[Qemu-devel] [PATCH v2 38/67] target/arm: Implement SVE Partition Break Group

[Richard Henderson <richard.henderson@linaro.org>]

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/arm/helper-sve.h    |  18 ++++
 target/arm/sve_helper.c    | 247 +++++++++++++++++++++++++++++++++++++++++++++
 target/arm/translate-sve.c |  96 ++++++++++++++++++
 target/arm/sve.decode      |  19 ++++
 4 files changed, 380 insertions(+)

diff --git a/target/arm/helper-sve.h b/target/arm/helper-sve.h
index ae38c0a4be..f0a3ed3414 100644
--- a/target/arm/helper-sve.h
+++ b/target/arm/helper-sve.h
@@ -658,3 +658,21 @@ DEF_HELPER_FLAGS_5(sve_orn_pppp, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32)
 DEF_HELPER_FLAGS_5(sve_nor_pppp, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32)
 DEF_HELPER_FLAGS_5(sve_nand_pppp, TCG_CALL_NO_RWG,
                    void, ptr, ptr, ptr, ptr, i32)
+
+DEF_HELPER_FLAGS_5(sve_brkpa, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_5(sve_brkpb, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_5(sve_brkpas, TCG_CALL_NO_RWG, i32, ptr, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_5(sve_brkpbs, TCG_CALL_NO_RWG, i32, ptr, ptr, ptr, ptr, i32)
+
+DEF_HELPER_FLAGS_4(sve_brka_z, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(sve_brkb_z, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(sve_brka_m, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(sve_brkb_m, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+
+DEF_HELPER_FLAGS_4(sve_brkas_z, TCG_CALL_NO_RWG, i32, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(sve_brkbs_z, TCG_CALL_NO_RWG, i32, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(sve_brkas_m, TCG_CALL_NO_RWG, i32, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(sve_brkbs_m, TCG_CALL_NO_RWG, i32, ptr, ptr, ptr, i32)
+
+DEF_HELPER_FLAGS_4(sve_brkn, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(sve_brkns, TCG_CALL_NO_RWG, i32, ptr, ptr, ptr, i32)
diff --git a/target/arm/sve_helper.c b/target/arm/sve_helper.c
index b74db681f2..d6d2220f8b 100644
--- a/target/arm/sve_helper.c
+++ b/target/arm/sve_helper.c
@@ -2455,3 +2455,250 @@ DO_CMP_PPZI_D(sve_cmpls_ppzi_d, uint64_t, <=)
 #undef DO_CMP_PPZI_S
 #undef DO_CMP_PPZI_D
 #undef DO_CMP_PPZI
+
+/* Similar to the ARM LastActive pseudocode function.  */
+static bool last_active_pred(void *vd, void *vg, intptr_t oprsz)
+{
+    intptr_t i;
+
+    for (i = QEMU_ALIGN_UP(oprsz, 8) - 8; i >= 0; i -= 8) {
+        uint64_t pg = *(uint64_t *)(vg + i);
+        if (pg) {
+            return (pow2floor(pg) & *(uint64_t *)(vd + i)) != 0;
+        }
+    }
+    return 0;
+}
+
+/* Compute a mask into RETB that is true for all G, up to and including
+ * (if after) or excluding (if !after) the first G & N.
+ * Return true if BRK found.
+ */
+static bool compute_brk(uint64_t *retb, uint64_t n, uint64_t g,
+                        bool brk, bool after)
+{
+    uint64_t b;
+
+    if (brk) {
+        b = 0;
+    } else if ((g & n) == 0) {
+        /* For all G, no N are set; break not found.  */
+        b = g;
+    } else {
+        /* Break somewhere in N.  Locate it.  */
+        b = g & n;            /* guard true, pred true*/
+        b = b & -b;           /* first such */
+        if (after) {
+            b = b | (b - 1);  /* break after same */
+        } else {
