aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorDavid Hildenbrand <david@redhat.com>2021-06-08 11:23:34 +0200
committerCornelia Huck <cohuck@redhat.com>2021-06-21 08:48:21 +0200
commitda4807527f3bda115606b4647fdc2f87928d0f15 (patch)
tree5b55c4dbcca3073edb341c9413b40489820816dc
parentf02497306ec2efb8315f604597ddf8142f00336d (diff)
s390x/tcg: Implement VECTOR FP (MAXIMUM|MINIMUM)
For IEEE functions, we can reuse the softfloat implementations. For the other functions, implement it generically for 32bit/64bit/128bit - carefully taking care of all weird special cases according to the tables defined in the PoP. Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Signed-off-by: David Hildenbrand <david@redhat.com> Message-Id: <20210608092337.12221-24-david@redhat.com> Signed-off-by: Cornelia Huck <cohuck@redhat.com>
-rw-r--r--target/s390x/helper.h6
-rw-r--r--target/s390x/insn-data.def4
-rw-r--r--target/s390x/internal.h9
-rw-r--r--target/s390x/translate_vx.c.inc44
-rw-r--r--target/s390x/vec_fpu_helper.c328
5 files changed, 391 insertions, 0 deletions
diff --git a/target/s390x/helper.h b/target/s390x/helper.h
index 913967ce4e..ba045f559d 100644
--- a/target/s390x/helper.h
+++ b/target/s390x/helper.h
@@ -291,6 +291,12 @@ DEF_HELPER_FLAGS_4(gvec_vflr128, TCG_CALL_NO_WG, void, ptr, cptr, env, i32)
DEF_HELPER_FLAGS_5(gvec_vfm32, 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_vfm128, TCG_CALL_NO_WG, void, ptr, cptr, cptr, env, i32)
+DEF_HELPER_FLAGS_5(gvec_vfmax32, TCG_CALL_NO_WG, void, ptr, cptr, cptr, env, i32)
+DEF_HELPER_FLAGS_5(gvec_vfmax64, TCG_CALL_NO_WG, void, ptr, cptr, cptr, env, i32)
+DEF_HELPER_FLAGS_5(gvec_vfmax128, TCG_CALL_NO_WG, void, ptr, cptr, cptr, env, i32)
+DEF_HELPER_FLAGS_5(gvec_vfmin32, TCG_CALL_NO_WG, void, ptr, cptr, cptr, env, i32)
+DEF_HELPER_FLAGS_5(gvec_vfmin64, TCG_CALL_NO_WG, void, ptr, cptr, cptr, env, i32)
+DEF_HELPER_FLAGS_5(gvec_vfmin128, 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_vfma64, 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)
diff --git a/target/s390x/insn-data.def b/target/s390x/insn-data.def
index 19b02dffca..3e5594210c 100644
--- a/target/s390x/insn-data.def
+++ b/target/s390x/insn-data.def
@@ -1253,6 +1253,10 @@
F(0xe7c4, VFLL, VRR_a, V, 0, 0, 0, 0, vfll, 0, IF_VEC)
/* VECTOR FP LOAD ROUNDED */
F(0xe7c5, VFLR, VRR_a, V, 0, 0, 0, 0, vcdg, 0, IF_VEC)
+/* VECTOR FP MAXIMUM */
+ F(0xe7ef, VFMAX, VRR_c, VE, 0, 0, 0, 0, vfmax, 0, IF_VEC)
+/* VECTOR FP MINIMUM */
+ F(0xe7ee, VFMIN, VRR_c, VE, 0, 0, 0, 0, vfmax, 0, IF_VEC)
/* VECTOR FP MULTIPLY */
F(0xe7e7, VFM, VRR_c, V, 0, 0, 0, 0, vfa, 0, IF_VEC)
/* VECTOR FP MULTIPLY AND ADD */
diff --git a/target/s390x/internal.h b/target/s390x/internal.h
index 11515bb617..d62dfc4dc6 100644
--- a/target/s390x/internal.h
+++ b/target/s390x/internal.h
