[PULL 16/21] target/hexagon: introduce new helper functions

[Taylor Simpson]

From: Niccolò Izzo <nizzo@rev.ng>

These helpers will be employed by the idef-parser generated code, to
correctly implement instruction semantics. "Helper" functions, in the
context of this patch, refers to functions which provide a manual TCG
implementation of certain features.

Signed-off-by: Alessandro Di Federico <ale@rev.ng>
Signed-off-by: Niccolò Izzo <nizzo@rev.ng>
Signed-off-by: Anton Johansson <anjo@rev.ng>
Signed-off-by: Taylor Simpson <tsimpson@quicinc.com>
Reviewed-by: Taylor Simpson <tsimpson@quicinc.com>
Message-Id: <20220923173831.227551-6-anjo@rev.ng>
---
 target/hexagon/genptr.h |  13 ++++
 target/hexagon/macros.h |   9 +++
 target/hexagon/genptr.c | 146 +++++++++++++++++++++++++++++++++++++++-
 3 files changed, 166 insertions(+), 2 deletions(-)

diff --git a/target/hexagon/genptr.h b/target/hexagon/genptr.h
index e126c46255..591b059698 100644
--- a/target/hexagon/genptr.h
+++ b/target/hexagon/genptr.h
@@ -33,9 +33,22 @@ void gen_store1i(TCGv_env cpu_env, TCGv vaddr, int32_t src, 
uint32_t slot);
 void gen_store2i(TCGv_env cpu_env, TCGv vaddr, int32_t src, uint32_t slot);
 void gen_store4i(TCGv_env cpu_env, TCGv vaddr, int32_t src, uint32_t slot);
 void gen_store8i(TCGv_env cpu_env, TCGv vaddr, int64_t src, uint32_t slot);
+TCGv gen_read_reg(TCGv result, int num);
 TCGv gen_read_preg(TCGv pred, uint8_t num);
 void gen_log_reg_write(int rnum, TCGv val);
 void gen_log_pred_write(DisasContext *ctx, int pnum, TCGv val);
+void gen_set_usr_field(int field, TCGv val);
+void gen_set_usr_fieldi(int field, int x);
+void gen_set_usr_field_if(int field, TCGv val);
+void gen_sat_i32(TCGv dest, TCGv source, int width);
+void gen_sat_i32_ovfl(TCGv ovfl, TCGv dest, TCGv source, int width);
+void gen_satu_i32(TCGv dest, TCGv source, int width);
+void gen_satu_i32_ovfl(TCGv ovfl, TCGv dest, TCGv source, int width);
+void gen_sat_i64(TCGv_i64 dest, TCGv_i64 source, int width);
+void gen_sat_i64_ovfl(TCGv ovfl, TCGv_i64 dest, TCGv_i64 source, int width);
+void gen_satu_i64(TCGv_i64 dest, TCGv_i64 source, int width);
+void gen_satu_i64_ovfl(TCGv ovfl, TCGv_i64 dest, TCGv_i64 source, int width);
+void gen_add_sat_i64(TCGv_i64 ret, TCGv_i64 a, TCGv_i64 b);
 TCGv gen_8bitsof(TCGv result, TCGv value);
 void gen_set_byte_i64(int N, TCGv_i64 result, TCGv src);
 TCGv gen_get_byte(TCGv result, int N, TCGv src, bool sign);
diff --git a/target/hexagon/macros.h b/target/hexagon/macros.h
index 1a31542010..cd64bb8eec 100644
--- a/target/hexagon/macros.h
+++ b/target/hexagon/macros.h
@@ -197,7 +197,16 @@
 #define MEM_STORE8(VA, DATA, SLOT) log_store64(env, VA, DATA, 8, SLOT)
 #endif
 
+#ifdef QEMU_GENERATE
+static inline void gen_cancel(uint32_t slot)
+{
+    tcg_gen_ori_tl(hex_slot_cancelled, hex_slot_cancelled, 1 << slot);
+}
+
+#define CANCEL gen_cancel(slot);
+#else
 #define CANCEL cancel_slot(env, slot)
+#endif
 
