From: Paul Brook <paul@nowt.org>
These helpers need to take special care to avoid overwriting source values
before the wole result has been calculated. Currently they use a dummy
Reg typed variable to store the result then assign the whole register.
This will cause 128 bit operations to corrupt the upper half of the register,
so replace it with explicit temporaries and element assignments.
Signed-off-by: Paul Brook <paul@nowt.org>
Message-Id: <20220424220204.2493824-14-paul@nowt.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
target/i386/ops_sse.h | 565 +++++++++++++++++++++---------------------
1 file changed, 284 insertions(+), 281 deletions(-)
diff --git a/target/i386/ops_sse.h b/target/i386/ops_sse.h
index 4d1fcbd3ae..5265005f1e 100644
--- a/target/i386/ops_sse.h
+++ b/target/i386/ops_sse.h
@@ -475,71 +475,68 @@ void glue(helper_movq_mm_T0, SUFFIX)(Reg *d, uint64_t val)
}
#endif
+#define SHUFFLE4(F, a, b, offset) do { \
+ r0 = a->F((order & 3) + offset); \
+ r1 = a->F(((order >> 2) & 3) + offset); \
+ r2 = b->F(((order >> 4) & 3) + offset); \
+ r3 = b->F(((order >> 6) & 3) + offset); \
+ d->F(offset) = r0; \
+ d->F(offset + 1) = r1; \
+ d->F(offset + 2) = r2; \
+ d->F(offset + 3) = r3; \
+ } while (0)
+
#if SHIFT == 0
void glue(helper_pshufw, SUFFIX)(Reg *d, Reg *s, int order)
{
- Reg r;
+ uint16_t r0, r1, r2, r3;
- r.W(0) = s->W(order & 3);
- r.W(1) = s->W((order >> 2) & 3);
- r.W(2) = s->W((order >> 4) & 3);
- r.W(3) = s->W((order >> 6) & 3);
- MOVE(*d, r);
+ SHUFFLE4(W, s, s, 0);
}
#else
void helper_shufps(Reg *d, Reg *s, int order)
{
- Reg r;
+ Reg *v = d;
+ uint32_t r0, r1, r2, r3;
- r.L(0) = d->L(order & 3);
- r.L(1) = d->L((order >> 2) & 3);
- r.L(2) = s->L((order >> 4) & 3);
- r.L(3) = s->L((order >> 6) & 3);
- MOVE(*d, r);
+ SHUFFLE4(L, v, s, 0);
}
void helper_shufpd(Reg *d, Reg *s, int order)
{
- Reg r;
+ Reg *v = d;
+ uint64_t r0, r1;
- r.Q(0) = d->Q(order & 1);
- r.Q(1) = s->Q((order >> 1) & 1);
- MOVE(*d, r);
+ r0 = v->Q(order & 1);
+ r1 = s->Q((order >> 1) & 1);
+ d->Q(0) = r0;
+ d->Q(1) = r1;
}
void glue(helper_pshufd, SUFFIX)(Reg *d, Reg *s, int order)
{
- Reg r;
+ uint32_t r0, r1, r2, r3;
- r.L(0) = s->L(order & 3);
- r.L(1) = s->L((order >> 2) & 3);
- r.L(2) = s->L((order >> 4) & 3);
- r.L(3) = s->L((order >> 6) & 3);
- MOVE(*d, r);
+ SHUFFLE4(L, s, s, 0);
+#if SHIFT == 2
+ SHUFFLE4(L, s, s, 4);
+#endif
}
void glue(helper_pshuflw, SUFFIX)(Reg *d, Reg *s, int order)
{
- Reg r;
+ uint16_t r0, r1, r2, r3;
- r.W(0) = s->W(order & 3);
- r.W(1) = s->W((order >> 2) & 3);
- r.W(2) = s->W((order >> 4) & 3);
- r.W(3) = s->W((order >> 6) & 3);
- r.Q(1) = s->Q(1);
- MOVE(*d, r);
+ SHUFFLE4(W, s, s, 0);
+ d->Q(1) = s->Q(1);
}
void glue(helper_pshufhw, SUFFIX)(Reg *d, Reg *s, int order)
{
- Reg r;
+ uint16_t r0, r1, r2, r3;
- r.Q(0) = s->Q(0);
- r.W(4) = s->W(4 + (order & 3));
- r.W(5) = s->W(4 + ((order >> 2) & 3));
- r.W(6) = s->W(4 + ((order >> 4) & 3));
- r.W(7) = s->W(4 + ((order >> 6) & 3));
- MOVE(*d, r);