@@ -288,6 +288,15 @@ uint8_t s390_softfloat_exc_to_ieee(unsigned int exc);
int s390_swap_bfp_rounding_mode(CPUS390XState *env, int m3);
void s390_restore_bfp_rounding_mode(CPUS390XState *env, int old_mode);
int float_comp_to_cc(CPUS390XState *env, int float_compare);
+
+#define DCMASK_ZERO 0x0c00
+#define DCMASK_NORMAL 0x0300
+#define DCMASK_SUBNORMAL 0x00c0
+#define DCMASK_INFINITY 0x0030
+#define DCMASK_QUIET_NAN 0x000c
+#define DCMASK_SIGNALING_NAN 0x0003
+#define DCMASK_NAN 0x000f
+#define DCMASK_NEGATIVE 0x0555
uint16_t float32_dcmask(CPUS390XState *env, float32 f1);
uint16_t float64_dcmask(CPUS390XState *env, float64 f1);
uint16_t float128_dcmask(CPUS390XState *env, float128 f1);
diff --git a/target/s390x/translate_vx.c.inc b/target/s390x/translate_vx.c.inc
index 200d83e783..a9d51b1f4c 100644
--- a/target/s390x/translate_vx.c.inc
+++ b/target/s390x/translate_vx.c.inc
@@ -2814,6 +2814,50 @@ static DisasJumpType op_vfll(DisasContext *s, DisasOps *o)
return DISAS_NEXT;
}
+static DisasJumpType op_vfmax(DisasContext *s, DisasOps *o)
+{
+ const uint8_t fpf = get_field(s, m4);
+ const uint8_t m6 = get_field(s, m6);
+ const uint8_t m5 = get_field(s, m5);
+ gen_helper_gvec_3_ptr *fn;
+
+ if (m6 == 5 || m6 == 6 || m6 == 7 || m6 > 13) {
+ gen_program_exception(s, PGM_SPECIFICATION);
+ return DISAS_NORETURN;
+ }
+
+ switch (fpf) {
+ case FPF_SHORT:
+ if (s->fields.op2 == 0xef) {
+ fn = gen_helper_gvec_vfmax32;
+ } else {
+ fn = gen_helper_gvec_vfmin32;
+ }
+ break;
+ case FPF_LONG:
+ if (s->fields.op2 == 0xef) {
+ fn = gen_helper_gvec_vfmax64;
+ } else {
+ fn = gen_helper_gvec_vfmin64;
+ }
+ break;
+ case FPF_EXT:
+ if (s->fields.op2 == 0xef) {
+ fn = gen_helper_gvec_vfmax128;
+ } else {
+ fn = gen_helper_gvec_vfmin128;
+ }
+ break;
+ default:
+ gen_program_exception(s, PGM_SPECIFICATION);
+ return DISAS_NORETURN;
+ }
+
+ gen_gvec_3_ptr(get_field(s, v1), get_field(s, v2), get_field(s, v3),
+ cpu_env, deposit32(m5, 4, 4, m6), fn);
+ return DISAS_NEXT;
+}
+
static DisasJumpType op_vfma(DisasContext *s, DisasOps *o)
{
const uint8_t m5 = get_field(s, m5);
diff --git a/target/s390x/vec_fpu_helper.c b/target/s390x/vec_fpu_helper.c
index dc9bcc90a7..8e2b274547 100644
--- a/target/s390x/vec_fpu_helper.c
+++ b/target/s390x/vec_fpu_helper.c
@@ -742,3 +742,331 @@ void HELPER(gvec_vftci128)(void *v1, const void *v2, CPUS390XState *env,
s390_vec_write_element64(v1, 1, 0);
}
}
+
+typedef enum S390MinMaxType {
+ S390_MINMAX_TYPE_IEEE = 0,
+ S390_MINMAX_TYPE_JAVA,
+ S390_MINMAX_TYPE_C_MACRO,
+ S390_MINMAX_TYPE_CPP,
+ S390_MINMAX_TYPE_F,
+} S390MinMaxType;
+
+typedef enum S390MinMaxRes {
+ S390_MINMAX_RES_MINMAX = 0,
+ S390_MINMAX_RES_A,
+ S390_MINMAX_RES_B,
+ S390_MINMAX_RES_SILENCE_A,
+ S390_MINMAX_RES_SILENCE_B,
+} S390MinMaxRes;
+
+static S390MinMaxRes vfmin_res(uint16_t dcmask_a, uint16_t dcmask_b,
