[v2,15/67] target/arm: Implement SVE Integer Multiply-Add Group

Message ID 20180217182323.25885-16-richard.henderson@linaro.org
State Superseded
Headers show
Series
  • target/arm: Scalable Vector Extension
Related show

Commit Message

Richard Henderson Feb. 17, 2018, 6:22 p.m.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

---
 target/arm/helper-sve.h    | 18 ++++++++++++++
 target/arm/sve_helper.c    | 58 +++++++++++++++++++++++++++++++++++++++++++++-
 target/arm/translate-sve.c | 31 +++++++++++++++++++++++++
 target/arm/sve.decode      | 17 ++++++++++++++
 4 files changed, 123 insertions(+), 1 deletion(-)

-- 
2.14.3

Comments

Peter Maydell Feb. 23, 2018, 1:12 p.m. | #1
On 17 February 2018 at 18:22, Richard Henderson
<richard.henderson@linaro.org> wrote:
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

> ---

>  target/arm/helper-sve.h    | 18 ++++++++++++++

>  target/arm/sve_helper.c    | 58 +++++++++++++++++++++++++++++++++++++++++++++-

>  target/arm/translate-sve.c | 31 +++++++++++++++++++++++++

>  target/arm/sve.decode      | 17 ++++++++++++++

>  4 files changed, 123 insertions(+), 1 deletion(-)

>

> diff --git a/target/arm/helper-sve.h b/target/arm/helper-sve.h

> index 11644125d1..b31d497f31 100644

> --- a/target/arm/helper-sve.h

> +++ b/target/arm/helper-sve.h

> @@ -345,6 +345,24 @@ DEF_HELPER_FLAGS_4(sve_neg_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)

>  DEF_HELPER_FLAGS_4(sve_neg_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)

>  DEF_HELPER_FLAGS_4(sve_neg_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)

>

> +DEF_HELPER_FLAGS_6(sve_mla_b, TCG_CALL_NO_RWG,

> +                   void, ptr, ptr, ptr, ptr, ptr, i32)

> +DEF_HELPER_FLAGS_6(sve_mla_h, TCG_CALL_NO_RWG,

> +                   void, ptr, ptr, ptr, ptr, ptr, i32)

> +DEF_HELPER_FLAGS_6(sve_mla_s, TCG_CALL_NO_RWG,

> +                   void, ptr, ptr, ptr, ptr, ptr, i32)

> +DEF_HELPER_FLAGS_6(sve_mla_d, TCG_CALL_NO_RWG,

> +                   void, ptr, ptr, ptr, ptr, ptr, i32)

> +

> +DEF_HELPER_FLAGS_6(sve_mls_b, TCG_CALL_NO_RWG,

> +                   void, ptr, ptr, ptr, ptr, ptr, i32)

> +DEF_HELPER_FLAGS_6(sve_mls_h, TCG_CALL_NO_RWG,

> +                   void, ptr, ptr, ptr, ptr, ptr, i32)

> +DEF_HELPER_FLAGS_6(sve_mls_s, TCG_CALL_NO_RWG,

> +                   void, ptr, ptr, ptr, ptr, ptr, i32)

> +DEF_HELPER_FLAGS_6(sve_mls_d, TCG_CALL_NO_RWG,

> +                   void, ptr, ptr, ptr, ptr, ptr, i32)

> +

>  DEF_HELPER_FLAGS_5(sve_and_pppp, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32)

>  DEF_HELPER_FLAGS_5(sve_bic_pppp, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32)

>  DEF_HELPER_FLAGS_5(sve_eor_pppp, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32)

> diff --git a/target/arm/sve_helper.c b/target/arm/sve_helper.c

> index e11823a727..4b08a38ce8 100644

> --- a/target/arm/sve_helper.c

> +++ b/target/arm/sve_helper.c

> @@ -932,6 +932,62 @@ DO_ZPZI_D(sve_asrd_d, int64_t, DO_ASRD)

>  #undef DO_SHR

>  #undef DO_SHL

>  #undef DO_ASRD

> -

>  #undef DO_ZPZI

>  #undef DO_ZPZI_D


Deletion of blank line should be dropped or squashed into earlier patch.