+ d->Q(0) = s->Q(0);
+ SHUFFLE4(W, s, s, 4);
}
#endif
@@ -1092,156 +1089,157 @@ uint32_t glue(helper_pmovmskb, SUFFIX)(CPUX86State *env, Reg *s)
return val;
}
-void glue(helper_packsswb, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
-{
- Reg r;
+#if SHIFT == 0
+#define PACK_WIDTH 4
+#else
+#define PACK_WIDTH 8
+#endif
- r.B(0) = satsb((int16_t)d->W(0));
- r.B(1) = satsb((int16_t)d->W(1));
- r.B(2) = satsb((int16_t)d->W(2));
- r.B(3) = satsb((int16_t)d->W(3));
-#if SHIFT == 1
- r.B(4) = satsb((int16_t)d->W(4));
- r.B(5) = satsb((int16_t)d->W(5));
- r.B(6) = satsb((int16_t)d->W(6));
- r.B(7) = satsb((int16_t)d->W(7));
-#endif
- r.B((4 << SHIFT) + 0) = satsb((int16_t)s->W(0));
- r.B((4 << SHIFT) + 1) = satsb((int16_t)s->W(1));
- r.B((4 << SHIFT) + 2) = satsb((int16_t)s->W(2));
- r.B((4 << SHIFT) + 3) = satsb((int16_t)s->W(3));
-#if SHIFT == 1
- r.B(12) = satsb((int16_t)s->W(4));
- r.B(13) = satsb((int16_t)s->W(5));
- r.B(14) = satsb((int16_t)s->W(6));
- r.B(15) = satsb((int16_t)s->W(7));
-#endif
- MOVE(*d, r);
+#define PACK4(F, to, reg, from) do { \
+ r[to + 0] = F((int16_t)reg->W(from + 0)); \
+ r[to + 1] = F((int16_t)reg->W(from + 1)); \
+ r[to + 2] = F((int16_t)reg->W(from + 2)); \
+ r[to + 3] = F((int16_t)reg->W(from + 3)); \
+ } while (0)
+
+#define PACK_HELPER_B(name, F) \
+void glue(helper_pack ## name, SUFFIX)(CPUX86State *env, \
+ Reg *d, Reg *s) \
+{ \
+ Reg *v = d; \
+ uint8_t r[PACK_WIDTH * 2]; \
+ int i; \
+ PACK4(F, 0, v, 0); \
+ PACK4(F, PACK_WIDTH, s, 0); \
+ XMM_ONLY( \
+ PACK4(F, 4, v, 4); \
+ PACK4(F, 12, s, 4); \
+ ) \
+ for (i = 0; i < PACK_WIDTH * 2; i++) { \
+ d->B(i) = r[i]; \
+ } \
}
-void glue(helper_packuswb, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
-{
- Reg r;
-
- r.B(0) = satub((int16_t)d->W(0));
- r.B(1) = satub((int16_t)d->W(1));
- r.B(2) = satub((int16_t)d->W(2));
- r.B(3) = satub((int16_t)d->W(3));
-#if SHIFT == 1
- r.B(4) = satub((int16_t)d->W(4));
- r.B(5) = satub((int16_t)d->W(5));
- r.B(6) = satub((int16_t)d->W(6));
- r.B(7) = satub((int16_t)d->W(7));
-#endif
- r.B((4 << SHIFT) + 0) = satub((int16_t)s->W(0));
- r.B((4 << SHIFT) + 1) = satub((int16_t)s->W(1));
- r.B((4 << SHIFT) + 2) = satub((int16_t)s->W(2));
- r.B((4 << SHIFT) + 3) = satub((int16_t)s->W(3));
-#if SHIFT == 1
- r.B(12) = satub((int16_t)s->W(4));
- r.B(13) = satub((int16_t)s->W(5));
- r.B(14) = satub((int16_t)s->W(6));
- r.B(15) = satub((int16_t)s->W(7));
-#endif
- MOVE(*d, r);
-}
+PACK_HELPER_B(sswb, satsb)
+PACK_HELPER_B(uswb, satub)
void glue(helper_packssdw, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
{
- Reg r;
+ Reg *v = d;
+ uint16_t r[PACK_WIDTH];
+ int i, j, k;
- r.W(0) = satsw(d->L(0));
- r.W(1) = satsw(d->L(1));
-#if SHIFT == 1
- r.W(2) = satsw(d->L(2));
- r.W(3) = satsw(d->L(3));
+ for (i = 0, j = 0; i <= 2 << SHIFT; i += 8, j += 4) {
+ r[0] = satsw(v->L(j));
+ r[1] = satsw(v->L(j + 1));
+ r[PACK_WIDTH / 2 + 0] = satsw(s->L(j));
+ r[PACK_WIDTH / 2 + 1] = satsw(s->L(j + 1));
