[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[PATCH v1 16/20] s390x/tcg: Implement 32/128bit for VECTOR FP MULTIPLY A
|
From: |
David Hildenbrand |
|
Subject: |
[PATCH v1 16/20] s390x/tcg: Implement 32/128bit for VECTOR FP MULTIPLY AND (ADD|SUBTRACT) |
|
Date: |
Wed, 30 Sep 2020 16:55:19 +0200 |
Signed-off-by: David Hildenbrand <david@redhat.com>
---
target/s390x/helper.h | 6 ++
target/s390x/translate_vx.c.inc | 48 +++++++++++++---
target/s390x/vec_fpu_helper.c | 98 ++++++++++++++++++++-------------
3 files changed, 107 insertions(+), 45 deletions(-)
diff --git a/target/s390x/helper.h b/target/s390x/helper.h
index c2ded83669..e4d60299dc 100644
--- a/target/s390x/helper.h
+++ b/target/s390x/helper.h
@@ -317,10 +317,16 @@ DEF_HELPER_FLAGS_5(gvec_vfm32s, TCG_CALL_NO_WG, void,
ptr, cptr, cptr, env, i32)
DEF_HELPER_FLAGS_5(gvec_vfm64, TCG_CALL_NO_WG, void, ptr, cptr, cptr, env, i32)
DEF_HELPER_FLAGS_5(gvec_vfm64s, TCG_CALL_NO_WG, void, ptr, cptr, cptr, env,
i32)
DEF_HELPER_FLAGS_5(gvec_vfm128, TCG_CALL_NO_WG, void, ptr, cptr, cptr, env,
i32)
+DEF_HELPER_FLAGS_6(gvec_vfma32, TCG_CALL_NO_WG, void, ptr, cptr, cptr, cptr,
env, i32)
+DEF_HELPER_FLAGS_6(gvec_vfma32s, TCG_CALL_NO_WG, void, ptr, cptr, cptr, cptr,
env, i32)
DEF_HELPER_FLAGS_6(gvec_vfma64, TCG_CALL_NO_WG, void, ptr, cptr, cptr, cptr,
env, i32)
DEF_HELPER_FLAGS_6(gvec_vfma64s, TCG_CALL_NO_WG, void, ptr, cptr, cptr, cptr,
env, i32)
+DEF_HELPER_FLAGS_6(gvec_vfma128, TCG_CALL_NO_WG, void, ptr, cptr, cptr, cptr,
env, i32)
+DEF_HELPER_FLAGS_6(gvec_vfms32, TCG_CALL_NO_WG, void, ptr, cptr, cptr, cptr,
env, i32)
+DEF_HELPER_FLAGS_6(gvec_vfms32s, TCG_CALL_NO_WG, void, ptr, cptr, cptr, cptr,
env, i32)
DEF_HELPER_FLAGS_6(gvec_vfms64, TCG_CALL_NO_WG, void, ptr, cptr, cptr, cptr,
env, i32)
DEF_HELPER_FLAGS_6(gvec_vfms64s, TCG_CALL_NO_WG, void, ptr, cptr, cptr, cptr,
env, i32)
+DEF_HELPER_FLAGS_6(gvec_vfms128, TCG_CALL_NO_WG, void, ptr, cptr, cptr, cptr,
env, i32)
DEF_HELPER_FLAGS_4(gvec_vfsq32, TCG_CALL_NO_WG, void, ptr, cptr, env, i32)
DEF_HELPER_FLAGS_4(gvec_vfsq32s, TCG_CALL_NO_WG, void, ptr, cptr, env, i32)
DEF_HELPER_FLAGS_4(gvec_vfsq64, TCG_CALL_NO_WG, void, ptr, cptr, env, i32)
diff --git a/target/s390x/translate_vx.c.inc b/target/s390x/translate_vx.c.inc
index 6bd599b319..5d31498cc1 100644
--- a/target/s390x/translate_vx.c.inc
+++ b/target/s390x/translate_vx.c.inc
@@ -2847,18 +2847,52 @@ static DisasJumpType op_vfma(DisasContext *s, DisasOps
*o)
const uint8_t m5 = get_field(s, m5);
const uint8_t fpf = get_field(s, m6);
const bool se = extract32(m5, 3, 1);
- gen_helper_gvec_4_ptr *fn;
+ gen_helper_gvec_4_ptr *fn = NULL;
- if (fpf != FPF_LONG || extract32(m5, 0, 3)) {
+ switch (s->fields.op2) {
+ case 0x8f:
+ switch (fpf) {
+ case FPF_SHORT:
+ if (s390_has_feat(S390_FEAT_VECTOR_ENH)) {
+ fn = se ? gen_helper_gvec_vfma32s : gen_helper_gvec_vfma32;
+ }
+ break;
+ case FPF_LONG:
+ fn = se ? gen_helper_gvec_vfma64s : gen_helper_gvec_vfma64;
+ break;
+ default:
+ if (s390_has_feat(S390_FEAT_VECTOR_ENH)) {
+ fn = gen_helper_gvec_vfma128;
+ }
+ break;
+ }
+ break;
+ case 0x8e:
+ switch (fpf) {
+ case FPF_SHORT:
+ if (s390_has_feat(S390_FEAT_VECTOR_ENH)) {
