target/arm: Implement SVE2 FMMLA

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Stephen Long <steplong@quicinc.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210525010358.152808-47-richard.henderson@linaro.org
Message-Id: <20200422165503.13511-1-steplong@quicinc.com>
[rth: Fix indexing in helpers, expand macro to straight functions.]
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

5 files changed, 125 insertions, 0 deletions

diff --git a/target/arm/cpu.h b/target/arm/cpu.h
index 39919a7..7ad79ff 100644
--- a/target/arm/cpu.h
+++ b/target/arm/cpu.h
@@ -4256,6 +4256,16 @@ static inline bool isar_feature_aa64_sve2_bitperm(const ARMISARegisters *id)
     return FIELD_EX64(id->id_aa64zfr0, ID_AA64ZFR0, BITPERM) != 0;
 }
 
+static inline bool isar_feature_aa64_sve_f32mm(const ARMISARegisters *id)
+{
+    return FIELD_EX64(id->id_aa64zfr0, ID_AA64ZFR0, F32MM) != 0;
+}
+
+static inline bool isar_feature_aa64_sve_f64mm(const ARMISARegisters *id)
+{
+    return FIELD_EX64(id->id_aa64zfr0, ID_AA64ZFR0, F64MM) != 0;
+}
+
 /*
  * Feature tests for "does this exist in either 32-bit or 64-bit?"
  */
diff --git a/target/arm/helper-sve.h b/target/arm/helper-sve.h
index 28b8f00..7e99dcd 100644
--- a/target/arm/helper-sve.h
+++ b/target/arm/helper-sve.h
@@ -2662,3 +2662,6 @@ DEF_HELPER_FLAGS_5(sve2_sqrdcmlah_zzzz_s, TCG_CALL_NO_RWG,
                    void, ptr, ptr, ptr, ptr, i32)
 DEF_HELPER_FLAGS_5(sve2_sqrdcmlah_zzzz_d, TCG_CALL_NO_RWG,
                    void, ptr, ptr, ptr, ptr, i32)
+
+DEF_HELPER_FLAGS_6(fmmla_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_6(fmmla_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, ptr, i32)
diff --git a/target/arm/sve.decode b/target/arm/sve.decode
index c3958be..cb2ee86 100644
--- a/target/arm/sve.decode
+++ b/target/arm/sve.decode
@@ -1389,6 +1389,10 @@ UMLSLT_zzzw     01000100 .. 0 ..... 010 111 ..... .....  @rda_rn_rm
 CMLA_zzzz       01000100 esz:2 0 rm:5 0010 rot:2 rn:5 rd:5  ra=%reg_movprfx
 SQRDCMLAH_zzzz  01000100 esz:2 0 rm:5 0011 rot:2 rn:5 rd:5  ra=%reg_movprfx
 
+### SVE2 floating point matrix multiply accumulate
+
+FMMLA           01100100 .. 1 ..... 111001 ..... .....  @rda_rn_rm
+
 ### SVE2 Memory Gather Load Group
 