+            b = b - 1;        /* break before same */
+        }
+        brk = true;
+    }
+
+    *retb = b;
+    return brk;
+}
+
+/* Compute a zeroing BRK.  */
+static void compute_brk_z(uint64_t *d, uint64_t *n, uint64_t *g,
+                          intptr_t oprsz, bool after)
+{
+    bool brk = false;
+    intptr_t i;
+
+    for (i = 0; i < DIV_ROUND_UP(oprsz, 8); ++i) {
+        uint64_t this_b, this_g = g[i];
+
+        brk = compute_brk(&this_b, n[i], this_g, brk, after);
+        d[i] = this_b & this_g;
+    }
+}
+
+/* Likewise, but also compute flags.  */
+static uint32_t compute_brks_z(uint64_t *d, uint64_t *n, uint64_t *g,
+                               intptr_t oprsz, bool after)
+{
+    uint32_t flags = PREDTEST_INIT;
+    bool brk = false;
+    intptr_t i;
+
+    for (i = 0; i < DIV_ROUND_UP(oprsz, 8); ++i) {
+        uint64_t this_b, this_d, this_g = g[i];
+
+        brk = compute_brk(&this_b, n[i], this_g, brk, after);
+        d[i] = this_d = this_b & this_g;
+        flags = iter_predtest_fwd(this_d, this_g, flags);
+    }
+    return flags;
+}
+
+/* Given a computation function, compute a merging BRK.  */
+static void compute_brk_m(uint64_t *d, uint64_t *n, uint64_t *g,
+                          intptr_t oprsz, bool after)
+{
+    bool brk = false;
+    intptr_t i;
+
+    for (i = 0; i < DIV_ROUND_UP(oprsz, 8); ++i) {
+        uint64_t this_b, this_g = g[i];
+
+        brk = compute_brk(&this_b, n[i], this_g, brk, after);
+        d[i] = (this_b & this_g) | (d[i] & ~this_g);
+    }
+}
+
+/* Likewise, but also compute flags.  */
+static uint32_t compute_brks_m(uint64_t *d, uint64_t *n, uint64_t *g,
+                               intptr_t oprsz, bool after)
+{
+    uint32_t flags = PREDTEST_INIT;
+    bool brk = false;
+    intptr_t i;
+
+    for (i = 0; i < oprsz / 8; ++i) {
+        uint64_t this_b, this_d = d[i], this_g = g[i];
+
+        brk = compute_brk(&this_b, n[i], this_g, brk, after);
+        d[i] = this_d = (this_b & this_g) | (this_d & ~this_g);
+        flags = iter_predtest_fwd(this_d, this_g, flags);
+    }
+    return flags;
+}
+
+static uint32_t do_zero(ARMPredicateReg *d, intptr_t oprsz)
+{
+    /* It is quicker to zero the whole predicate than loop on OPRSZ.
+       The compiler should turn this into 4 64-bit integer stores.  */
+    memset(d, 0, sizeof(ARMPredicateReg));
+    return PREDTEST_INIT;
+}
+
+void HELPER(sve_brkpa)(void *vd, void *vn, void *vm, void *vg,
+                       uint32_t pred_desc)
+{
+    intptr_t oprsz = extract32(pred_desc, 0, SIMD_OPRSZ_BITS) + 2;
+    if (last_active_pred(vn, vg, oprsz)) {
+        compute_brk_z(vd, vm, vg, oprsz, true);
+    } else {
+        do_zero(vd, oprsz);
+    }
+}
+
+uint32_t HELPER(sve_brkpas)(void *vd, void *vn, void *vm, void *vg,
+                            uint32_t pred_desc)
+{
+    intptr_t oprsz = extract32(pred_desc, 0, SIMD_OPRSZ_BITS) + 2;
+    if (last_active_pred(vn, vg, oprsz)) {
+        return compute_brks_z(vd, vm, vg, oprsz, true);
+    } else {
+        return do_zero(vd, oprsz);
+    }
+}
+
+void HELPER(sve_brkpb)(void *vd, void *vn, void *vm, void *vg,
+                       uint32_t pred_desc)
+{
+    intptr_t oprsz = extract32(pred_desc, 0, SIMD_OPRSZ_BITS) + 2;
+    if (last_active_pred(vn, vg, oprsz)) {