+#if SHIFT >= 1
+ r[2] = satsw(v->L(j + 2));
+ r[3] = satsw(v->L(j + 3));
+ r[6] = satsw(s->L(j + 2));
+ r[7] = satsw(s->L(j + 3));
#endif
- r.W((2 << SHIFT) + 0) = satsw(s->L(0));
- r.W((2 << SHIFT) + 1) = satsw(s->L(1));
-#if SHIFT == 1
- r.W(6) = satsw(s->L(2));
- r.W(7) = satsw(s->L(3));
-#endif
- MOVE(*d, r);
+ for (k = 0; k < PACK_WIDTH; k++) {
+ d->W(i + k) = r[k];
+ }
+ }
}
#define UNPCK_OP(base_name, base) \
\
void glue(helper_punpck ## base_name ## bw, SUFFIX)(CPUX86State *env,\
- Reg *d, Reg *s) \
+ Reg *d, Reg *s) \
{ \
- Reg r; \
+ Reg *v = d; \
+ uint8_t r[PACK_WIDTH * 2]; \
+ int i; \
\
- r.B(0) = d->B((base << (SHIFT + 2)) + 0); \
- r.B(1) = s->B((base << (SHIFT + 2)) + 0); \
- r.B(2) = d->B((base << (SHIFT + 2)) + 1); \
- r.B(3) = s->B((base << (SHIFT + 2)) + 1); \
- r.B(4) = d->B((base << (SHIFT + 2)) + 2); \
- r.B(5) = s->B((base << (SHIFT + 2)) + 2); \
- r.B(6) = d->B((base << (SHIFT + 2)) + 3); \
- r.B(7) = s->B((base << (SHIFT + 2)) + 3); \
+ r[0] = v->B((base * PACK_WIDTH) + 0); \
+ r[1] = s->B((base * PACK_WIDTH) + 0); \
+ r[2] = v->B((base * PACK_WIDTH) + 1); \
+ r[3] = s->B((base * PACK_WIDTH) + 1); \
+ r[4] = v->B((base * PACK_WIDTH) + 2); \
+ r[5] = s->B((base * PACK_WIDTH) + 2); \
+ r[6] = v->B((base * PACK_WIDTH) + 3); \
+ r[7] = s->B((base * PACK_WIDTH) + 3); \
XMM_ONLY( \
- r.B(8) = d->B((base << (SHIFT + 2)) + 4); \
- r.B(9) = s->B((base << (SHIFT + 2)) + 4); \
- r.B(10) = d->B((base << (SHIFT + 2)) + 5); \
- r.B(11) = s->B((base << (SHIFT + 2)) + 5); \
- r.B(12) = d->B((base << (SHIFT + 2)) + 6); \
- r.B(13) = s->B((base << (SHIFT + 2)) + 6); \
- r.B(14) = d->B((base << (SHIFT + 2)) + 7); \
- r.B(15) = s->B((base << (SHIFT + 2)) + 7); \
+ r[8] = v->B((base * PACK_WIDTH) + 4); \
+ r[9] = s->B((base * PACK_WIDTH) + 4); \
+ r[10] = v->B((base * PACK_WIDTH) + 5); \
+ r[11] = s->B((base * PACK_WIDTH) + 5); \
+ r[12] = v->B((base * PACK_WIDTH) + 6); \
+ r[13] = s->B((base * PACK_WIDTH) + 6); \
+ r[14] = v->B((base * PACK_WIDTH) + 7); \
+ r[15] = s->B((base * PACK_WIDTH) + 7); \
) \
- MOVE(*d, r); \
+ for (i = 0; i < PACK_WIDTH * 2; i++) { \
+ d->B(i) = r[i]; \
+ } \
} \
\
void glue(helper_punpck ## base_name ## wd, SUFFIX)(CPUX86State *env,\
- Reg *d, Reg *s) \
+ Reg *d, Reg *s) \
{ \
- Reg r; \
+ Reg *v = d; \
+ uint16_t r[PACK_WIDTH]; \
+ int i; \
\
- r.W(0) = d->W((base << (SHIFT + 1)) + 0); \
- r.W(1) = s->W((base << (SHIFT + 1)) + 0); \
- r.W(2) = d->W((base << (SHIFT + 1)) + 1); \
- r.W(3) = s->W((base << (SHIFT + 1)) + 1); \
+ r[0] = v->W((base * (PACK_WIDTH / 2)) + 0); \
+ r[1] = s->W((base * (PACK_WIDTH / 2)) + 0); \
+ r[2] = v->W((base * (PACK_WIDTH / 2)) + 1); \
+ r[3] = s->W((base * (PACK_WIDTH / 2)) + 1); \
XMM_ONLY( \
- r.W(4) = d->W((base << (SHIFT + 1)) + 2); \
- r.W(5) = s->W((base << (SHIFT + 1)) + 2); \
- r.W(6) = d->W((base << (SHIFT + 1)) + 3); \
- r.W(7) = s->W((base << (SHIFT + 1)) + 3); \
+ r[4] = v->W((base * 4) + 2); \
+ r[5] = s->W((base * 4) + 2); \
+ r[6] = v->W((base * 4) + 3); \