+ S390MinMaxType type, float_status *s)
+{
+ const bool neg_a = dcmask_a & DCMASK_NEGATIVE;
+ const bool nan_a = dcmask_a & DCMASK_NAN;
+ const bool nan_b = dcmask_b & DCMASK_NAN;
+
+ g_assert(type > S390_MINMAX_TYPE_IEEE && type <= S390_MINMAX_TYPE_F);
+
+ if (unlikely((dcmask_a | dcmask_b) & DCMASK_NAN)) {
+ const bool sig_a = dcmask_a & DCMASK_SIGNALING_NAN;
+ const bool sig_b = dcmask_b & DCMASK_SIGNALING_NAN;
+
+ if ((dcmask_a | dcmask_b) & DCMASK_SIGNALING_NAN) {
+ s->float_exception_flags |= float_flag_invalid;
+ }
+ switch (type) {
+ case S390_MINMAX_TYPE_JAVA:
+ if (sig_a) {
+ return S390_MINMAX_RES_SILENCE_A;
+ } else if (sig_b) {
+ return S390_MINMAX_RES_SILENCE_B;
+ }
+ return nan_a ? S390_MINMAX_RES_A : S390_MINMAX_RES_B;
+ case S390_MINMAX_TYPE_F:
+ return nan_b ? S390_MINMAX_RES_A : S390_MINMAX_RES_B;
+ case S390_MINMAX_TYPE_C_MACRO:
+ s->float_exception_flags |= float_flag_invalid;
+ return S390_MINMAX_RES_B;
+ case S390_MINMAX_TYPE_CPP:
+ s->float_exception_flags |= float_flag_invalid;
+ return S390_MINMAX_RES_A;
+ default:
+ g_assert_not_reached();
+ }
+ } else if (unlikely(dcmask_a & dcmask_b & DCMASK_ZERO)) {
+ switch (type) {
+ case S390_MINMAX_TYPE_JAVA:
+ return neg_a ? S390_MINMAX_RES_A : S390_MINMAX_RES_B;
+ case S390_MINMAX_TYPE_C_MACRO:
+ return S390_MINMAX_RES_B;
+ case S390_MINMAX_TYPE_F:
+ return !neg_a ? S390_MINMAX_RES_B : S390_MINMAX_RES_A;
+ case S390_MINMAX_TYPE_CPP:
+ return S390_MINMAX_RES_A;
+ default:
+ g_assert_not_reached();
+ }
+ }
+ return S390_MINMAX_RES_MINMAX;
+}
+
+static S390MinMaxRes vfmax_res(uint16_t dcmask_a, uint16_t dcmask_b,
+ S390MinMaxType type, float_status *s)
+{
+ g_assert(type > S390_MINMAX_TYPE_IEEE && type <= S390_MINMAX_TYPE_F);
+
+ if (unlikely((dcmask_a | dcmask_b) & DCMASK_NAN)) {
+ const bool sig_a = dcmask_a & DCMASK_SIGNALING_NAN;
+ const bool sig_b = dcmask_b & DCMASK_SIGNALING_NAN;
+ const bool nan_a = dcmask_a & DCMASK_NAN;
+ const bool nan_b = dcmask_b & DCMASK_NAN;
+
+ if ((dcmask_a | dcmask_b) & DCMASK_SIGNALING_NAN) {
+ s->float_exception_flags |= float_flag_invalid;
+ }
+ switch (type) {
+ case S390_MINMAX_TYPE_JAVA:
+ if (sig_a) {
+ return S390_MINMAX_RES_SILENCE_A;
+ } else if (sig_b) {
+ return S390_MINMAX_RES_SILENCE_B;
+ }
+ return nan_a ? S390_MINMAX_RES_A : S390_MINMAX_RES_B;
+ case S390_MINMAX_TYPE_F:
+ return nan_b ? S390_MINMAX_RES_A : S390_MINMAX_RES_B;
+ case S390_MINMAX_TYPE_C_MACRO:
+ s->float_exception_flags |= float_flag_invalid;
+ return S390_MINMAX_RES_B;
+ case S390_MINMAX_TYPE_CPP:
+ s->float_exception_flags |= float_flag_invalid;
+ return S390_MINMAX_RES_A;
+ default:
+ g_assert_not_reached();
+ }
+ } else if (unlikely(dcmask_a & dcmask_b & DCMASK_ZERO)) {
+ const bool neg_a = dcmask_a & DCMASK_NEGATIVE;
+
+ switch (type) {
+ case S390_MINMAX_TYPE_JAVA:
