[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[PATCH v5 73/81] target/arm: Implement SVE2 fp multiply-add long
From: |
Richard Henderson |
Subject: |
[PATCH v5 73/81] target/arm: Implement SVE2 fp multiply-add long |
Date: |
Fri, 16 Apr 2021 14:02:32 -0700 |
From: Stephen Long <steplong@quicinc.com>
Implements both vectored and indexed FMLALB, FMLALT, FMLSLB, FMLSLT
Signed-off-by: Stephen Long <steplong@quicinc.com>
Message-Id: <20200504171240.11220-1-steplong@quicinc.com>
[rth: Rearrange to use float16_to_float32_by_bits.]
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
target/arm/helper.h | 5 +++
target/arm/sve.decode | 14 +++++++
target/arm/translate-sve.c | 75 ++++++++++++++++++++++++++++++++++++++
target/arm/vec_helper.c | 51 ++++++++++++++++++++++++++
4 files changed, 145 insertions(+)
diff --git a/target/arm/helper.h b/target/arm/helper.h
index 86f938c938..e8b16a401f 100644
--- a/target/arm/helper.h
+++ b/target/arm/helper.h
@@ -987,6 +987,11 @@ DEF_HELPER_FLAGS_4(sve2_sqrdmulh_idx_s, TCG_CALL_NO_RWG,
DEF_HELPER_FLAGS_4(sve2_sqrdmulh_idx_d, TCG_CALL_NO_RWG,
void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_6(sve2_fmlal_zzzw_s, TCG_CALL_NO_RWG,
+ void, ptr, ptr, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_6(sve2_fmlal_zzxw_s, TCG_CALL_NO_RWG,
+ void, ptr, ptr, ptr, ptr, ptr, i32)
+
DEF_HELPER_FLAGS_4(gvec_xar_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
#ifdef TARGET_AARCH64
diff --git a/target/arm/sve.decode b/target/arm/sve.decode
index cfdee8955b..63870b7539 100644
--- a/target/arm/sve.decode
+++ b/target/arm/sve.decode
@@ -132,6 +132,8 @@
&rrrr_esz ra=%reg_movprfx
# Four operand with unused vector element size
+@rda_rn_rm_e0 ........ ... rm:5 ... ... rn:5 rd:5 \
+ &rrrr_esz esz=0 ra=%reg_movprfx
@rdn_ra_rm_e0 ........ ... rm:5 ... ... ra:5 rd:5 \
&rrrr_esz esz=0 rn=%reg_movprfx
@@ -1559,3 +1561,15 @@ FCVTLT_sd 01100100 11 0010 11 101 ... ..... .....
@rd_pg_rn_e0
### SVE2 floating-point convert to integer
FLOGB 01100101 00 011 esz:2 0101 pg:3 rn:5 rd:5 &rpr_esz
+
+### SVE2 floating-point multiply-add long (vectors)
+FMLALB_zzzw 01100100 10 1 ..... 10 0 00 0 ..... ..... @rda_rn_rm_e0
+FMLALT_zzzw 01100100 10 1 ..... 10 0 00 1 ..... ..... @rda_rn_rm_e0
+FMLSLB_zzzw 01100100 10 1 ..... 10 1 00 0 ..... ..... @rda_rn_rm_e0
+FMLSLT_zzzw 01100100 10 1 ..... 10 1 00 1 ..... ..... @rda_rn_rm_e0
+
+### SVE2 floating-point multiply-add long (indexed)
+FMLALB_zzxw 01100100 10 1 ..... 0100.0 ..... ..... @rrxr_3a esz=2
+FMLALT_zzxw 01100100 10 1 ..... 0100.1 ..... ..... @rrxr_3a esz=2
+FMLSLB_zzxw 01100100 10 1 ..... 0110.0 ..... ..... @rrxr_3a esz=2
+FMLSLT_zzxw 01100100 10 1 ..... 0110.1 ..... ..... @rrxr_3a esz=2
diff --git a/target/arm/translate-sve.c b/target/arm/translate-sve.c
index 26377b1227..ae628968da 100644
--- a/target/arm/translate-sve.c
+++ b/target/arm/translate-sve.c
@@ -8468,3 +8468,78 @@ static bool trans_FLOGB(DisasContext *s, arg_rpr_esz *a)
}
return true;
}
+
+static bool do_FMLAL_zzzw(DisasContext *s, arg_rrrr_esz *a, bool sub, bool sel)
+{
+ if (!dc_isar_feature(aa64_sve2, s)) {
+ return false;
+ }
+ if (sve_access_check(s)) {
+ unsigned vsz = vec_full_reg_size(s);