+ fn = se ? gen_helper_gvec_vfms32s : gen_helper_gvec_vfms32;
+ }
+ break;
+ case FPF_LONG:
+ fn = se ? gen_helper_gvec_vfms64s : gen_helper_gvec_vfms64;
+ break;
+ default:
+ if (s390_has_feat(S390_FEAT_VECTOR_ENH)) {
+ fn = gen_helper_gvec_vfms128;
+ }
+ break;
+ }
+ break;
+ default:
+ g_assert_not_reached();
+ }
+
+ if (!fn || extract32(m5, 0, 3)) {
gen_program_exception(s, PGM_SPECIFICATION);
return DISAS_NORETURN;
}
- if (s->fields.op2 == 0x8f) {
- fn = se ? gen_helper_gvec_vfma64s : gen_helper_gvec_vfma64;
- } else {
- fn = se ? gen_helper_gvec_vfms64s : gen_helper_gvec_vfms64;
- }
gen_gvec_4_ptr(get_field(s, v1), get_field(s, v2),
get_field(s, v3), get_field(s, v4), cpu_env,
0, fn);
diff --git a/target/s390x/vec_fpu_helper.c b/target/s390x/vec_fpu_helper.c
index f18f0ae8e2..0b25718365 100644
--- a/target/s390x/vec_fpu_helper.c
+++ b/target/s390x/vec_fpu_helper.c
@@ -699,53 +699,75 @@ void HELPER(gvec_vfm##BITS##s)(void *v1, const void *v2,
const void *v3, \
DEF_GVEC_FVM_S(32)
DEF_GVEC_FVM_S(64)
-static void vfma64(S390Vector *v1, const S390Vector *v2, const S390Vector *v3,
- const S390Vector *v4, CPUS390XState *env, bool s, int flags,
- uintptr_t retaddr)
-{
- uint8_t vxc, vec_exc = 0;
- S390Vector tmp = {};
- int i;
-
- for (i = 0; i < 2; i++) {
- const uint64_t a = s390_vec_read_element64(v2, i);
- const uint64_t b = s390_vec_read_element64(v3, i);
- const uint64_t c = s390_vec_read_element64(v4, i);
- uint64_t ret = float64_muladd(a, b, c, flags, &env->fpu_status);
-
- s390_vec_write_element64(&tmp, i, ret);
- vxc = check_ieee_exc(env, i, false, &vec_exc);
- if (s || vxc) {
- break;
- }
- }
- handle_ieee_exc(env, vxc, vec_exc, retaddr);
- *v1 = tmp;
+#define DEF_VFMA(BITS)
\
+static void vfma##BITS(S390Vector *v1, const S390Vector *v2,
\
+ const S390Vector *v3, const S390Vector *v4,
\
+ CPUS390XState *env, bool s, int flags,
\
+ uintptr_t retaddr)
\
+{
\
+ uint8_t vxc, vec_exc = 0;
\
+ S390Vector tmp = {};
\
+ int i;
\
+
\
+ for (i = 0; i < (128 / BITS); i++) {
\
+ const float##BITS a = s390_vec_read_float##BITS(v2, i);
\
+ const float##BITS b = s390_vec_read_float##BITS(v3, i);
\
+ const float##BITS c = s390_vec_read_float##BITS(v4, i);
\
+ float##BITS ret = float##BITS##_muladd(a, b, c, flags,
\
+ &env->fpu_status);
\
+
\
+ s390_vec_write_float##BITS(&tmp, i, ret);
\
+ vxc = check_ieee_exc(env, i, false, &vec_exc);
\
+ if (s || vxc) {
\
+ break;
\
+ }
\
+ }
\
+ handle_ieee_exc(env, vxc, vec_exc, retaddr);
\
+ *v1 = tmp;
\
}
+DEF_VFMA(32)
+DEF_VFMA(64)
+DEF_VFMA(128)
-void HELPER(gvec_vfma64)(void *v1, const void *v2, const void *v3,
- const void *v4, CPUS390XState *env, uint32_t desc)
-{
- vfma64(v1, v2, v3, v4, env, false, 0, GETPC());
+#define DEF_GVEC_VFMA(BITS)
\
+void HELPER(gvec_vfma##BITS)(void *v1, const void *v2, const void *v3,
\
+ const void *v4, CPUS390XState *env, uint32_t
desc)\
+{
\
+ vfma##BITS(v1, v2, v3, v4, env, false, 0, GETPC());
\
}
+DEF_GVEC_VFMA(32)
+DEF_GVEC_VFMA(64)
+DEF_GVEC_VFMA(128)
-void HELPER(gvec_vfma64s)(void *v1, const void *v2, const void *v3,
- const void *v4, CPUS390XState *env, uint32_t desc)
-{
- vfma64(v1, v2, v3, v4, env, true, 0, GETPC());
+#define DEF_GVEC_VFMA_S(BITS)
\
