The FMAXNMQV and FMINNMQV insns use the default NaN as their identity
value for inactive source vector elements. We open-coded this in
sve_helper.c, hoping to avoid a function call. However, this fails
to account for FPCR.AH=1 changing the default NaN value to set the
sign bit. Use a call to floatN_default_nan() to obtain this value.
Fixes: 1de7ecfc12d05 ("target/arm: Implement FADDQV, F{MIN, MAX}{NM}QV for
SVE2p1")
Signed-off-by: Peter Maydell <peter.mayd...@linaro.org>
---
target/arm/tcg/sve_helper.c | 29 +++++++++++++++++------------
1 file changed, 17 insertions(+), 12 deletions(-)
diff --git a/target/arm/tcg/sve_helper.c b/target/arm/tcg/sve_helper.c
index bf894f0bf13..803f0a094dc 100644
--- a/target/arm/tcg/sve_helper.c
+++ b/target/arm/tcg/sve_helper.c
@@ -4484,33 +4484,35 @@ static TYPE FUNC##_reduce(TYPE *data, float_status
*status, uintptr_t n) \
} \
} \
uint64_t helper_sve_##NAME##v_##SUF(void *vn, void *vg, \
- float_status *s, uint32_t desc) \
+ float_status *status, uint32_t desc) \
{ \
uintptr_t i, oprsz = simd_oprsz(desc), maxsz = simd_data(desc); \
TYPE data[sizeof(ARMVectorReg) / sizeof(TYPE)]; \
+ TYPE ident = IDENT; \
for (i = 0; i < oprsz; ) { \
uint16_t pg = *(uint16_t *)(vg + H1_2(i >> 3)); \
do { \
TYPE nn = *(TYPE *)(vn + H(i)); \
- *(TYPE *)((void *)data + i) = (pg & 1 ? nn : IDENT); \
+ *(TYPE *)((void *)data + i) = (pg & 1 ? nn : ident); \
i += sizeof(TYPE), pg >>= sizeof(TYPE); \
} while (i & 15); \
} \
for (; i < maxsz; i += sizeof(TYPE)) { \
- *(TYPE *)((void *)data + i) = IDENT; \
+ *(TYPE *)((void *)data + i) = ident; \
} \
- return FUNC##_reduce(data, s, maxsz / sizeof(TYPE)); \
+ return FUNC##_reduce(data, status, maxsz / sizeof(TYPE)); \
} \
void helper_sve2p1_##NAME##qv_##SUF(void *vd, void *vn, void *vg, \
float_status *status, uint32_t desc) \
{ \
unsigned oprsz = simd_oprsz(desc), segments = oprsz / 16; \
+ TYPE ident = IDENT; \
for (unsigned e = 0; e < 16; e += sizeof(TYPE)) { \
TYPE data[ARM_MAX_VQ]; \
for (unsigned s = 0; s < segments; s++) { \
uint16_t pg = *(uint16_t *)(vg + H1_2(s * 2)); \
TYPE nn = *(TYPE *)(vn + (s * 16 + H(e))); \
- data[s] = (pg >> e) & 1 ? nn : IDENT; \
+ data[s] = (pg >> e) & 1 ? nn : ident; \
} \
*(TYPE *)(vd + H(e)) = FUNC##_reduce(data, status, segments); \
} \
@@ -4521,14 +4523,17 @@ DO_REDUCE(fadd,h, float16, H1_2, float16_add,
float16_zero)
DO_REDUCE(fadd,s, float32, H1_4, float32_add, float32_zero)
DO_REDUCE(fadd,d, float64, H1_8, float64_add, float64_zero)
-/* Identity is floatN_default_nan, without the function call. */
-DO_REDUCE(fminnm,h, float16, H1_2, float16_minnum, 0x7E00)
-DO_REDUCE(fminnm,s, float32, H1_4, float32_minnum, 0x7FC00000)
-DO_REDUCE(fminnm,d, float64, H1_8, float64_minnum, 0x7FF8000000000000ULL)
+/*
+ * We can't avoid the function call for the default NaN value, because
+ * it changes when FPCR.AH is set.
+ */
+DO_REDUCE(fminnm,h, float16, H1_2, float16_minnum, float16_default_nan(status))
+DO_REDUCE(fminnm,s, float32, H1_4, float32_minnum, float32_default_nan(status))
+DO_REDUCE(fminnm,d, float64, H1_8, float64_minnum, float64_default_nan(status))
-DO_REDUCE(fmaxnm,h, float16, H1_2, float16_maxnum, 0x7E00)
-DO_REDUCE(fmaxnm,s, float32, H1_4, float32_maxnum, 0x7FC00000)
-DO_REDUCE(fmaxnm,d, float64, H1_8, float64_maxnum, 0x7FF8000000000000ULL)
+DO_REDUCE(fmaxnm,h, float16, H1_2, float16_maxnum, float16_default_nan(status))
+DO_REDUCE(fmaxnm,s, float32, H1_4, float32_maxnum, float32_default_nan(status))
+DO_REDUCE(fmaxnm,d, float64, H1_8, float64_maxnum, float64_default_nan(status))
DO_REDUCE(fmin,h, float16, H1_2, float16_min, float16_infinity)
DO_REDUCE(fmin,s, float32, H1_4, float32_min, float32_infinity)
--
2.43.0