+        compute_brk_z(vd, vm, vg, oprsz, false);
+    } else {
+        do_zero(vd, oprsz);
+    }
+}
+
+uint32_t HELPER(sve_brkpbs)(void *vd, void *vn, void *vm, void *vg,
+                            uint32_t pred_desc)
+{
+    intptr_t oprsz = extract32(pred_desc, 0, SIMD_OPRSZ_BITS) + 2;
+    if (last_active_pred(vn, vg, oprsz)) {
+        return compute_brks_z(vd, vm, vg, oprsz, false);
+    } else {
+        return do_zero(vd, oprsz);
+    }
+}
+
+void HELPER(sve_brka_z)(void *vd, void *vn, void *vg, uint32_t pred_desc)
+{
+    intptr_t oprsz = extract32(pred_desc, 0, SIMD_OPRSZ_BITS) + 2;
+    compute_brk_z(vd, vn, vg, oprsz, true);
+}
+
+uint32_t HELPER(sve_brkas_z)(void *vd, void *vn, void *vg, uint32_t pred_desc)
+{
+    intptr_t oprsz = extract32(pred_desc, 0, SIMD_OPRSZ_BITS) + 2;
+    return compute_brks_z(vd, vn, vg, oprsz, true);
+}
+
+void HELPER(sve_brkb_z)(void *vd, void *vn, void *vg, uint32_t pred_desc)
+{
+    intptr_t oprsz = extract32(pred_desc, 0, SIMD_OPRSZ_BITS) + 2;
+    compute_brk_z(vd, vn, vg, oprsz, false);
+}
+
+uint32_t HELPER(sve_brkbs_z)(void *vd, void *vn, void *vg, uint32_t pred_desc)
+{
+    intptr_t oprsz = extract32(pred_desc, 0, SIMD_OPRSZ_BITS) + 2;
+    return compute_brks_z(vd, vn, vg, oprsz, false);
+}
+
+void HELPER(sve_brka_m)(void *vd, void *vn, void *vg, uint32_t pred_desc)
+{
+    intptr_t oprsz = extract32(pred_desc, 0, SIMD_OPRSZ_BITS) + 2;
+    compute_brk_m(vd, vn, vg, oprsz, true);
+}
+
+uint32_t HELPER(sve_brkas_m)(void *vd, void *vn, void *vg, uint32_t pred_desc)
+{
+    intptr_t oprsz = extract32(pred_desc, 0, SIMD_OPRSZ_BITS) + 2;
+    return compute_brks_m(vd, vn, vg, oprsz, true);
+}
+
+void HELPER(sve_brkb_m)(void *vd, void *vn, void *vg, uint32_t pred_desc)
+{
+    intptr_t oprsz = extract32(pred_desc, 0, SIMD_OPRSZ_BITS) + 2;
+    compute_brk_m(vd, vn, vg, oprsz, false);
+}
+
+uint32_t HELPER(sve_brkbs_m)(void *vd, void *vn, void *vg, uint32_t pred_desc)
+{
+    intptr_t oprsz = extract32(pred_desc, 0, SIMD_OPRSZ_BITS) + 2;
+    return compute_brks_m(vd, vn, vg, oprsz, false);
+}
+
+void HELPER(sve_brkn)(void *vd, void *vn, void *vg, uint32_t pred_desc)
+{
+    intptr_t oprsz = extract32(pred_desc, 0, SIMD_OPRSZ_BITS) + 2;
+
+    if (!last_active_pred(vn, vg, oprsz)) {
+        do_zero(vd, oprsz);
+    }
+}
+
+/* As if PredTest(Ones(PL), D, esz).  */
+static uint32_t predtest_ones(ARMPredicateReg *d, intptr_t oprsz,
+                              uint64_t esz_mask)
+{
+    uint32_t flags = PREDTEST_INIT;
+    intptr_t i;
+
+    for (i = 0; i < oprsz / 8; i++) {
+        flags = iter_predtest_fwd(d->p[i], esz_mask, flags);
+    }
+    if (oprsz & 7) {
+        uint64_t mask = ~(-1ULL << (8 * (oprsz & 7)));
+        flags = iter_predtest_fwd(d->p[i], esz_mask & mask, flags);
+    }
+    return flags;
+}
+
+uint32_t HELPER(sve_brkns)(void *vd, void *vn, void *vg, uint32_t pred_desc)
+{
+    intptr_t oprsz = extract32(pred_desc, 0, SIMD_OPRSZ_BITS) + 2;
+
+    if (last_active_pred(vn, vg, oprsz)) {
+        return predtest_ones(vd, oprsz, -1);
+    } else {
+        return do_zero(vd, oprsz);
+    }
+}
diff --git a/target/arm/translate-sve.c b/target/arm/translate-sve.c
index a7eeb122e3..dc95d68867 100644
--- a/target/arm/translate-sve.c
+++ b/target/arm/translate-sve.c
@@ -2635,6 +2635,102 @@ DO_PPZI(CMPLS, cmpls)
 