+ case S390_MINMAX_TYPE_F:
+ return neg_a ? S390_MINMAX_RES_B : S390_MINMAX_RES_A;
+ case S390_MINMAX_TYPE_C_MACRO:
+ return S390_MINMAX_RES_B;
+ case S390_MINMAX_TYPE_CPP:
+ return S390_MINMAX_RES_A;
+ default:
+ g_assert_not_reached();
+ }
+ }
+ return S390_MINMAX_RES_MINMAX;
+}
+
+static S390MinMaxRes vfminmax_res(uint16_t dcmask_a, uint16_t dcmask_b,
+ S390MinMaxType type, bool is_min,
+ float_status *s)
+{
+ return is_min ? vfmin_res(dcmask_a, dcmask_b, type, s) :
+ vfmax_res(dcmask_a, dcmask_b, type, s);
+}
+
+static void vfminmax32(S390Vector *v1, const S390Vector *v2,
+ const S390Vector *v3, CPUS390XState *env,
+ S390MinMaxType type, bool is_min, bool is_abs, bool se,
+ uintptr_t retaddr)
+{
+ float_status *s = &env->fpu_status;
+ uint8_t vxc, vec_exc = 0;
+ S390Vector tmp = {};
+ int i;
+
+ for (i = 0; i < 4; i++) {
+ float32 a = s390_vec_read_float32(v2, i);
+ float32 b = s390_vec_read_float32(v3, i);
+ float32 result;
+
+ if (type != S390_MINMAX_TYPE_IEEE) {
+ S390MinMaxRes res;
+
+ if (is_abs) {
+ a = float32_abs(a);
+ b = float32_abs(b);
+ }
+
+ res = vfminmax_res(float32_dcmask(env, a), float32_dcmask(env, b),
+ type, is_min, s);
+ switch (res) {
+ case S390_MINMAX_RES_MINMAX:
+ result = is_min ? float32_min(a, b, s) : float32_max(a, b, s);
+ break;
+ case S390_MINMAX_RES_A:
+ result = a;
+ break;
+ case S390_MINMAX_RES_B:
+ result = b;
+ break;
+ case S390_MINMAX_RES_SILENCE_A:
+ result = float32_silence_nan(a, s);
+ break;
+ case S390_MINMAX_RES_SILENCE_B:
+ result = float32_silence_nan(b, s);
+ break;
+ default:
+ g_assert_not_reached();
+ }
+ } else if (!is_abs) {
+ result = is_min ? float32_minnum(a, b, &env->fpu_status) :
+ float32_maxnum(a, b, &env->fpu_status);
+ } else {
+ result = is_min ? float32_minnummag(a, b, &env->fpu_status) :
+ float32_maxnummag(a, b, &env->fpu_status);
+ }
+
+ s390_vec_write_float32(&tmp, i, result);
+ vxc = check_ieee_exc(env, i, false, &vec_exc);
+ if (se || vxc) {
+ break;
+ }
+ }
+ handle_ieee_exc(env, vxc, vec_exc, retaddr);
+ *v1 = tmp;
+}
+
+static void vfminmax64(S390Vector *v1, const S390Vector *v2,
+ const S390Vector *v3, CPUS390XState *env,
+ S390MinMaxType type, bool is_min, bool is_abs, bool se,
+ uintptr_t retaddr)
+{
+ float_status *s = &env->fpu_status;
+ uint8_t vxc, vec_exc = 0;
+ S390Vector tmp = {};
+ int i;
+
+ for (i = 0; i < 2; i++) {
+ float64 a = s390_vec_read_float64(v2, i);
+ float64 b = s390_vec_read_float64(v3, i);
+ float64 result;
+
+ if (type != S390_MINMAX_TYPE_IEEE) {
+ S390MinMaxRes res;
+
+ if (is_abs) {
+ a = float64_abs(a);
+ b = float64_abs(b);
+ }
+
+ res = vfminmax_res(float64_dcmask(env, a), float64_dcmask(env, b),
+ type, is_min, s);
+ switch (res) {
+ case S390_MINMAX_RES_MINMAX:
+ result = is_min ? float64_min(a, b, s) : float64_max(a, b, s);
+ break;
+ case S390_MINMAX_RES_A:
