target/arm: Implement FMOP4 (widening, 2-way fp16 to fp32)

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20260702204314.79224-7-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

diff --git a/target/arm/tcg/helper-sme-defs.h b/target/arm/tcg/helper-sme-defs.h
index 0e4bb12b38d008d22fa729080325d7d5e8b66ece..780b8f162a7f70abf4f0b2cf7a3b034cb2b68677 100644 (file)
--- a/target/arm/tcg/helper-sme-defs.h
+++ b/target/arm/tcg/helper-sme-defs.h
@@ -375,3 +375,7 @@ DEF_HELPER_FLAGS_5(sme_ah_bfmop4s_hh, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst
 DEF_HELPER_FLAGS_5(sme_bfmop4a_sh, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, env, i32)
 DEF_HELPER_FLAGS_5(sme_bfmop4s_sh, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, env, i32)
 DEF_HELPER_FLAGS_5(sme_ah_bfmop4s_sh, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, env, i32)
+
+DEF_HELPER_FLAGS_5(sme_fmop4a_sh, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, env, i32)
+DEF_HELPER_FLAGS_5(sme_fmop4s_sh, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, env, i32)
+DEF_HELPER_FLAGS_5(sme_ah_fmop4s_sh, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, env, i32)

diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode
index 12a3a6143276b1dfb46f7aa4e665d4d11a9e85b9..ea9f459e38cb1972d1fc01bdb5ac3c9d91b18dbb 100644 (file)
--- a/target/arm/tcg/sme.decode
+++ b/target/arm/tcg/sme.decode
@@ -1107,3 +1107,4 @@ FMOP4_ss        1000 0000 000. ...0 0000 00.. ..0. 00..         @mop4_o2
 FMOP4_dd        1000 0000 110. ...0 0000 00.. ..0. 1...         @mop4_o3
 
 BFMOP4_sh       1000 0001 000. ...0 0000 00.. ..0. 00..         @mop4_o2
+FMOP4_sh        1000 0001 001. ...0 0000 00.. ..0. 00..         @mop4_o2

diff --git a/target/arm/tcg/sme_helper.c b/target/arm/tcg/sme_helper.c
index 5a661ca616ef06a20b1d29237422da06336ceb22..123f4593421bc9505495b013fc8a0550ea38882a 100644 (file)
--- a/target/arm/tcg/sme_helper.c
+++ b/target/arm/tcg/sme_helper.c
@@ -2887,3 +2887,54 @@ void HELPER(sme_ah_bfmop4s_sh)(void *vza, void *vzn, void *vzm,
              is_ebf(env, &fpst) ? inner_ebf_ah_bfmop4s_sh
                                 : inner_ah_bfmop4s_sh);
 }
+
+static void inner_fmop4a_sh(void *vd, void *vn, void *vm, void *vinfo)
+{
+    float32 *d = vd;
+    uint32_t *n = vn, *m = vm;
+    CPUArchState *env = vinfo;
+
+    *d = f16_dotadd(*d, *n, *m,
+                    &env->vfp.fp_status[FPST_ZA_F16],
+                    &env->vfp.fp_status[FPST_ZA]);
+}
+
+void HELPER(sme_fmop4a_sh)(void *vza, void *vzn, void *vzm,
+                           CPUArchState *env, uint32_t desc)
+{
+    sme_mop4(vza, vzn, vzm, env, desc, sizeof(float32), inner_fmop4a_sh);
+}
+
+static void inner_fmop4s_sh(void *vd, void *vn, void *vm, void *vinfo)
+{
+    float32 *d = vd;
+    uint32_t *n = vn, *m = vm;
+    CPUArchState *env = vinfo;
+
+    *d = f16_dotadd(*d, *n ^ 0x80008000u, *m,
+                    &env->vfp.fp_status[FPST_ZA_F16],
+                    &env->vfp.fp_status[FPST_ZA]);
+}
+
+void HELPER(sme_fmop4s_sh)(void *vza, void *vzn, void *vzm,
+                           CPUArchState *env, uint32_t desc)
+{
+    sme_mop4(vza, vzn, vzm, env, desc, sizeof(float32), inner_fmop4s_sh);
+}
+
+static void inner_ah_fmop4s_sh(void *vd, void *vn, void *vm, void *vinfo)
+{
+    float32 *d = vd;
+    uint32_t *n = vn, *m = vm;
+    CPUArchState *env = vinfo;
+
+    *d = f16_dotadd(*d, f16mop_ah_neg_adj_pair(*n, -1), *m,
+                    &env->vfp.fp_status[FPST_ZA_F16],
+                    &env->vfp.fp_status[FPST_ZA]);
+}
+
+void HELPER(sme_ah_fmop4s_sh)(void *vza, void *vzn, void *vzm,
+                              CPUArchState *env, uint32_t desc)
+{
+    sme_mop4(vza, vzn, vzm, env, desc, sizeof(float32), inner_ah_fmop4s_sh);
+}

diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c
index df4a5c212530c18c16c06902db9588cc7c507f15..544a825595afd14c70804e65f31daf1722bbe57c 100644 (file)
--- a/target/arm/tcg/translate-sme.c
+++ b/target/arm/tcg/translate-sme.c
@@ -2065,3 +2065,10 @@ static gen_helper_gvec_3_ptr * const bfmop4_sh[3] = {
     gen_helper_sme_ah_bfmop4s_sh
 };
 TRANS_FEAT(BFMOP4_sh, aa64_sme_mop4, do_mop4_fp, a, MO_32, FPST_ENV, bfmop4_sh)
+
+static gen_helper_gvec_3_ptr * const fmop4_sh[3] = {
+    gen_helper_sme_fmop4a_sh,
+    gen_helper_sme_fmop4s_sh,
+    gen_helper_sme_ah_fmop4s_sh
+};
+TRANS_FEAT(FMOP4_sh, aa64_sme_mop4, do_mop4_fp, a, MO_32, FPST_ENV, fmop4_sh)