 #undef DO_PPZI
 
+/*
+ *** SVE Partition Break Group
+ */
+
+static void do_brk3(DisasContext *s, arg_rprr_s *a,
+                    gen_helper_gvec_4 *fn, gen_helper_gvec_flags_4 *fn_s)
+{
+    unsigned vsz = pred_full_reg_size(s);
+
+    /* Predicate sizes may be smaller and cannot use simd_desc.  */
+    TCGv_ptr d = tcg_temp_new_ptr();
+    TCGv_ptr n = tcg_temp_new_ptr();
+    TCGv_ptr m = tcg_temp_new_ptr();
+    TCGv_ptr g = tcg_temp_new_ptr();
+    TCGv_i32 t = tcg_const_i32(vsz - 2);
+
+    tcg_gen_addi_ptr(d, cpu_env, pred_full_reg_offset(s, a->rd));
+    tcg_gen_addi_ptr(n, cpu_env, pred_full_reg_offset(s, a->rn));
+    tcg_gen_addi_ptr(m, cpu_env, pred_full_reg_offset(s, a->rm));
+    tcg_gen_addi_ptr(g, cpu_env, pred_full_reg_offset(s, a->pg));
+
+    if (a->s) {
+        fn_s(t, d, n, m, g, t);
+        do_pred_flags(t);
+    } else {
+        fn(d, n, m, g, t);
+    }
+    tcg_temp_free_ptr(d);
+    tcg_temp_free_ptr(n);
+    tcg_temp_free_ptr(m);
+    tcg_temp_free_ptr(g);
+    tcg_temp_free_i32(t);
+}
+
+static void do_brk2(DisasContext *s, arg_rpr_s *a,
+                    gen_helper_gvec_3 *fn, gen_helper_gvec_flags_3 *fn_s)
+{
+    unsigned vsz = pred_full_reg_size(s);
+
+    /* Predicate sizes may be smaller and cannot use simd_desc.  */
+    TCGv_ptr d = tcg_temp_new_ptr();
+    TCGv_ptr n = tcg_temp_new_ptr();
+    TCGv_ptr g = tcg_temp_new_ptr();
+    TCGv_i32 t = tcg_const_i32(vsz - 2);
+
+    tcg_gen_addi_ptr(d, cpu_env, pred_full_reg_offset(s, a->rd));
+    tcg_gen_addi_ptr(n, cpu_env, pred_full_reg_offset(s, a->rn));
+    tcg_gen_addi_ptr(g, cpu_env, pred_full_reg_offset(s, a->pg));
+
+    if (a->s) {
+        fn_s(t, d, n, g, t);
+        do_pred_flags(t);
+    } else {
+        fn(d, n, g, t);
+    }
+    tcg_temp_free_ptr(d);
+    tcg_temp_free_ptr(n);
+    tcg_temp_free_ptr(g);
+    tcg_temp_free_i32(t);
+}
+
+void trans_BRKPA(DisasContext *s, arg_rprr_s *a, uint32_t insn)
+{
+    do_brk3(s, a, gen_helper_sve_brkpa, gen_helper_sve_brkpas);
+}
+
+void trans_BRKPB(DisasContext *s, arg_rprr_s *a, uint32_t insn)
+{
+    do_brk3(s, a, gen_helper_sve_brkpb, gen_helper_sve_brkpbs);
+}
+
+void trans_BRKA_m(DisasContext *s, arg_rpr_s *a, uint32_t insn)
+{
+    do_brk2(s, a, gen_helper_sve_brka_m, gen_helper_sve_brkas_m);
+}
+
+void trans_BRKB_m(DisasContext *s, arg_rpr_s *a, uint32_t insn)
+{
+    do_brk2(s, a, gen_helper_sve_brkb_m, gen_helper_sve_brkbs_m);
+}
+
+void trans_BRKA_z(DisasContext *s, arg_rpr_s *a, uint32_t insn)
+{
+    do_brk2(s, a, gen_helper_sve_brka_z, gen_helper_sve_brkas_z);
+}
+
+void trans_BRKB_z(DisasContext *s, arg_rpr_s *a, uint32_t insn)
+{
+    do_brk2(s, a, gen_helper_sve_brkb_z, gen_helper_sve_brkbs_z);
+}
+
+void trans_BRKN(DisasContext *s, arg_rpr_s *a, uint32_t insn)
+{
+    do_brk2(s, a, gen_helper_sve_brkn, gen_helper_sve_brkns);
+}
+
 /*
  *** SVE Memory - 32-bit Gather and Unsized Contiguous Group
  */
diff --git a/target/arm/sve.decode b/target/arm/sve.decode
index 0e317d7d48..1c19129e55 100644
--- a/target/arm/sve.decode
+++ b/target/arm/sve.decode
@@ -60,6 +60,7 @@
 &rri_esz	rd rn imm esz
 &rrr_esz	rd rn rm esz
 &rpr_esz	rd pg rn esz
+&rpr_s		rd pg rn s
 &rprr_s		rd pg rn rm s
 &rprr_esz	rd pg rn rm esz
 &rprrr_esz	rd pg rn rm ra esz
@@ -79,6 +80,9 @@
 @pd_pn		........ esz:2 .. .... ....... rn:4 . rd:4	&rr_esz
 @rd_rn		........ esz:2 ...... ...... rn:5 rd:5		&rr_esz
 
