fpu: Process float_muladd_negate_result after rounding

Changing the sign before rounding affects the correctness of
the asymmetric rouding modes: float_round_up and float_round_down.

Reported-by: WANG Rui <wangrui@loongson.cn>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

2 files changed, 38 insertions, 0 deletions

diff --git a/tests/tcg/multiarch/Makefile.target b/tests/tcg/multiarch/Makefile.target
index 45c9cfe18c..bfdf7197a7 100644
--- a/tests/tcg/multiarch/Makefile.target
+++ b/tests/tcg/multiarch/Makefile.target
@@ -29,6 +29,7 @@ run-float_%: float_%
 	$(call run-test,$<, $(QEMU) $(QEMU_OPTS) $<)
 	$(call conditional-diff-out,$<,$(SRC_PATH)/tests/tcg/$(TARGET_NAME)/$<.ref)
 
+fnmsub: LDFLAGS+=-lm
 
 testthread: LDFLAGS+=-lpthread
 
diff --git a/tests/tcg/multiarch/fnmsub.c b/tests/tcg/multiarch/fnmsub.c
new file mode 100644
index 0000000000..15dd41d3bd
--- /dev/null
+++ b/tests/tcg/multiarch/fnmsub.c
@@ -0,0 +1,37 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+
+#include <stdio.h>
+#include <math.h>
+#include <fenv.h>
+
+union U {
+  double d;
+  unsigned long long l;
+};
+
+union U x = { .l = 0x4ff0000000000000ULL };
+union U y = { .l = 0x2ff0000000000000ULL };
+union U r;
+
+int main()
+{
+#ifdef FE_DOWNWARD
+    fesetround(FE_DOWNWARD);
+
+#if defined(__loongarch__)
+    asm("fnmsub.d %0, %1, %1, %2" : "=f"(r.d) : "f"(x.d), "f"(y.d));
+#elif defined(__powerpc64__)
+    asm("fnmsub %0,%1,%1,%2" : "=f"(r.d) : "f"(x.d), "f"(y.d));
+#elif defined(__s390x__) && 0 /* need -march=z14 */
+    asm("vfnms %0,%1,%1,%2,0,3" : "=f"(r.d) : "f"(x.d), "f"(y.d));
+#else
+    r.d = -fma(x.d, x.d, -y.d);
+#endif
+
+    if (r.l != 0xdfefffffffffffffULL) {
+        printf("r = %.18a (%016llx)\n", r.d, r.l);
+        return 1;
+    }
+#endif
+    return 0;
+}