e13df99b2af0ce1362999af09fab004d4435a8"));
+	};
+
 	runTest("Hash test 1a (interpreter)", stringsEqual(RANDOMX_ARGON_SALT, "RandomX\x03"), test_a);
 
 	runTest("Hash test 1b (interpreter)", stringsEqual(RANDOMX_ARGON_SALT, "RandomX\x03"), test_b);
@@ -1011,6 +1025,8 @@ int main() {
 
 	runTest("Hash test 1e (interpreter)", stringsEqual(RANDOMX_ARGON_SALT, "RandomX\x03"), test_e);
 
+	runTest("Hash test 1f (ISUB_R edge case, interpreter)", stringsEqual(RANDOMX_ARGON_SALT, "RandomX\x03"), test_f);
+
 	if (RANDOMX_HAVE_COMPILER) {
 		randomx_release_cache(cache);
 		randomx_destroy_vm(vm);
@@ -1034,6 +1050,8 @@ int main() {
 
 	runTest("Hash test 2e (compiler)", RANDOMX_HAVE_COMPILER && stringsEqual(RANDOMX_ARGON_SALT, "RandomX\x03"), test_e);
 
+	runTest("Hash test 2f (ISUB_R edge case, compiler)", stringsEqual(RANDOMX_ARGON_SALT, "RandomX\x03"), test_f);
+
 	auto flags = randomx_get_flags();
 
 	randomx_release_cache(cache);