+ result = a;
+ break;
+ case S390_MINMAX_RES_B:
+ result = b;
+ break;
+ case S390_MINMAX_RES_SILENCE_A:
+ result = float64_silence_nan(a, s);
+ break;
+ case S390_MINMAX_RES_SILENCE_B:
+ result = float64_silence_nan(b, s);
+ break;
+ default:
+ g_assert_not_reached();
+ }
+ } else if (!is_abs) {
+ result = is_min ? float64_minnum(a, b, &env->fpu_status) :
+ float64_maxnum(a, b, &env->fpu_status);
+ } else {
+ result = is_min ? float64_minnummag(a, b, &env->fpu_status) :
+ float64_maxnummag(a, b, &env->fpu_status);
+ }
+
+ s390_vec_write_float64(&tmp, i, result);
+ vxc = check_ieee_exc(env, i, false, &vec_exc);
+ if (se || vxc) {
+ break;
+ }
+ }
+ handle_ieee_exc(env, vxc, vec_exc, retaddr);
+ *v1 = tmp;
+}
+
+static void vfminmax128(S390Vector *v1, const S390Vector *v2,
+ const S390Vector *v3, CPUS390XState *env,
+ S390MinMaxType type, bool is_min, bool is_abs, bool se,
+ uintptr_t retaddr)
+{
+ float128 a = s390_vec_read_float128(v2);
+ float128 b = s390_vec_read_float128(v3);
+ float_status *s = &env->fpu_status;
+ uint8_t vxc, vec_exc = 0;
+ float128 result;
+
+ if (type != S390_MINMAX_TYPE_IEEE) {
+ S390MinMaxRes res;
+
+ if (is_abs) {
+ a = float128_abs(a);
+ b = float128_abs(b);
+ }
+
+ res = vfminmax_res(float128_dcmask(env, a), float128_dcmask(env, b),
+ type, is_min, s);
+ switch (res) {
+ case S390_MINMAX_RES_MINMAX:
+ result = is_min ? float128_min(a, b, s) : float128_max(a, b, s);
+ break;
+ case S390_MINMAX_RES_A:
+ result = a;
+ break;
+ case S390_MINMAX_RES_B:
+ result = b;
+ break;
+ case S390_MINMAX_RES_SILENCE_A:
+ result = float128_silence_nan(a, s);
+ break;
+ case S390_MINMAX_RES_SILENCE_B:
+ result = float128_silence_nan(b, s);
+ break;
+ default:
+ g_assert_not_reached();
+ }
+ } else if (!is_abs) {
+ result = is_min ? float128_minnum(a, b, &env->fpu_status) :
+ float128_maxnum(a, b, &env->fpu_status);
+ } else {
+ result = is_min ? float128_minnummag(a, b, &env->fpu_status) :
+ float128_maxnummag(a, b, &env->fpu_status);
+ }
+
+ vxc = check_ieee_exc(env, 0, false, &vec_exc);
+ handle_ieee_exc(env, vxc, vec_exc, retaddr);
+ s390_vec_write_float128(v1, result);
+}
+
+#define DEF_GVEC_VFMINMAX_B(NAME, IS_MIN, BITS) \
+void HELPER(gvec_##NAME##BITS)(void *v1, const void *v2, const void *v3, \
+ CPUS390XState *env, uint32_t desc) \
+{ \
+ const bool se = extract32(simd_data(desc), 3, 1); \
+ uint8_t type = extract32(simd_data(desc), 4, 4); \
+ bool is_abs = false; \
+ \
+ if (type >= 8) { \
+ is_abs = true; \
+ type -= 8; \
+ } \
+ \
+ vfminmax##BITS(v1, v2, v3, env, type, IS_MIN, is_abs, se, GETPC()); \
+}
+
+#define DEF_GVEC_VFMINMAX(NAME, IS_MIN) \
+ DEF_GVEC_VFMINMAX_B(NAME, IS_MIN, 32) \
+ DEF_GVEC_VFMINMAX_B(NAME, IS_MIN, 64) \
+ DEF_GVEC_VFMINMAX_B(NAME, IS_MIN, 128)
+
+DEF_GVEC_VFMINMAX(vfmax, false)
+DEF_GVEC_VFMINMAX(vfmin, true)