+ r[7] = s->W((base * 4) + 3); \
) \
- MOVE(*d, r); \
+ for (i = 0; i < PACK_WIDTH; i++) { \
+ d->W(i) = r[i]; \
+ } \
} \
\
void glue(helper_punpck ## base_name ## dq, SUFFIX)(CPUX86State *env,\
- Reg *d, Reg *s) \
+ Reg *d, Reg *s) \
{ \
- Reg r; \
+ Reg *v = d; \
+ uint32_t r[4]; \
\
- r.L(0) = d->L((base << SHIFT) + 0); \
- r.L(1) = s->L((base << SHIFT) + 0); \
+ r[0] = v->L((base * (PACK_WIDTH / 4)) + 0); \
+ r[1] = s->L((base * (PACK_WIDTH / 4)) + 0); \
XMM_ONLY( \
- r.L(2) = d->L((base << SHIFT) + 1); \
- r.L(3) = s->L((base << SHIFT) + 1); \
+ r[2] = v->L((base * 2) + 1); \
+ r[3] = s->L((base * 2) + 1); \
+ d->L(2) = r[2]; \
+ d->L(3) = r[3]; \
) \
- MOVE(*d, r); \
+ d->L(0) = r[0]; \
+ d->L(1) = r[1]; \
} \
Reg *s) \
+ void glue(helper_punpck ## base_name ## qdq, SUFFIX)( \
+ CPUX86State *env, Reg *d, Reg *s) \
{ \
- Reg r; \
+ Reg *v = d; \
+ uint64_t r[2]; \
\
- r.Q(0) = d->Q(base); \
- r.Q(1) = s->Q(base); \
- MOVE(*d, r); \
+ r[0] = v->Q(base); \
+ r[1] = s->Q(base); \
+ d->Q(0) = r[0]; \
+ d->Q(1) = r[1]; \
} \
)
UNPCK_OP(l, 0)
UNPCK_OP(h, 1)
+#undef PACK_WIDTH
+#undef PACK_HELPER_B
+#undef PACK4
+
+
/* 3DNow! float ops */
#if SHIFT == 0
void helper_pi2fd(CPUX86State *env, MMXReg *d, MMXReg *s)
@@ -1394,122 +1392,113 @@ void helper_pswapd(CPUX86State *env, MMXReg *d, MMXReg *s)
/* SSSE3 op helpers */
void glue(helper_pshufb, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
{
+ Reg *v = d;
int i;
- Reg r;
+#if SHIFT == 0
+ uint8_t r[8];
- for (i = 0; i < (8 << SHIFT); i++) {
- r.B(i) = (s->B(i) & 0x80) ? 0 : (d->B(s->B(i) & ((8 << SHIFT) - 1)));
+ for (i = 0; i < 8; i++) {
+ r[i] = (s->B(i) & 0x80) ? 0 : (v->B(s->B(i) & 7));
}
+ for (i = 0; i < 8; i++) {
+ d->B(i) = r[i];
+ }
+#else
+ uint8_t r[8 << SHIFT];
- MOVE(*d, r);
-}
-
-void glue(helper_phaddw, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
-{
-
- Reg r;
-
- r.W(0) = (int16_t)d->W(0) + (int16_t)d->W(1);
- r.W(1) = (int16_t)d->W(2) + (int16_t)d->W(3);
- XMM_ONLY(r.W(2) = (int16_t)d->W(4) + (int16_t)d->W(5));
- XMM_ONLY(r.W(3) = (int16_t)d->W(6) + (int16_t)d->W(7));
- r.W((2 << SHIFT) + 0) = (int16_t)s->W(0) + (int16_t)s->W(1);
- r.W((2 << SHIFT) + 1) = (int16_t)s->W(2) + (int16_t)s->W(3);
- XMM_ONLY(r.W(6) = (int16_t)s->W(4) + (int16_t)s->W(5));
- XMM_ONLY(r.W(7) = (int16_t)s->W(6) + (int16_t)s->W(7));
-
- MOVE(*d, r);
-}
-
-void glue(helper_phaddd, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
-{
- Reg r;
-
- r.L(0) = (int32_t)d->L(0) + (int32_t)d->L(1);
- XMM_ONLY(r.L(1) = (int32_t)d->L(2) + (int32_t)d->L(3));
- r.L((1 << SHIFT) + 0) = (int32_t)s->L(0) + (int32_t)s->L(1);
- XMM_ONLY(r.L(3) = (int32_t)s->L(2) + (int32_t)s->L(3));
-
- MOVE(*d, r);
-}
-
-void glue(helper_phaddsw, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
-{
- Reg r;
-
- r.W(0) = satsw((int16_t)d->W(0) + (int16_t)d->W(1));
- r.W(1) = satsw((int16_t)d->W(2) + (int16_t)d->W(3));
- XMM_ONLY(r.W(2) = satsw((int16_t)d->W(4) + (int16_t)d->W(5)));