+void HELPER(gvec_vfma##BITS##s)(void *v1, const void *v2, const void *v3,
\
+ const void *v4, CPUS390XState *env,
\
+ uint32_t desc)
\
+{
\
+ vfma##BITS(v1, v2, v3, v4, env, true, 0, GETPC());
\
}
+DEF_GVEC_VFMA_S(32)
+DEF_GVEC_VFMA_S(64)
-void HELPER(gvec_vfms64)(void *v1, const void *v2, const void *v3,
- const void *v4, CPUS390XState *env, uint32_t desc)
-{
- vfma64(v1, v2, v3, v4, env, false, float_muladd_negate_c, GETPC());
+#define DEF_GVEC_VFMS(BITS)
\
+void HELPER(gvec_vfms##BITS)(void *v1, const void *v2, const void *v3,
\
+ const void *v4, CPUS390XState *env, uint32_t
desc)\
+{
\
+ vfma##BITS(v1, v2, v3, v4, env, false, float_muladd_negate_c, GETPC());
\
}
+DEF_GVEC_VFMS(32)
+DEF_GVEC_VFMS(64)
+DEF_GVEC_VFMS(128)
-void HELPER(gvec_vfms64s)(void *v1, const void *v2, const void *v3,
- const void *v4, CPUS390XState *env, uint32_t desc)
-{
- vfma64(v1, v2, v3, v4, env, true, float_muladd_negate_c, GETPC());
+#define DEF_GVEC_VFMS_S(BITS)
\
+void HELPER(gvec_vfms##BITS##s)(void *v1, const void *v2, const void *v3,
\
+ const void *v4, CPUS390XState *env,
\
+ uint32_t desc)
\
+{
\
+ vfma##BITS(v1, v2, v3, v4, env, true, float_muladd_negate_c, GETPC());
\
}
+DEF_GVEC_VFMS_S(32)
+DEF_GVEC_VFMS_S(64)
#define DEF_GVEC_VFSQ(BITS)
\
void HELPER(gvec_vfsq##BITS)(void *v1, const void *v2, CPUS390XState *env,
\
--
2.26.2
- [PATCH v1 06/20] s390x/tcg: Implement 32/128 bit for VECTOR FP MULTIPLY, (continued)
- [PATCH v1 06/20] s390x/tcg: Implement 32/128 bit for VECTOR FP MULTIPLY, David Hildenbrand, 2020/09/30
- [PATCH v1 07/20] s390x/tcg: Implement 32/128 bit for VECTOR FP SUBTRACT, David Hildenbrand, 2020/09/30
- [PATCH v1 09/20] s390x/tcg: Implement 32/128 bit for VECTOR FP COMPARE *, David Hildenbrand, 2020/09/30
- [PATCH v1 08/20] s390x/tcg: Implement 32/128 bit for VECTOR FP COMPARE (AND SIGNAL) SCALAR, David Hildenbrand, 2020/09/30
- [PATCH v1 12/20] s390x/tcg: Implement 128 bit for VECTOR FP LOAD ROUNDED, David Hildenbrand, 2020/09/30
- [PATCH v1 10/20] s390x/tcg: Implement 32/128 bit for VECTOR LOAD FP INTEGER, David Hildenbrand, 2020/09/30
- [PATCH v1 13/20] s390x/tcg: Implement 32/128 bit for VECTOR FP PERFORM SIGN OPERATION, David Hildenbrand, 2020/09/30
- [PATCH v1 11/20] s390x/tcg: Implement 64 bit for VECTOR FP LOAD LENGTHENED, David Hildenbrand, 2020/09/30
- [PATCH v1 15/20] s390x/tcg: Implement 32/128 bit for VECTOR FP TEST DATA CLASS IMMEDIATE, David Hildenbrand, 2020/09/30
- [PATCH v1 17/20] s390x/tcg: Implement VECTOR FP NEGATIVE MULTIPLY AND (ADD|SUBTRACT), David Hildenbrand, 2020/09/30
- [PATCH v1 16/20] s390x/tcg: Implement 32/128bit for VECTOR FP MULTIPLY AND (ADD|SUBTRACT),
David Hildenbrand <=
- [PATCH v1 14/20] s390x/tcg: Implement 32/128 bit for VECTOR FP SQUARE ROOT, David Hildenbrand, 2020/09/30
- [PATCH v1 18/20] s390x/tcg: Implement VECTOR FP (MAXIMUM|MINIMUM), David Hildenbrand, 2020/09/30
- [PATCH v1 19/20] s390x/tcg: We support Vector enhancements facility, David Hildenbrand, 2020/09/30
- [PATCH v1 20/20] s390x/cpumodel: Bump up QEMU model to a stripped-down IBM z14 GA2, David Hildenbrand, 2020/09/30
- Re: [PATCH v1 00/20] s390x/tcg: Implement Vector enhancements facility and switch to z14, no-reply, 2020/09/30