target/arm: Implement bfloat widening fma (vector)

This is BFMLAL{B,T} for both AArch64 AdvSIMD and SVE,
and VFMA{B,T}.BF16 for AArch32 NEON.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210525225817.400336-10-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

7 files changed, 73 insertions, 4 deletions

diff --git a/target/arm/helper.h b/target/arm/helper.h
index af75d7f25f..36b3c9dd2d 100644
--- a/target/arm/helper.h
+++ b/target/arm/helper.h
@@ -1010,6 +1010,9 @@ DEF_HELPER_FLAGS_5(gvec_bfdot_idx, TCG_CALL_NO_RWG,
 DEF_HELPER_FLAGS_5(gvec_bfmmla, TCG_CALL_NO_RWG,
                    void, ptr, ptr, ptr, ptr, i32)
 
+DEF_HELPER_FLAGS_6(gvec_bfmlal, TCG_CALL_NO_RWG,
+                   void, ptr, ptr, ptr, ptr, ptr, i32)
+
 #ifdef TARGET_AARCH64
 #include "helper-a64.h"
 #include "helper-sve.h"
diff --git a/target/arm/neon-shared.decode b/target/arm/neon-shared.decode
index 4e0a25d27c..b61addd98b 100644
--- a/target/arm/neon-shared.decode
+++ b/target/arm/neon-shared.decode
@@ -70,6 +70,9 @@ VUSMMLA        1111 1100 1.10 .... .... 1100 .1.0 .... \
 VMMLA_b16      1111 1100 0.00 .... .... 1100 .1.0 .... \
                vm=%vm_dp vn=%vn_dp vd=%vd_dp
 
+VFMA_b16       1111 110 0 0.11 .... .... 1000 . q:1 . 1 .... \
+               vm=%vm_dp vn=%vn_dp vd=%vd_dp
+
 VCMLA_scalar   1111 1110 0 . rot:2 .... .... 1000 . q:1 index:1 0 vm:4 \
                vn=%vn_dp vd=%vd_dp size=1
 VCMLA_scalar   1111 1110 1 . rot:2 .... .... 1000 . q:1 . 0 .... \
diff --git a/target/arm/sve.decode b/target/arm/sve.decode
index 6c17898dee..5281164eea 100644
--- a/target/arm/sve.decode
+++ b/target/arm/sve.decode
@@ -1627,6 +1627,9 @@ 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
 
+BFMLALB_zzzw    01100100 11 1 ..... 10 0 00 0 ..... .....  @rda_rn_rm_e0
+BFMLALT_zzzw    01100100 11 1 ..... 10 0 00 1 ..... .....  @rda_rn_rm_e0
+
 ### SVE2 floating-point bfloat16 dot-product
 BFDOT_zzzz      01100100 01 1 ..... 10 0 00 0 ..... .....  @rda_rn_rm_e0
 