Otherwise

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>


thanks
-- PMM

Patch

diff --git a/target/arm/helper-sve.h b/target/arm/helper-sve.h
index 11644125d1..b31d497f31 100644
--- a/target/arm/helper-sve.h
+++ b/target/arm/helper-sve.h
@@ -345,6 +345,24 @@  DEF_HELPER_FLAGS_4(sve_neg_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
 DEF_HELPER_FLAGS_4(sve_neg_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
 DEF_HELPER_FLAGS_4(sve_neg_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
 
+DEF_HELPER_FLAGS_6(sve_mla_b, TCG_CALL_NO_RWG,
+                   void, ptr, ptr, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_6(sve_mla_h, TCG_CALL_NO_RWG,
+                   void, ptr, ptr, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_6(sve_mla_s, TCG_CALL_NO_RWG,
+                   void, ptr, ptr, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_6(sve_mla_d, TCG_CALL_NO_RWG,
+                   void, ptr, ptr, ptr, ptr, ptr, i32)
+
+DEF_HELPER_FLAGS_6(sve_mls_b, TCG_CALL_NO_RWG,
+                   void, ptr, ptr, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_6(sve_mls_h, TCG_CALL_NO_RWG,
+                   void, ptr, ptr, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_6(sve_mls_s, TCG_CALL_NO_RWG,
+                   void, ptr, ptr, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_6(sve_mls_d, TCG_CALL_NO_RWG,
+                   void, ptr, ptr, ptr, ptr, ptr, i32)
+
 DEF_HELPER_FLAGS_5(sve_and_pppp, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32)
 DEF_HELPER_FLAGS_5(sve_bic_pppp, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32)
 DEF_HELPER_FLAGS_5(sve_eor_pppp, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32)
diff --git a/target/arm/sve_helper.c b/target/arm/sve_helper.c
index e11823a727..4b08a38ce8 100644
--- a/target/arm/sve_helper.c
+++ b/target/arm/sve_helper.c
@@ -932,6 +932,62 @@  DO_ZPZI_D(sve_asrd_d, int64_t, DO_ASRD)
 #undef DO_SHR
 #undef DO_SHL
 #undef DO_ASRD
-
 #undef DO_ZPZI
 #undef DO_ZPZI_D
+
+/* Fully general four-operand expander, controlled by a predicate.
+ */
+#define DO_ZPZZZ(NAME, TYPE, H, OP)                           \
+void HELPER(NAME)(void *vd, void *va, void *vn, void *vm,     \
+                  void *vg, uint32_t desc)                    \
+{                                                             \
+    intptr_t i, opr_sz = simd_oprsz(desc);                    \
+    for (i = 0; i < opr_sz; ) {                               \
+        uint16_t pg = *(uint16_t *)(vg + H1_2(i >> 3));       \
+        do {                                                  \
+            if (pg & 1) {                                     \
+                TYPE nn = *(TYPE *)(vn + H(i));               \
+                TYPE mm = *(TYPE *)(vm + H(i));               \
+                TYPE aa = *(TYPE *)(va + H(i));               \
+                *(TYPE *)(vd + H(i)) = OP(aa, nn, mm);        \
+            }                                                 \
+            i += sizeof(TYPE), pg >>= sizeof(TYPE);           \
+        } while (i & 15);                                     \
+    }                                                         \
+}
+
+/* Similarly, specialized for 64-bit operands.  */
+#define DO_ZPZZZ_D(NAME, TYPE, OP)                            \
+void HELPER(NAME)(void *vd, void *va, void *vn, void *vm,     \
+                  void *vg, uint32_t desc)                    \
+{                                                             \
+    intptr_t i, opr_sz = simd_oprsz(desc) / 8;                \
+    TYPE *d = vd, *a = va, *n = vn, *m = vm;                  \
+    uint8_t *pg = vg;                                         \
+    for (i = 0; i < opr_sz; i += 1) {                         \
+        if (pg[H1(i)] & 1) {                                  \
+            TYPE aa = a[i], nn = n[i], mm = m[i];             \
+            d[i] = OP(aa, nn, mm);                            \
+        }                                                     \
+    }                                                         \
+}
+
+#define DO_MLA(A, N, M)  (A + N * M)
+#define DO_MLS(A, N, M)  (A - N * M)
+
+DO_ZPZZZ(sve_mla_b, uint8_t, H1, DO_MLA)
+DO_ZPZZZ(sve_mls_b, uint8_t, H1, DO_MLS)
+
+DO_ZPZZZ(sve_mla_h, uint16_t, H1_2, DO_MLA)
+DO_ZPZZZ(sve_mls_h, uint16_t, H1_2, DO_MLS)
+
+DO_ZPZZZ(sve_mla_s, uint32_t, H1_4, DO_MLA)
+DO_ZPZZZ(sve_mls_s, uint32_t, H1_4, DO_MLS)
+
+DO_ZPZZZ_D(sve_mla_d, uint64_t, DO_MLA)
+DO_ZPZZZ_D(sve_mls_d, uint64_t, DO_MLS)
+
+#undef DO_MLA
+#undef DO_MLS
+#undef DO_ZPZZZ
+#undef DO_ZPZZZ_D
diff --git a/target/arm/translate-sve.c b/target/arm/translate-sve.c
index dce8ba8dc0..b956d87636 100644
--- a/target/arm/translate-sve.c
+++ b/target/arm/translate-sve.c
@@ -609,6 +609,37 @@  DO_ZPZW(LSL, lsl)
 