 # SVE2 64-bit gather non-temporal load
diff --git a/target/arm/sve_helper.c b/target/arm/sve_helper.c
index 5b62929..fa96e28 100644
--- a/target/arm/sve_helper.c
+++ b/target/arm/sve_helper.c
@@ -7241,3 +7241,77 @@ void HELPER(sve2_xar_s)(void *vd, void *vn, void *vm, uint32_t desc)
         d[i] = ror32(n[i] ^ m[i], shr);
     }
 }
+
+void HELPER(fmmla_s)(void *vd, void *vn, void *vm, void *va,
+                     void *status, uint32_t desc)
+{
+    intptr_t s, opr_sz = simd_oprsz(desc) / (sizeof(float32) * 4);
+
+    for (s = 0; s < opr_sz; ++s) {
+        float32 *n = vn + s * sizeof(float32) * 4;
+        float32 *m = vm + s * sizeof(float32) * 4;
+        float32 *a = va + s * sizeof(float32) * 4;
+        float32 *d = vd + s * sizeof(float32) * 4;
+        float32 n00 = n[H4(0)], n01 = n[H4(1)];
+        float32 n10 = n[H4(2)], n11 = n[H4(3)];
+        float32 m00 = m[H4(0)], m01 = m[H4(1)];
+        float32 m10 = m[H4(2)], m11 = m[H4(3)];
+        float32 p0, p1;
+
+        /* i = 0, j = 0 */
+        p0 = float32_mul(n00, m00, status);
+        p1 = float32_mul(n01, m01, status);
+        d[H4(0)] = float32_add(a[H4(0)], float32_add(p0, p1, status), status);
+
+        /* i = 0, j = 1 */
+        p0 = float32_mul(n00, m10, status);
+        p1 = float32_mul(n01, m11, status);
+        d[H4(1)] = float32_add(a[H4(1)], float32_add(p0, p1, status), status);
+
+        /* i = 1, j = 0 */
+        p0 = float32_mul(n10, m00, status);
+        p1 = float32_mul(n11, m01, status);
+        d[H4(2)] = float32_add(a[H4(2)], float32_add(p0, p1, status), status);
+
+        /* i = 1, j = 1 */
+        p0 = float32_mul(n10, m10, status);
+        p1 = float32_mul(n11, m11, status);
+        d[H4(3)] = float32_add(a[H4(3)], float32_add(p0, p1, status), status);
+    }
+}
+
+void HELPER(fmmla_d)(void *vd, void *vn, void *vm, void *va,
+                     void *status, uint32_t desc)
+{
+    intptr_t s, opr_sz = simd_oprsz(desc) / (sizeof(float64) * 4);
+
+    for (s = 0; s < opr_sz; ++s) {
+        float64 *n = vn + s * sizeof(float64) * 4;
+        float64 *m = vm + s * sizeof(float64) * 4;
+        float64 *a = va + s * sizeof(float64) * 4;
+        float64 *d = vd + s * sizeof(float64) * 4;
+        float64 n00 = n[0], n01 = n[1], n10 = n[2], n11 = n[3];
+        float64 m00 = m[0], m01 = m[1], m10 = m[2], m11 = m[3];
+        float64 p0, p1;
+
+        /* i = 0, j = 0 */
+        p0 = float64_mul(n00, m00, status);
+        p1 = float64_mul(n01, m01, status);
+        d[0] = float64_add(a[0], float64_add(p0, p1, status), status);
+
+        /* i = 0, j = 1 */
+        p0 = float64_mul(n00, m10, status);
+        p1 = float64_mul(n01, m11, status);
+        d[1] = float64_add(a[1], float64_add(p0, p1, status), status);
+
+        /* i = 1, j = 0 */
+        p0 = float64_mul(n10, m00, status);
+        p1 = float64_mul(n11, m01, status);
+        d[2] = float64_add(a[2], float64_add(p0, p1, status), status);
+
+        /* i = 1, j = 1 */
+        p0 = float64_mul(n10, m10, status);
+        p1 = float64_mul(n11, m11, status);
+        d[3] = float64_add(a[3], float64_add(p0, p1, status), status);
+    }
+}
diff --git a/target/arm/translate-sve.c b/target/arm/translate-sve.c
index a64ad04..a94b399 100644
--- a/target/arm/translate-sve.c
+++ b/target/arm/translate-sve.c
@@ -7672,6 +7672,40 @@ DO_SVE2_ZPZZ_FP(FMINP, fminp)
  * SVE Integer Multiply-Add (unpredicated)
  */
 
+static bool trans_FMMLA(DisasContext *s, arg_rrrr_esz *a)
+{
+    gen_helper_gvec_4_ptr *fn;
+
+    switch (a->esz) {
+    case MO_32:
+        if (!dc_isar_feature(aa64_sve_f32mm, s)) {
+            return false;
+        }
+        fn = gen_helper_fmmla_s;
+        break;
+    case MO_64:
+        if (!dc_isar_feature(aa64_sve_f64mm, s)) {
+            return false;
+        }
+        fn = gen_helper_fmmla_d;
+        break;
+    default:
+        return false;
+    }
+
+    if (sve_access_check(s)) {
+        unsigned vsz = vec_full_reg_size(s);
+        TCGv_ptr status = fpstatus_ptr(FPST_FPCR);
+        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, 0, fn);
+        tcg_temp_free_ptr(status);
+    }
+    return true;
+}
+
 static bool do_sqdmlal_zzzw(DisasContext *s, arg_rrrr_esz *a,
                             bool sel1, bool sel2)
 {