motorhead/qemu
1
0
Fork 0
mirror of https://github.com/Motorhead1991/qemu.git synced 2025-08-04 08:13:54 -06:00
Code Issues Projects Releases Packages Wiki Activity Actions

target/arm: Implement MVE VQDMULL scalar

Browse source 
Implement the MVE VQDMULL scalar insn. This multiplies the top or
bottom half of each element by the scalar, doubles and saturates
to a double-width result.

Note that this encoding overlaps with VQADD and VQSUB; it uses
what in VQADD and VQSUB would be the 'size=0b11' encoding.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210617121628.20116-30-peter.maydell@linaro.org
This commit is contained in:
Peter Maydell 2021-06-17 13:16:13 +01:00
parent 66c0576754
commit a88903537d
4 changed files with 119 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

65
target/arm/mve_helper.c
View file

@ -610,6 +610,71 @@ DO_2OP_SAT_SCALAR(vqrdmulh_scalarb, 1, int8_t, DO_QRDMULH_B)
DO_2OP_SAT_SCALAR(vqrdmulh_scalarh, 2, int16_t, DO_QRDMULH_H)
DO_2OP_SAT_SCALAR(vqrdmulh_scalarw, 4, int32_t, DO_QRDMULH_W)
/*
* Long saturating scalar ops. As with DO_2OP_L, TYPE and H are for the
* input (smaller) type and LESIZE, LTYPE, LH for the output (long) type.
* SATMASK specifies which bits of the predicate mask matter for determining
* whether to propagate a saturation indication into FPSCR.QC -- for
* the 16x16->32 case we must check only the bit corresponding to the T or B
* half that we used, but for the 32x32->64 case we propagate if the mask
* bit is set for either half.
*/
#define DO_2OP_SAT_SCALAR_L(OP, TOP, ESIZE, TYPE, LESIZE, LTYPE, FN, SATMASK) \
void HELPER(glue(mve_, OP))(CPUARMState *env, void *vd, void *vn, \
uint32_t rm) \
{ \
LTYPE *d = vd; \
TYPE *n = vn; \
TYPE m = rm; \
uint16_t mask = mve_element_mask(env); \
unsigned le; \
bool qc = false; \
for (le = 0; le < 16 / LESIZE; le++, mask >>= LESIZE) { \
bool sat = false; \
LTYPE r = FN((LTYPE)n[H##ESIZE(le * 2 + TOP)], m, &sat); \
mergemask(&d[H##LESIZE(le)], r, mask); \
qc |= sat && (mask & SATMASK); \
} \
if (qc) { \
env->vfp.qc[0] = qc; \
} \
mve_advance_vpt(env); \
}
static inline int32_t do_qdmullh(int16_t n, int16_t m, bool *sat)
{
int64_t r = ((int64_t)n * m) * 2;
return do_sat_bhw(r, INT32_MIN, INT32_MAX, sat);
}
static inline int64_t do_qdmullw(int32_t n, int32_t m, bool *sat)
{
/* The multiply can't overflow, but the doubling might */
int64_t r = (int64_t)n * m;
if (r > INT64_MAX / 2) {
*sat = true;
return INT64_MAX;
} else if (r < INT64_MIN / 2) {
*sat = true;
return INT64_MIN;
} else {
return r * 2;
}
}
#define SATMASK16B 1
#define SATMASK16T (1 << 2)
#define SATMASK32 ((1 << 4) | 1)
DO_2OP_SAT_SCALAR_L(vqdmullb_scalarh, 0, 2, int16_t, 4, int32_t, \
do_qdmullh, SATMASK16B)
DO_2OP_SAT_SCALAR_L(vqdmullb_scalarw, 0, 4, int32_t, 8, int64_t, \
do_qdmullw, SATMASK32)
DO_2OP_SAT_SCALAR_L(vqdmullt_scalarh, 1, 2, int16_t, 4, int32_t, \
do_qdmullh, SATMASK16T)
DO_2OP_SAT_SCALAR_L(vqdmullt_scalarw, 1, 4, int32_t, 8, int64_t, \
do_qdmullw, SATMASK32)
static inline uint32_t do_vbrsrb(uint32_t n, uint32_t m)
{
m &= 0xff;