+# Two operand with governing predicate, flags setting
+@pd_pg_pn_s	........ . s:1 ...... .. pg:4 . rn:4 . rd:4	&rpr_s
+
 # Three operand with unused vector element size
 @rd_rn_rm_e0	........ ... rm:5 ... ... rn:5 rd:5		&rrr_esz esz=0
 
@@ -568,6 +572,21 @@ PFIRST		00100101 01 011 000 11000 00 .... 0 ....	@pd_pn_e0
 # SVE predicate next active
 PNEXT		00100101 .. 011 001 11000 10 .... 0 ....	@pd_pn
 
+### SVE Partition Break Group
+
+# SVE propagate break from previous partition
+BRKPA		00100101 0. 00 .... 11 .... 0 .... 0 ....	@pd_pg_pn_pm_s
+BRKPB		00100101 0. 00 .... 11 .... 0 .... 1 ....	@pd_pg_pn_pm_s
+
+# SVE partition break condition
+BRKA_z		00100101 0. 01000001 .... 0 .... 0 ....		@pd_pg_pn_s
+BRKB_z		00100101 1. 01000001 .... 0 .... 0 ....		@pd_pg_pn_s
+BRKA_m		00100101 0. 01000001 .... 0 .... 1 ....		@pd_pg_pn_s
+BRKB_m		00100101 1. 01000001 .... 0 .... 1 ....		@pd_pg_pn_s
+
+# SVE propagate break to next partition
+BRKN		00100101 0. 01100001 .... 0 .... 0 ....		@pd_pg_pn_s
+
 ### SVE Memory - 32-bit Gather and Unsized Contiguous Group
 
 # SVE load predicate register
-- 
2.14.3