 #define LOAD_CANCEL(EA) do { CANCEL; } while (0)
 
diff --git a/target/hexagon/genptr.c b/target/hexagon/genptr.c
index 3fe279ce8c..6cf2e0ed43 100644
--- a/target/hexagon/genptr.c
+++ b/target/hexagon/genptr.c
@@ -31,6 +31,12 @@
 #include "gen_tcg_hvx.h"
 #include "genptr.h"
 
+TCGv gen_read_reg(TCGv result, int num)
+{
+    tcg_gen_mov_tl(result, hex_gpr[num]);
+    return result;
+}
+
 TCGv gen_read_preg(TCGv pred, uint8_t num)
 {
     tcg_gen_mov_tl(pred, hex_pred[num]);
@@ -506,7 +512,7 @@ static void gen_write_new_pc_pcrel(DisasContext *ctx, int 
pc_off,
     }
 }
 
-static void gen_set_usr_field(int field, TCGv val)
+void gen_set_usr_field(int field, TCGv val)
 {
     tcg_gen_deposit_tl(hex_new_value[HEX_REG_USR], hex_new_value[HEX_REG_USR],
                        val,
@@ -514,7 +520,7 @@ static void gen_set_usr_field(int field, TCGv val)
                        reg_field_info[field].width);
 }
 
-static void gen_set_usr_fieldi(int field, int x)
+void gen_set_usr_fieldi(int field, int x)
 {
     if (reg_field_info[field].width == 1) {
         target_ulong bit = 1 << reg_field_info[field].offset;
@@ -1028,5 +1034,141 @@ void probe_noshuf_load(TCGv va, int s, int mi)
     gen_helper_probe_noshuf_load(cpu_env, va, size, mem_idx);
 }
 