 #undef DO_ZPZW
 
+/*
+ *** SVE Integer Multiply-Add Group
+ */
+
+static void do_zpzzz_ool(DisasContext *s, arg_rprrr_esz *a,
+                         gen_helper_gvec_5 *fn)
+{
+    unsigned vsz = vec_full_reg_size(s);
+    tcg_gen_gvec_5_ool(vec_full_reg_offset(s, a->rd),
+                       vec_full_reg_offset(s, a->ra),
+                       vec_full_reg_offset(s, a->rn),
+                       vec_full_reg_offset(s, a->rm),
+                       pred_full_reg_offset(s, a->pg),
+                       vsz, vsz, 0, fn);
+}
+
+#define DO_ZPZZZ(NAME, name) \
+static void trans_##NAME(DisasContext *s, arg_rprrr_esz *a, uint32_t insn) \
+{                                                                    \
+    static gen_helper_gvec_5 * const fns[4] = {                      \
+        gen_helper_sve_##name##_b, gen_helper_sve_##name##_h,        \
+        gen_helper_sve_##name##_s, gen_helper_sve_##name##_d,        \
+    };                                                               \
+    do_zpzzz_ool(s, a, fns[a->esz]);                                 \
+}
+
+DO_ZPZZZ(MLA, mla)
+DO_ZPZZZ(MLS, mls)
+
+#undef DO_ZPZZZ
+
 /*
  *** SVE Predicate Logical Operations Group
  */
diff --git a/target/arm/sve.decode b/target/arm/sve.decode
index b875501475..68a1823b72 100644
--- a/target/arm/sve.decode
+++ b/target/arm/sve.decode
@@ -48,6 +48,7 @@ 
 &rpr_esz	rd pg rn esz
 &rprr_s		rd pg rn rm s
 &rprr_esz	rd pg rn rm esz
+&rprrr_esz	rd pg rn rm ra esz
 &rpri_esz	rd pg rn imm esz
 &ptrue		rd esz pat s
 
@@ -73,6 +74,12 @@ 
 @rdm_pg_rn	........ esz:2 ... ... ... pg:3 rn:5 rd:5 \
 		&rprr_esz rm=%reg_movprfx
 
+# Three register operand, with governing predicate, vector element size
+@rda_pg_rn_rm	........ esz:2 . rm:5  ... pg:3 rn:5 rd:5 \
+		&rprrr_esz ra=%reg_movprfx
+@rdn_pg_ra_rm	........ esz:2 . rm:5  ... pg:3 ra:5 rd:5 \
+		&rprrr_esz rn=%reg_movprfx
+
 # One register operand, with governing predicate, vector element size
 @rd_pg_rn	........ esz:2 ... ... ... pg:3 rn:5 rd:5	&rpr_esz
 
@@ -188,6 +195,16 @@  UXTH		00000100 .. 010 011 101 ... ..... .....		@rd_pg_rn
 SXTW		00000100 .. 010 100 101 ... ..... .....		@rd_pg_rn
 UXTW		00000100 .. 010 101 101 ... ..... .....		@rd_pg_rn
 
+### SVE Integer Multiply-Add Group
+
+# SVE integer multiply-add writing addend (predicated)
+MLA		00000100 .. 0 ..... 010 ... ..... .....   @rda_pg_rn_rm
+MLS		00000100 .. 0 ..... 011 ... ..... .....   @rda_pg_rn_rm
+
+# SVE integer multiply-add writing multiplicand (predicated)
+MLA		00000100 .. 0 ..... 110 ... ..... .....   @rdn_pg_ra_rm # MAD
+MLS		00000100 .. 0 ..... 111 ... ..... .....   @rdn_pg_ra_rm # MSB
+
 ### SVE Logical - Unpredicated Group
 
 # SVE bitwise logical operations (unpredicated)