diff --git a/target/arm/translate-a64.c b/target/arm/translate-a64.c
index 9ce2f5a7d4..8dcb15ac0f 100644
--- a/target/arm/translate-a64.c
+++ b/target/arm/translate-a64.c
@@ -12242,9 +12242,10 @@ static void disas_simd_three_reg_same_extra(DisasContext *s, uint32_t insn)
         }
         feature = dc_isar_feature(aa64_bf16, s);
         break;
-    case 0x1f: /* BFDOT */
+    case 0x1f:
         switch (size) {
-        case 1:
+        case 1: /* BFDOT */
+        case 3: /* BFMLAL{B,T} */
             feature = dc_isar_feature(aa64_bf16, s);
             break;
         default:
@@ -12338,11 +12339,15 @@ static void disas_simd_three_reg_same_extra(DisasContext *s, uint32_t insn)
     case 0xd: /* BFMMLA */
         gen_gvec_op4_ool(s, is_q, rd, rn, rm, rd, 0, gen_helper_gvec_bfmmla);
         return;
-    case 0xf: /* BFDOT */
+    case 0xf:
         switch (size) {
-        case 1:
+        case 1: /* BFDOT */
             gen_gvec_op4_ool(s, is_q, rd, rn, rm, rd, 0, gen_helper_gvec_bfdot);
             break;
+        case 3: /* BFMLAL{B,T} */
+            gen_gvec_op4_fpst(s, 1, rd, rn, rm, rd, false, is_q,
+                              gen_helper_gvec_bfmlal);
+            break;
         default:
             g_assert_not_reached();
         }
diff --git a/target/arm/translate-neon.c b/target/arm/translate-neon.c
index 9d227a1e13..4d0c2494dc 100644
--- a/target/arm/translate-neon.c
+++ b/target/arm/translate-neon.c
@@ -4135,3 +4135,12 @@ static bool trans_VMMLA_b16(DisasContext *s, arg_VMMLA_b16 *a)
     return do_neon_ddda(s, 7, a->vd, a->vn, a->vm, 0,
                         gen_helper_gvec_bfmmla);
 }
+
+static bool trans_VFMA_b16(DisasContext *s, arg_VFMA_b16 *a)
+{
+    if (!dc_isar_feature(aa32_bf16, s)) {
+        return false;
+    }
+    return do_neon_ddda_fpst(s, 7, a->vd, a->vn, a->vm, a->q, FPST_STD,
+                             gen_helper_gvec_bfmlal);
+}
diff --git a/target/arm/translate-sve.c b/target/arm/translate-sve.c
index 4f575dc334..ba8f5d7b7d 100644
--- a/target/arm/translate-sve.c
+++ b/target/arm/translate-sve.c
@@ -8689,3 +8689,33 @@ static bool trans_BFMMLA(DisasContext *s, arg_rrrr_esz *a)
     }
     return true;
 }
+
+static bool do_BFMLAL_zzzw(DisasContext *s, arg_rrrr_esz *a, bool sel)
+{
+    if (!dc_isar_feature(aa64_sve_bf16, s)) {
+        return false;
+    }
+    if (sve_access_check(s)) {
+        TCGv_ptr status = fpstatus_ptr(FPST_FPCR);
+        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),
+                           status, vsz, vsz, sel,
+                           gen_helper_gvec_bfmlal);
+        tcg_temp_free_ptr(status);
+    }
+    return true;
+}
+
+static bool trans_BFMLALB_zzzw(DisasContext *s, arg_rrrr_esz *a)
+{
+    return do_BFMLAL_zzzw(s, a, false);
+}
+
+static bool trans_BFMLALT_zzzw(DisasContext *s, arg_rrrr_esz *a)
+{
+    return do_BFMLAL_zzzw(s, a, true);
+}
diff --git a/target/arm/vec_helper.c b/target/arm/vec_helper.c
index 27e9bdd329..d82736b5e6 100644
--- a/target/arm/vec_helper.c
+++ b/target/arm/vec_helper.c
@@ -2512,3 +2512,19 @@ void HELPER(gvec_bfmmla)(void *vd, void *vn, void *vm, void *va, uint32_t desc)
     }
     clear_tail(d, opr_sz, simd_maxsz(desc));
 }
+
+void HELPER(gvec_bfmlal)(void *vd, void *vn, void *vm, void *va,
+                         void *stat, uint32_t desc)
+{
+    intptr_t i, opr_sz = simd_oprsz(desc);
+    intptr_t sel = simd_data(desc);
+    float32 *d = vd, *a = va;
+    bfloat16 *n = vn, *m = vm;
+
+    for (i = 0; i < opr_sz / 4; ++i) {
+        float32 nn = n[H2(i * 2 + sel)] << 16;
+        float32 mm = m[H2(i * 2 + sel)] << 16;
+        d[H4(i)] = float32_muladd(nn, mm, a[H4(i)], 0, stat);
+    }
+    clear_tail(d, opr_sz, simd_maxsz(desc));
+}