+/*
+ * Note: Since this function might branch, `val` is
+ * required to be a `tcg_temp_local`.
+ */
+void gen_set_usr_field_if(int field, TCGv val)
+{
+    /* Sets the USR field if `val` is non-zero */
+    if (reg_field_info[field].width == 1) {
+        TCGv tmp = tcg_temp_new();
+        tcg_gen_extract_tl(tmp, val, 0, reg_field_info[field].width);
+        tcg_gen_shli_tl(tmp, tmp, reg_field_info[field].offset);
+        tcg_gen_or_tl(hex_new_value[HEX_REG_USR],
+                      hex_new_value[HEX_REG_USR],
+                      tmp);
+        tcg_temp_free(tmp);
+    } else {
+        TCGLabel *skip_label = gen_new_label();
+        tcg_gen_brcondi_tl(TCG_COND_EQ, val, 0, skip_label);
+        gen_set_usr_field(field, val);
+        gen_set_label(skip_label);
+    }
+}
+
+void gen_sat_i32(TCGv dest, TCGv source, int width)
+{
+    TCGv max_val = tcg_constant_tl((1 << (width - 1)) - 1);
+    TCGv min_val = tcg_constant_tl(-(1 << (width - 1)));
+    tcg_gen_smin_tl(dest, source, max_val);
+    tcg_gen_smax_tl(dest, dest, min_val);
+}
+
+void gen_sat_i32_ovfl(TCGv ovfl, TCGv dest, TCGv source, int width)
+{
+    gen_sat_i32(dest, source, width);
+    tcg_gen_setcond_tl(TCG_COND_NE, ovfl, source, dest);
+}
+
+void gen_satu_i32(TCGv dest, TCGv source, int width)
+{
+    TCGv max_val = tcg_constant_tl((1 << width) - 1);
+    TCGv zero = tcg_constant_tl(0);
+    tcg_gen_movcond_tl(TCG_COND_GTU, dest, source, max_val, max_val, source);
+    tcg_gen_movcond_tl(TCG_COND_LT, dest, source, zero, zero, dest);
+}
+
+void gen_satu_i32_ovfl(TCGv ovfl, TCGv dest, TCGv source, int width)
+{
+    gen_satu_i32(dest, source, width);
+    tcg_gen_setcond_tl(TCG_COND_NE, ovfl, source, dest);
+}
+
+void gen_sat_i64(TCGv_i64 dest, TCGv_i64 source, int width)
+{
+    TCGv_i64 max_val = tcg_constant_i64((1LL << (width - 1)) - 1LL);
+    TCGv_i64 min_val = tcg_constant_i64(-(1LL << (width - 1)));
+    tcg_gen_smin_i64(dest, source, max_val);
+    tcg_gen_smax_i64(dest, dest, min_val);
+}
+
+void gen_sat_i64_ovfl(TCGv ovfl, TCGv_i64 dest, TCGv_i64 source, int width)
+{
+    TCGv_i64 ovfl_64;
+    gen_sat_i64(dest, source, width);
+    ovfl_64 = tcg_temp_new_i64();
+    tcg_gen_setcond_i64(TCG_COND_NE, ovfl_64, dest, source);
+    tcg_gen_trunc_i64_tl(ovfl, ovfl_64);
+    tcg_temp_free_i64(ovfl_64);
+}
+
+void gen_satu_i64(TCGv_i64 dest, TCGv_i64 source, int width)
+{
+    TCGv_i64 max_val = tcg_constant_i64((1LL << width) - 1LL);
+    TCGv_i64 zero = tcg_constant_i64(0);
+    tcg_gen_movcond_i64(TCG_COND_GTU, dest, source, max_val, max_val, source);
+    tcg_gen_movcond_i64(TCG_COND_LT, dest, source, zero, zero, dest);
+}
+
+void gen_satu_i64_ovfl(TCGv ovfl, TCGv_i64 dest, TCGv_i64 source, int width)
+{
+    TCGv_i64 ovfl_64;
+    gen_satu_i64(dest, source, width);
+    ovfl_64 = tcg_temp_new_i64();
+    tcg_gen_setcond_i64(TCG_COND_NE, ovfl_64, dest, source);
+    tcg_gen_trunc_i64_tl(ovfl, ovfl_64);
+    tcg_temp_free_i64(ovfl_64);
+}
+
+/* Implements the fADDSAT64 macro in TCG */
+void gen_add_sat_i64(TCGv_i64 ret, TCGv_i64 a, TCGv_i64 b)
+{
+    TCGv_i64 sum = tcg_temp_local_new_i64();
+    TCGv_i64 xor = tcg_temp_new_i64();
+    TCGv_i64 cond1 = tcg_temp_new_i64();
+    TCGv_i64 cond2 = tcg_temp_local_new_i64();
+    TCGv_i64 cond3 = tcg_temp_new_i64();
+    TCGv_i64 mask = tcg_constant_i64(0x8000000000000000ULL);
+    TCGv_i64 max_pos = tcg_constant_i64(0x7FFFFFFFFFFFFFFFLL);
+    TCGv_i64 max_neg = tcg_constant_i64(0x8000000000000000LL);
+    TCGv_i64 zero = tcg_constant_i64(0);
+    TCGLabel *no_ovfl_label = gen_new_label();
+    TCGLabel *ovfl_label = gen_new_label();
+    TCGLabel *ret_label = gen_new_label();
+
+    tcg_gen_add_i64(sum, a, b);
+    tcg_gen_xor_i64(xor, a, b);
+
+    /* if (xor & mask) */
+    tcg_gen_and_i64(cond1, xor, mask);
+    tcg_temp_free_i64(xor);
+    tcg_gen_brcondi_i64(TCG_COND_NE, cond1, 0, no_ovfl_label);
+    tcg_temp_free_i64(cond1);
+
+    /* else if ((a ^ sum) & mask) */
+    tcg_gen_xor_i64(cond2, a, sum);
+    tcg_gen_and_i64(cond2, cond2, mask);
+    tcg_gen_brcondi_i64(TCG_COND_NE, cond2, 0, ovfl_label);
+    tcg_temp_free_i64(cond2);
+    /* fallthrough to no_ovfl_label branch */
+
+    /* if branch */
+    gen_set_label(no_ovfl_label);
+    tcg_gen_mov_i64(ret, sum);
+    tcg_gen_br(ret_label);
+
+    /* else if branch */
+    gen_set_label(ovfl_label);
+    tcg_gen_and_i64(cond3, sum, mask);
+    tcg_temp_free_i64(mask);
+    tcg_temp_free_i64(sum);
+    tcg_gen_movcond_i64(TCG_COND_NE, ret, cond3, zero, max_pos, max_neg);
+    tcg_temp_free_i64(cond3);
+    SET_USR_FIELD(USR_OVF, 1);
+
+    gen_set_label(ret_label);
+}
+
 #include "tcg_funcs_generated.c.inc"
 #include "tcg_func_table_generated.c.inc"
-- 
2.17.1