- XMM_ONLY(r.W(3) = satsw((int16_t)d->W(6) + (int16_t)d->W(7)));
- r.W((2 << SHIFT) + 0) = satsw((int16_t)s->W(0) + (int16_t)s->W(1));
- r.W((2 << SHIFT) + 1) = satsw((int16_t)s->W(2) + (int16_t)s->W(3));
- XMM_ONLY(r.W(6) = satsw((int16_t)s->W(4) + (int16_t)s->W(5)));
- XMM_ONLY(r.W(7) = satsw((int16_t)s->W(6) + (int16_t)s->W(7)));
-
- MOVE(*d, r);
-}
-
-void glue(helper_pmaddubsw, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
-{
- d->W(0) = satsw((int8_t)s->B(0) * (uint8_t)d->B(0) +
- (int8_t)s->B(1) * (uint8_t)d->B(1));
- d->W(1) = satsw((int8_t)s->B(2) * (uint8_t)d->B(2) +
- (int8_t)s->B(3) * (uint8_t)d->B(3));
- d->W(2) = satsw((int8_t)s->B(4) * (uint8_t)d->B(4) +
- (int8_t)s->B(5) * (uint8_t)d->B(5));
- d->W(3) = satsw((int8_t)s->B(6) * (uint8_t)d->B(6) +
- (int8_t)s->B(7) * (uint8_t)d->B(7));
-#if SHIFT == 1
- d->W(4) = satsw((int8_t)s->B(8) * (uint8_t)d->B(8) +
- (int8_t)s->B(9) * (uint8_t)d->B(9));
- d->W(5) = satsw((int8_t)s->B(10) * (uint8_t)d->B(10) +
- (int8_t)s->B(11) * (uint8_t)d->B(11));
- d->W(6) = satsw((int8_t)s->B(12) * (uint8_t)d->B(12) +
- (int8_t)s->B(13) * (uint8_t)d->B(13));
- d->W(7) = satsw((int8_t)s->B(14) * (uint8_t)d->B(14) +
- (int8_t)s->B(15) * (uint8_t)d->B(15));
+ for (i = 0; i < 8 << SHIFT; i++) {
+ int j = i & ~0xf;
+ r[i] = (s->B(i) & 0x80) ? 0 : v->B(j | (s->B(i) & 0xf));
+ }
+ for (i = 0; i < 8 << SHIFT; i++) {
+ d->B(i) = r[i];
+ }
#endif
}
-void glue(helper_phsubw, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
-{
- Reg r;
+#if SHIFT == 0
- r.W(0) = (int16_t)d->W(0) - (int16_t)d->W(1);
- r.W(1) = (int16_t)d->W(2) - (int16_t)d->W(3);
- XMM_ONLY(r.W(2) = (int16_t)d->W(4) - (int16_t)d->W(5));
- XMM_ONLY(r.W(3) = (int16_t)d->W(6) - (int16_t)d->W(7));
- r.W((2 << SHIFT) + 0) = (int16_t)s->W(0) - (int16_t)s->W(1);
- r.W((2 << SHIFT) + 1) = (int16_t)s->W(2) - (int16_t)s->W(3);
- XMM_ONLY(r.W(6) = (int16_t)s->W(4) - (int16_t)s->W(5));
- XMM_ONLY(r.W(7) = (int16_t)s->W(6) - (int16_t)s->W(7));
- MOVE(*d, r);
+#define SSE_HELPER_HW(name, F) \
+void glue(helper_ ## name, SUFFIX)(CPUX86State *env, Reg *d, Reg *s) \
+{ \
+ Reg *v = d; \
+ uint16_t r[4]; \
+ r[0] = F(v->W(0), v->W(1)); \
+ r[1] = F(v->W(2), v->W(3)); \
+ r[2] = F(s->W(0), s->W(1)); \
+ r[3] = F(s->W(3), s->W(3)); \
+ d->W(0) = r[0]; \
+ d->W(1) = r[1]; \
+ d->W(2) = r[2]; \
+ d->W(3) = r[3]; \
}
-void glue(helper_phsubd, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
-{
- Reg r;
-
- r.L(0) = (int32_t)d->L(0) - (int32_t)d->L(1);
- XMM_ONLY(r.L(1) = (int32_t)d->L(2) - (int32_t)d->L(3));
- r.L((1 << SHIFT) + 0) = (int32_t)s->L(0) - (int32_t)s->L(1);
- XMM_ONLY(r.L(3) = (int32_t)s->L(2) - (int32_t)s->L(3));
- MOVE(*d, r);
+#define SSE_HELPER_HL(name, F) \
+void glue(helper_ ## name, SUFFIX)(CPUX86State *env, Reg *d, Reg *s) \
+{ \
+ Reg *v = d; \
+ uint32_t r0, r1; \
+ r0 = F(v->L(0), v->L(1)); \
+ r1 = F(s->L(0), s->L(1)); \
+ d->W(0) = r0; \
+ d->W(1) = r1; \
}
-void glue(helper_phsubsw, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
