Fix FPU_CVT_W_S: truncate toward zero and saturate like R5900 CVT.W.S

ps2xRuntime/include/ps2_runtime_macros.h

@@ -623,7 +623,17 @@ inline __m128i ps2_u64_to_epi64_pair(uint64_t value)
 #define FPU_FLOOR_W_S(a) ((int32_t)floorf((float)(a)))
 #define FPU_CVT_S_W(a) ((float)(int32_t)(a))
 #define FPU_CVT_S_L(a) ((float)(int64_t)(a))
-#define FPU_CVT_W_S(a) ((int32_t)nearbyintf((float)(a)))
+// CVT.W.S: truncate toward zero regardless of the rounding mode.
+// Clamp to either INT32_MIN/INT32_MAX on over/underflow.
+static inline int32_t ps2_fpu_cvt_w_s(float a)
+{
+    if (a >= 2147483648.0f)
+        return INT32_MAX;
+    if (a <= -2147483648.0f)
+        return INT32_MIN;
+    return (int32_t)a;
+}
+#define FPU_CVT_W_S(a) (ps2_fpu_cvt_w_s((float)(a)))
 #define FPU_CVT_L_S(a) ((int64_t)(float)(a))
 #define FPU_C_F_S(a, b) (0)
 #define FPU_C_UN_S(a, b) (isnan((float)(a)) || isnan((float)(b)))

ps2xTest/CMakeLists.txt

@@ -45,6 +45,7 @@ target_include_directories(ps2_test_function_table PRIVATE
 add_library(ps2_test_lib STATIC
     src/code_generator_tests.cpp
     src/r5900_decoder_tests.cpp
+    src/ps2_fpu_macros_tests.cpp
     src/elf_analyzer_tests.cpp
     src/pad_input_tests.cpp
     src/ps2_runtime_io_tests.cpp

ps2xTest/src/main.cpp

@@ -18,6 +18,7 @@ void register_ps2_sif_rpc_tests();
 void register_ps2_sif_dma_tests();
 void register_ps2_recompiler_tests();
 void register_ps2_runtime_expansion_tests();
+void register_ps2_fpu_macros_tests();
 void reset_ps2_test_function_table();
 
 int main()
@@ -40,6 +41,7 @@ int main()
     register_ps2_sif_dma_tests();
     register_ps2_recompiler_tests();
     register_ps2_runtime_expansion_tests();
+    register_ps2_fpu_macros_tests();
     int res = MiniTest::Run();
     std::cout.flush();
     std::cerr.flush();

ps2xTest/src/ps2_fpu_macros_tests.cpp

@@ -0,0 +1,41 @@
+#include "MiniTest.h"
+
+#include <cfenv>
+#include <cstdint>
+#include <cstring>
+
+#include "ps2_runtime_macros.h"
+
+// R5900 CVT.W.S reference behavior (matches the PCSX2 interpreter's CVT_W):
+// truncate toward zero, saturate to 0x7FFFFFFF / 0x80000000 on overflow.
+// The EE FPU ignores the host/guest rounding mode for this conversion.
+
+void register_ps2_fpu_macros_tests()
+{
+    MiniTest::Case("FpuMacros", [](TestCase &tc)
+                   {
+    tc.Run("CVT_W_S truncates toward zero", [](TestCase &t) {
+        t.Equals((uint32_t)FPU_CVT_W_S(2.7f), 2u, "2.7 must truncate to 2");
+        t.Equals((uint32_t)FPU_CVT_W_S(3.5f), 3u, "3.5 must truncate to 3 (not round to 4)");
+        t.Equals((uint32_t)FPU_CVT_W_S(0.99f), 0u, "0.99 must truncate to 0");
+        t.Equals((uint32_t)FPU_CVT_W_S(-2.7f), 0xFFFFFFFEu, "-2.7 must truncate to -2");
+    });
+
+    tc.Run("CVT_W_S ignores host rounding mode", [](TestCase &t) {
+        const int previous = std::fegetround();
+        std::fesetround(FE_TONEAREST);
+        t.Equals((uint32_t)FPU_CVT_W_S(2.7f), 2u, "must truncate even under FE_TONEAREST");
+        std::fesetround(FE_UPWARD);
+        t.Equals((uint32_t)FPU_CVT_W_S(2.1f), 2u, "must truncate even under FE_UPWARD");
+        std::fesetround(previous);
+    });
+
+    tc.Run("CVT_W_S saturates on overflow", [](TestCase &t) {
+        t.Equals((uint32_t)FPU_CVT_W_S(3e9f), 0x7FFFFFFFu, "positive overflow -> 0x7FFFFFFF");
+        t.Equals((uint32_t)FPU_CVT_W_S(-3e9f), 0x80000000u, "negative overflow -> 0x80000000");
+        t.Equals((uint32_t)FPU_CVT_W_S(2147483648.0f), 0x7FFFFFFFu, "2^31 -> 0x7FFFFFFF");
+        t.Equals((uint32_t)FPU_CVT_W_S(-2147483648.0f), 0x80000000u, "-2^31 -> 0x80000000");
+        t.Equals((uint32_t)FPU_CVT_W_S(2147483520.0f), 0x7FFFFF80u,
+                 "largest float below 2^31 converts exactly");
+    }); });
+}