+ tcg_gen_gvec_4_ptr(vec_full_reg_offset(s, a->rd),
+ vec_full_reg_offset(s, a->rn),
+ vec_full_reg_offset(s, a->rm),
+ vec_full_reg_offset(s, a->ra),
+ cpu_env, vsz, vsz, (sel << 1) | sub,
+ gen_helper_sve2_fmlal_zzzw_s);
+ }
+ return true;
+}
+
+static bool trans_FMLALB_zzzw(DisasContext *s, arg_rrrr_esz *a)
+{
+ return do_FMLAL_zzzw(s, a, false, false);
+}
+
+static bool trans_FMLALT_zzzw(DisasContext *s, arg_rrrr_esz *a)
+{
+ return do_FMLAL_zzzw(s, a, false, true);
+}
+
+static bool trans_FMLSLB_zzzw(DisasContext *s, arg_rrrr_esz *a)
+{
+ return do_FMLAL_zzzw(s, a, true, false);
+}
+
+static bool trans_FMLSLT_zzzw(DisasContext *s, arg_rrrr_esz *a)
+{
+ return do_FMLAL_zzzw(s, a, true, true);
+}
+
+static bool do_FMLAL_zzxw(DisasContext *s, arg_rrxr_esz *a, bool sub, bool sel)
+{
+ if (!dc_isar_feature(aa64_sve2, s)) {
+ return false;
+ }
+ if (sve_access_check(s)) {
+ unsigned vsz = vec_full_reg_size(s);
+ tcg_gen_gvec_4_ptr(vec_full_reg_offset(s, a->rd),
+ vec_full_reg_offset(s, a->rn),
+ vec_full_reg_offset(s, a->rm),
+ vec_full_reg_offset(s, a->ra),
+ cpu_env, vsz, vsz,
+ (a->index << 2) | (sel << 1) | sub,
+ gen_helper_sve2_fmlal_zzxw_s);
+ }
+ return true;
+}
+
+static bool trans_FMLALB_zzxw(DisasContext *s, arg_rrxr_esz *a)
+{
+ return do_FMLAL_zzxw(s, a, false, false);
+}
+
+static bool trans_FMLALT_zzxw(DisasContext *s, arg_rrxr_esz *a)
+{
+ return do_FMLAL_zzxw(s, a, false, true);
+}
+
+static bool trans_FMLSLB_zzxw(DisasContext *s, arg_rrxr_esz *a)
+{
+ return do_FMLAL_zzxw(s, a, true, false);
+}
+
+static bool trans_FMLSLT_zzxw(DisasContext *s, arg_rrxr_esz *a)
+{
+ return do_FMLAL_zzxw(s, a, true, true);
+}
diff --git a/target/arm/vec_helper.c b/target/arm/vec_helper.c
index 9b2a4d5b7e..ea08b15c55 100644
--- a/target/arm/vec_helper.c
+++ b/target/arm/vec_helper.c
@@ -29,10 +29,14 @@
so addressing units smaller than that needs a host-endian fixup. */
#ifdef HOST_WORDS_BIGENDIAN
#define H1(x) ((x) ^ 7)
+#define H1_2(x) ((x) ^ 6)
+#define H1_4(x) ((x) ^ 4)
#define H2(x) ((x) ^ 3)
#define H4(x) ((x) ^ 1)
#else
#define H1(x) (x)
+#define H1_2(x) (x)
+#define H1_4(x) (x)
#define H2(x) (x)
#define H4(x) (x)
#endif
@@ -1907,6 +1911,27 @@ void HELPER(gvec_fmlal_a64)(void *vd, void *vn, void *vm,
get_flush_inputs_to_zero(&env->vfp.fp_status_f16));
}
+void HELPER(sve2_fmlal_zzzw_s)(void *vd, void *vn, void *vm, void *va,
+ void *venv, uint32_t desc)
+{
+ intptr_t i, oprsz = simd_oprsz(desc);
+ uint16_t negn = extract32(desc, SIMD_DATA_SHIFT, 1) << 15;
+ intptr_t sel = extract32(desc, SIMD_DATA_SHIFT + 1, 1) * sizeof(float16);
+ CPUARMState *env = venv;
+ float_status *status = &env->vfp.fp_status;
+ bool fz16 = get_flush_inputs_to_zero(&env->vfp.fp_status_f16);
+
+ for (i = 0; i < oprsz; i += sizeof(float32)) {
+ float16 nn_16 = *(float16 *)(vn + H1_2(i + sel)) ^ negn;
+ float16 mm_16 = *(float16 *)(vm + H1_2(i + sel));
+ float32 nn = float16_to_float32_by_bits(nn_16, fz16);
+ float32 mm = float16_to_float32_by_bits(mm_16, fz16);
+ float32 aa = *(float32 *)(va + H1_4(i));
+
+ *(float32 *)(vd + H1_4(i)) = float32_muladd(nn, mm, aa, 0, status);
+ }
+}
+
static void do_fmlal_idx(float32 *d, void *vn, void *vm, float_status *fpst,
uint32_t desc, bool fz16)
{