-{
- Reg r;
+#else
- r.W(0) = satsw((int16_t)d->W(0) - (int16_t)d->W(1));
- r.W(1) = satsw((int16_t)d->W(2) - (int16_t)d->W(3));
- XMM_ONLY(r.W(2) = satsw((int16_t)d->W(4) - (int16_t)d->W(5)));
- XMM_ONLY(r.W(3) = satsw((int16_t)d->W(6) - (int16_t)d->W(7)));
- r.W((2 << SHIFT) + 0) = satsw((int16_t)s->W(0) - (int16_t)s->W(1));
- r.W((2 << SHIFT) + 1) = satsw((int16_t)s->W(2) - (int16_t)s->W(3));
- XMM_ONLY(r.W(6) = satsw((int16_t)s->W(4) - (int16_t)s->W(5)));
- XMM_ONLY(r.W(7) = satsw((int16_t)s->W(6) - (int16_t)s->W(7)));
- MOVE(*d, r);
+#define SSE_HELPER_HW(name, F) \
+void glue(helper_ ## name, SUFFIX)(CPUX86State *env, Reg *d, Reg *s) \
+{ \
+ Reg *v = d; \
+ int16_t r[4 << SHIFT]; \
+ int i, j; \
+ for (i = j = 0; j < 8; i++, j += 2) { \
+ r[i] = F(v->W(j), v->W(j + 1)); \
+ } \
+ for (j = 0; j < 8; i++, j += 2) { \
+ r[i] = F(s->W(j), s->W(j + 1)); \
+ } \
+ for (i = 0; i < 4 << SHIFT; i++) { \
+ d->W(i) = r[i]; \
+ } \
+}
+
+#define SSE_HELPER_HL(name, F) \
+void glue(helper_ ## name, SUFFIX)(CPUX86State *env, Reg *d, Reg *s) \
+{ \
+ Reg *v = d; \
+ int32_t r[2 << SHIFT]; \
+ int i, j; \
+ for (i = j = 0; j < 4; i++, j += 2) { \
+ r[i] = F(v->L(j), v->L(j + 1)); \
+ } \
+ for (j = 0; j < 4; i++, j += 2) { \
+ r[i] = F(s->L(j), s->L(j + 1)); \
+ } \
+ for (i = 0; i < 2 << SHIFT; i++) { \
+ d->L(i) = r[i]; \
+ } \
+}
+#endif
+
+SSE_HELPER_HW(phaddw, FADD)
+SSE_HELPER_HW(phsubw, FSUB)
+SSE_HELPER_HW(phaddsw, FADDSW)
+SSE_HELPER_HW(phsubsw, FSUBSW)
+SSE_HELPER_HL(phaddd, FADD)
+SSE_HELPER_HL(phsubd, FSUB)
+
+#undef SSE_HELPER_HW
+#undef SSE_HELPER_HL
+
+void glue(helper_pmaddubsw, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
+{
+ Reg *v = d;
+ int i;
+ for (i = 0; i < 4 << SHIFT; i++) {
+ d->W(i) = satsw((int8_t)s->B(i * 2) * (uint8_t)v->B(i * 2) +
+ (int8_t)s->B(i * 2 + 1) * (uint8_t)v->B(i * 2 + 1));
+ }
}
#define FABSB(x) (x > INT8_MAX ? -(int8_t)x : x)
@@ -1532,32 +1521,38 @@ SSE_HELPER_L(helper_psignd, FSIGNL)
void glue(helper_palignr, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
int32_t shift)
{
- Reg r;
+ Reg *v = d;
+ int i;
/* XXX could be checked during translation */
- if (shift >= (16 << SHIFT)) {
- r.Q(0) = 0;
- XMM_ONLY(r.Q(1) = 0);
+ if (shift >= (SHIFT ? 32 : 16)) {
+ for (i = 0; i < (1 << SHIFT); i++) {
+ d->Q(i) = 0;
+ }
} else {
shift <<= 3;
#define SHR(v, i) (i < 64 && i > -64 ? i > 0 ? v >> (i) : (v << -(i)) : 0)
#if SHIFT == 0
- r.Q(0) = SHR(s->Q(0), shift - 0) |
- SHR(d->Q(0), shift - 64);
+ d->Q(0) = SHR(s->Q(0), shift - 0) |
+ SHR(v->Q(0), shift - 64);
#else
- r.Q(0) = SHR(s->Q(0), shift - 0) |
- SHR(s->Q(1), shift - 64) |
- SHR(d->Q(0), shift - 128) |
- SHR(d->Q(1), shift - 192);
- r.Q(1) = SHR(s->Q(0), shift + 64) |
- SHR(s->Q(1), shift - 0) |
- SHR(d->Q(0), shift - 64) |
- SHR(d->Q(1), shift - 128);
+ for (i = 0; i < (1 << SHIFT); i += 2) {
+ uint64_t r0, r1;
+
+ r0 = SHR(s->Q(i), shift - 0) |
+ SHR(s->Q(i + 1), shift - 64) |
+ SHR(v->Q(i), shift - 128) |
+ SHR(v->Q(i + 1), shift - 192);
+ r1 = SHR(s->Q(i), shift + 64) |
+ SHR(s->Q(i + 1), shift - 0) |
+ SHR(v->Q(i), shift - 64) |
+ SHR(v->Q(i + 1), shift - 128);
+ d->Q(i) = r0;
+ d->Q(i + 1) = r1;
+ }
#endif
#undef SHR
}
-
- MOVE(*d, r);
}
#define XMM0 (env->xmm_regs[0])
@@ -1682,17 +1677,23 @@ SSE_HELPER_Q(helper_pcmpeqq, FCMPEQQ)
void glue(helper_packusdw, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
{
- Reg r;
+ Reg *v = d;
+ uint16_t r[8];
+ int i, j, k;
- r.W(0) = satuw((int32_t) d->L(0));
- r.W(1) = satuw((int32_t) d->L(1));
- r.W(2) = satuw((int32_t) d->L(2));
- r.W(3) = satuw((int32_t) d->L(3));
- r.W(4) = satuw((int32_t) s->L(0));
- r.W(5) = satuw((int32_t) s->L(1));
- r.W(6) = satuw((int32_t) s->L(2));
- r.W(7) = satuw((int32_t) s->L(3));
- MOVE(*d, r);
+ for (i = 0, j = 0; i <= 2 << SHIFT; i += 8, j += 4) {
+ r[0] = satuw(v->L(j));
+ r[1] = satuw(v->L(j + 1));
+ r[2] = satuw(v->L(j + 2));
+ r[3] = satuw(v->L(j + 3));
+ r[4] = satuw(s->L(j));
+ r[5] = satuw(s->L(j + 1));
+ r[6] = satuw(s->L(j + 2));
+ r[7] = satuw(s->L(j + 3));
+ for (k = 0; k < 8; k++) {
+ d->W(i + k) = r[k];
+ }
+ }
}
#define FMINSB(d, s) MIN((int8_t)d, (int8_t)s)
@@ -1948,20 +1949,22 @@ void glue(helper_dppd, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, uint32_t mask)
void glue(helper_mpsadbw, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
uint32_t offset)
{
+ Reg *v = d;
int s0 = (offset & 3) << 2;
int d0 = (offset & 4) << 0;
int i;
- Reg r;
+ uint16_t r[8];
for (i = 0; i < 8; i++, d0++) {
- r.W(i) = 0;
- r.W(i) += abs1(d->B(d0 + 0) - s->B(s0 + 0));
- r.W(i) += abs1(d->B(d0 + 1) - s->B(s0 + 1));
- r.W(i) += abs1(d->B(d0 + 2) - s->B(s0 + 2));
- r.W(i) += abs1(d->B(d0 + 3) - s->B(s0 + 3));
+ r[i] = 0;
+ r[i] += abs1(v->B(d0 + 0) - s->B(s0 + 0));
+ r[i] += abs1(v->B(d0 + 1) - s->B(s0 + 1));
+ r[i] += abs1(v->B(d0 + 2) - s->B(s0 + 2));
+ r[i] += abs1(v->B(d0 + 3) - s->B(s0 + 3));
+ }
+ for (i = 0; i < 8; i++) {
+ d->W(i) = r[i];
}
-
- MOVE(*d, r);
}
/* SSE4.2 op helpers */
--
2.37.1On 8/25/22 15:14, Paolo Bonzini wrote:
> void glue(helper_pshufd, SUFFIX)(Reg *d, Reg *s, int order)
> {
> - Reg r;
> + uint32_t r0, r1, r2, r3;
>
> - r.L(0) = s->L(order & 3);
> - r.L(1) = s->L((order >> 2) & 3);
> - r.L(2) = s->L((order >> 4) & 3);
> - r.L(3) = s->L((order >> 6) & 3);
> - MOVE(*d, r);
> + SHUFFLE4(L, s, s, 0);
> +#if SHIFT == 2
> + SHUFFLE4(L, s, s, 4);
> +#endif
> }
Why the if and not a loop?
for (int offset = 0; offset < 2 << SHIFT; offset += 4) {
SHUFFLE4(L, s, s, offset);
}
Most of the other shuffles haven't been updated for ymm.
Perhaps this too should be deferred to the future patch?
> +#if SHIFT == 0
> +#define PACK_WIDTH 4
> +#else
> +#define PACK_WIDTH 8
> +#endif
>
....
> +#define PACK4(F, to, reg, from) do { \
> + r[to + 0] = F((int16_t)reg->W(from + 0)); \
> + r[to + 1] = F((int16_t)reg->W(from + 1)); \
> + r[to + 2] = F((int16_t)reg->W(from + 2)); \
> + r[to + 3] = F((int16_t)reg->W(from + 3)); \
> + } while (0)
...