@@ -1951,6 +1976,32 @@ void HELPER(gvec_fmlal_idx_a64)(void *vd, void *vn, void
*vm,
get_flush_inputs_to_zero(&env->vfp.fp_status_f16));
}
+void HELPER(sve2_fmlal_zzxw_s)(void *vd, void *vn, void *vm, void *va,
+ void *venv, uint32_t desc)
+{
+ intptr_t i, j, oprsz = simd_oprsz(desc);
+ uint16_t negn = extract32(desc, SIMD_DATA_SHIFT, 1) << 15;
+ intptr_t sel = extract32(desc, SIMD_DATA_SHIFT + 1, 1) * sizeof(float16);
+ intptr_t idx = extract32(desc, SIMD_DATA_SHIFT + 2, 3) * sizeof(float16);
+ CPUARMState *env = venv;
+ float_status *status = &env->vfp.fp_status;
+ bool fz16 = get_flush_inputs_to_zero(&env->vfp.fp_status_f16);
+
+ for (i = 0; i < oprsz; i += 16) {
+ float16 mm_16 = *(float16 *)(vm + i + idx);
+ float32 mm = float16_to_float32_by_bits(mm_16, fz16);
+
+ for (j = 0; j < 16; j += sizeof(float32)) {
+ float16 nn_16 = *(float16 *)(vn + H1_2(i + j + sel)) ^ negn;
+ float32 nn = float16_to_float32_by_bits(nn_16, fz16);
+ float32 aa = *(float32 *)(va + H1_4(i + j));
+
+ *(float32 *)(vd + H1_4(i + j)) =
+ float32_muladd(nn, mm, aa, 0, status);
+ }
+ }
+}
+
void HELPER(gvec_sshl_b)(void *vd, void *vn, void *vm, uint32_t desc)
{
intptr_t i, opr_sz = simd_oprsz(desc);
--
2.25.1
- [PATCH v5 62/81] target/arm: Implement SVE2 crypto destructive binary operations, (continued)
- [PATCH v5 62/81] target/arm: Implement SVE2 crypto destructive binary operations, Richard Henderson, 2021/04/16
- [PATCH v5 66/81] target/arm: Implement SVE2 FCVTLT, Richard Henderson, 2021/04/16
- [PATCH v5 53/81] target/arm: Implement SVE2 integer multiply-add (indexed), Richard Henderson, 2021/04/16
- [PATCH v5 59/81] target/arm: Implement SVE mixed sign dot product (indexed), Richard Henderson, 2021/04/16
- [PATCH v5 65/81] target/arm: Implement SVE2 FCVTNT, Richard Henderson, 2021/04/16
- [PATCH v5 67/81] target/arm: Implement SVE2 FCVTXNT, FCVTX, Richard Henderson, 2021/04/16
- [PATCH v5 63/81] target/arm: Implement SVE2 crypto constructive binary operations, Richard Henderson, 2021/04/16
- [PATCH v5 70/81] target/arm: Implement SVE2 LD1RO, Richard Henderson, 2021/04/16
- [PATCH v5 71/81] target/arm: Implement 128-bit ZIP, UZP, TRN, Richard Henderson, 2021/04/16
- [PATCH v5 72/81] target/arm: Implement SVE2 bitwise shift immediate, Richard Henderson, 2021/04/16
- [PATCH v5 73/81] target/arm: Implement SVE2 fp multiply-add long,
Richard Henderson <=
- [PATCH v5 75/81] target/arm: Split out do_neon_ddda_fpst, Richard Henderson, 2021/04/16
- [PATCH v5 74/81] target/arm: Implement aarch64 SUDOT, USDOT, Richard Henderson, 2021/04/16
- [PATCH v5 78/81] target/arm: Split decode of VSDOT and VUDOT, Richard Henderson, 2021/04/16
- [PATCH v5 76/81] target/arm: Remove unused fpst from VDOT_scalar, Richard Henderson, 2021/04/16
- [PATCH v5 77/81] target/arm: Fix decode for VDOT (indexed), Richard Henderson, 2021/04/16
- [PATCH v5 79/81] target/arm: Implement aarch32 VSUDOT, VUSDOT, Richard Henderson, 2021/04/16
- [PATCH v5 80/81] target/arm: Implement integer matrix multiply accumulate, Richard Henderson, 2021/04/16
- [PATCH v5 81/81] target/arm: Enable SVE2 and some extensions, Richard Henderson, 2021/04/16
- Re: [PATCH v5 for-6.1 00/81] target/arm: Implement SVE2, no-reply, 2021/04/16