> +
> +#define PACK_HELPER_B(name, F) \
> +void glue(helper_pack ## name, SUFFIX)(CPUX86State *env, \
> + Reg *d, Reg *s) \
> +{ \
> + Reg *v = d; \
> + uint8_t r[PACK_WIDTH * 2]; \
> + int i; \
> + PACK4(F, 0, v, 0); \
> + PACK4(F, PACK_WIDTH, s, 0); \
> + XMM_ONLY( \
> + PACK4(F, 4, v, 4); \
> + PACK4(F, 12, s, 4); \
> + ) \
The 4 in PACK4 looks suspiciously like it should be PACK_WIDTH --
that would eliminate the XMM_ONLY bit here.
> void glue(helper_packssdw, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
> {
> + Reg *v = d;
> + uint16_t r[PACK_WIDTH];
> + int i, j, k;
>
> + for (i = 0, j = 0; i <= 2 << SHIFT; i += 8, j += 4) {
> + r[0] = satsw(v->L(j));
> + r[1] = satsw(v->L(j + 1));
> + r[PACK_WIDTH / 2 + 0] = satsw(s->L(j));
> + r[PACK_WIDTH / 2 + 1] = satsw(s->L(j + 1));
> +#if SHIFT >= 1
> + r[2] = satsw(v->L(j + 2));
> + r[3] = satsw(v->L(j + 3));
> + r[6] = satsw(s->L(j + 2));
> + r[7] = satsw(s->L(j + 3));
> #endif
> + for (k = 0; k < PACK_WIDTH; k++) {
> + d->W(i + k) = r[k];
> + }
> + }
Hmm. Better with nested loops?
for (j = 0; j <= 1 << SHIFT; j += 4) {
for (k = 0; k < PACK_WIDTH / 2; ++k) {
r[k] = satsw(v->L(j + k));
}
for (k = 0; k < PACK_WIDTH / 2; ++k) {
r[PACK_WIDTH / 2 + k] = satsw(s->L(j + k));
}
for (k = 0; k < PACK_WIDTH; k++) {
d->W(j * 2 + k) = r[k];
}
}
> #define UNPCK_OP(base_name, base) \
> \
> void glue(helper_punpck ## base_name ## bw, SUFFIX)(CPUX86State *env,\
> - Reg *d, Reg *s) \
> + Reg *d, Reg *s) \
> { \
> - Reg r; \
> + Reg *v = d; \
> + uint8_t r[PACK_WIDTH * 2]; \
> + int i; \
> \
> - r.B(0) = d->B((base << (SHIFT + 2)) + 0); \
> - r.B(1) = s->B((base << (SHIFT + 2)) + 0); \
> - r.B(2) = d->B((base << (SHIFT + 2)) + 1); \
> - r.B(3) = s->B((base << (SHIFT + 2)) + 1); \
> - r.B(4) = d->B((base << (SHIFT + 2)) + 2); \
> - r.B(5) = s->B((base << (SHIFT + 2)) + 2); \
> - r.B(6) = d->B((base << (SHIFT + 2)) + 3); \
> - r.B(7) = s->B((base << (SHIFT + 2)) + 3); \
> + r[0] = v->B((base * PACK_WIDTH) + 0); \
> + r[1] = s->B((base * PACK_WIDTH) + 0); \
> + r[2] = v->B((base * PACK_WIDTH) + 1); \
> + r[3] = s->B((base * PACK_WIDTH) + 1); \
> + r[4] = v->B((base * PACK_WIDTH) + 2); \
> + r[5] = s->B((base * PACK_WIDTH) + 2); \
> + r[6] = v->B((base * PACK_WIDTH) + 3); \
> + r[7] = s->B((base * PACK_WIDTH) + 3); \
> XMM_ONLY( \
> - r.B(8) = d->B((base << (SHIFT + 2)) + 4); \
> - r.B(9) = s->B((base << (SHIFT + 2)) + 4); \
> - r.B(10) = d->B((base << (SHIFT + 2)) + 5); \
> - r.B(11) = s->B((base << (SHIFT + 2)) + 5); \
> - r.B(12) = d->B((base << (SHIFT + 2)) + 6); \
> - r.B(13) = s->B((base << (SHIFT + 2)) + 6); \
> - r.B(14) = d->B((base << (SHIFT + 2)) + 7); \
> - r.B(15) = s->B((base << (SHIFT + 2)) + 7); \
> + r[8] = v->B((base * PACK_WIDTH) + 4); \
> + r[9] = s->B((base * PACK_WIDTH) + 4); \
> + r[10] = v->B((base * PACK_WIDTH) + 5); \
> + r[11] = s->B((base * PACK_WIDTH) + 5); \
> + r[12] = v->B((base * PACK_WIDTH) + 6); \
> + r[13] = s->B((base * PACK_WIDTH) + 6); \
> + r[14] = v->B((base * PACK_WIDTH) + 7); \
> + r[15] = s->B((base * PACK_WIDTH) + 7); \
> ) \
> - MOVE(*d, r); \
> + for (i = 0; i < PACK_WIDTH * 2; i++) { \
> + d->B(i) = r[i]; \
> + } \
> } \
Surely better with loops, and more semi-conversions following.
r~
© 2016 - 2026 Red Hat, Inc.