[PULL 01/15] Hexagon (target/hexagon) Analyze reads before writes

Brian Cain posted 15 patches 6 months, 3 weeks ago
Maintainers: Brian Cain <bcain@quicinc.com>, Alessandro Di Federico <ale@rev.ng>, Anton Johansson <anjo@rev.ng>
[PULL 01/15] Hexagon (target/hexagon) Analyze reads before writes
Posted by Brian Cain 6 months, 3 weeks ago
From: Taylor Simpson <ltaylorsimpson@gmail.com>

We divide gen_analyze_funcs.py into 3 phases
    Declare the operands
    Analyze the register reads
    Analyze the register writes

We also create special versions of ctx_log_*_read for new operands
    Check that the operand is written before the read

This is a precursor to improving the analysis for short-circuiting
the packet semantics in a subsequent commit

Signed-off-by: Taylor Simpson <ltaylorsimpson@gmail.com>
Reviewed-by: Brian Cain <bcain@quicinc.com>
Message-Id: <20240201103340.119081-2-ltaylorsimpson@gmail.com>
Signed-off-by: Brian Cain <bcain@quicinc.com>
---
 target/hexagon/README               |  9 +++--
 target/hexagon/gen_analyze_funcs.py | 34 ++++++++++------
 target/hexagon/hex_common.py        | 63 +++++++++++++++--------------
 target/hexagon/translate.h          | 26 +++++++++++-
 4 files changed, 83 insertions(+), 49 deletions(-)

diff --git a/target/hexagon/README b/target/hexagon/README
index 746ebec378..c1d8c8d0ab 100644
--- a/target/hexagon/README
+++ b/target/hexagon/README
@@ -183,10 +183,11 @@ when the override is present.
     }
 
 We also generate an analyze_<tag> function for each instruction.  Currently,
-these functions record the writes to registers by calling ctx_log_*.  During
-gen_start_packet, we invoke the analyze_<tag> function for each instruction in
-the packet, and we mark the implicit writes.  After the analysis is performed,
-we initialize the result register for each of the predicated assignments.
+these functions record the reads and writes to registers by calling ctx_log_*.
+During gen_start_packet, we invoke the analyze_<tag> function for each instruction in
+the packet, and we mark the implicit writes.  The analysis determines if the packet
+semantics can be short-circuited.  If not, we initialize the result register for each
+of the predicated assignments.
 
 In addition to instruction semantics, we use a generator to create the decode
 tree.  This generation is a four step process.
diff --git a/target/hexagon/gen_analyze_funcs.py b/target/hexagon/gen_analyze_funcs.py
index a9af666cef..890e6a3a95 100755
--- a/target/hexagon/gen_analyze_funcs.py
+++ b/target/hexagon/gen_analyze_funcs.py
@@ -1,7 +1,7 @@
 #!/usr/bin/env python3
 
 ##
-##  Copyright(c) 2022-2023 Qualcomm Innovation Center, Inc. All Rights Reserved.
+##  Copyright(c) 2022-2024 Qualcomm Innovation Center, Inc. All Rights Reserved.
 ##
 ##  This program is free software; you can redistribute it and/or modify
 ##  it under the terms of the GNU General Public License as published by
@@ -44,15 +44,25 @@ def gen_analyze_func(f, tag, regs, imms):
 
     f.write("    Insn *insn G_GNUC_UNUSED = ctx->insn;\n")
 
-    i = 0
-    ## Analyze all the registers
-    for regtype, regid in regs:
-        reg = hex_common.get_register(tag, regtype, regid)
+    ## Declare all the registers
+    for regno, register in enumerate(regs):
+        reg_type, reg_id = register
+        reg = hex_common.get_register(tag, reg_type, reg_id)
+        reg.decl_reg_num(f, regno)
+
+    ## Analyze the register reads
+    for regno, register in enumerate(regs):
+        reg_type, reg_id = register
+        reg = hex_common.get_register(tag, reg_type, reg_id)
+        if reg.is_read():
+            reg.analyze_read(f, regno)
+
+    ## Analyze the register writes
+    for regno, register in enumerate(regs):
+        reg_type, reg_id = register
+        reg = hex_common.get_register(tag, reg_type, reg_id)
         if reg.is_written():
-            reg.analyze_write(f, tag, i)
-        else:
-            reg.analyze_read(f, i)
-        i += 1
+            reg.analyze_write(f, tag, regno)
 
     has_generated_helper = not hex_common.skip_qemu_helper(
         tag
@@ -89,13 +99,13 @@ def main():
     tagimms = hex_common.get_tagimms()
 
     with open(sys.argv[-1], "w") as f:
-        f.write("#ifndef HEXAGON_TCG_FUNCS_H\n")
-        f.write("#define HEXAGON_TCG_FUNCS_H\n\n")
+        f.write("#ifndef HEXAGON_ANALYZE_FUNCS_C_INC\n")
+        f.write("#define HEXAGON_ANALYZE_FUNCS_C_INC\n\n")
 
         for tag in hex_common.tags:
             gen_analyze_func(f, tag, tagregs[tag], tagimms[tag])
 
-        f.write("#endif    /* HEXAGON_TCG_FUNCS_H */\n")
+        f.write("#endif    /* HEXAGON_ANALYZE_FUNCS_C_INC */\n")
 
 
 if __name__ == "__main__":
diff --git a/target/hexagon/hex_common.py b/target/hexagon/hex_common.py
index 195620c7ec..33801e4bd7 100755
--- a/target/hexagon/hex_common.py
+++ b/target/hexagon/hex_common.py
@@ -1,7 +1,7 @@
 #!/usr/bin/env python3
 
 ##
-##  Copyright(c) 2019-2023 Qualcomm Innovation Center, Inc. All Rights Reserved.
+##  Copyright(c) 2019-2024 Qualcomm Innovation Center, Inc. All Rights Reserved.
 ##
 ##  This program is free software; you can redistribute it and/or modify
 ##  it under the terms of the GNU General Public License as published by
@@ -425,7 +425,6 @@ def log_write(self, f, tag):
             gen_log_reg_write(ctx, {self.reg_num}, {self.reg_tcg()});
         """))
     def analyze_write(self, f, tag, regno):
-        self.decl_reg_num(f, regno)
         predicated = "true" if is_predicated(tag) else "false"
         f.write(code_fmt(f"""\
             ctx_log_reg_write(ctx, {self.reg_num}, {predicated});
@@ -438,7 +437,6 @@ def decl_tcg(self, f, tag, regno):
             TCGv {self.reg_tcg()} = hex_gpr[{self.reg_num}];
         """))
     def analyze_read(self, f, regno):
-        self.decl_reg_num(f, regno)
         f.write(code_fmt(f"""\
             ctx_log_reg_read(ctx, {self.reg_num});
         """))
@@ -449,9 +447,8 @@ def decl_tcg(self, f, tag, regno):
             TCGv {self.reg_tcg()} = get_result_gpr(ctx, insn->regno[{regno}]);
         """))
     def analyze_read(self, f, regno):
-        self.decl_reg_num(f, regno)
         f.write(code_fmt(f"""\
-            ctx_log_reg_read(ctx, {self.reg_num});
+            ctx_log_reg_read_new(ctx, {self.reg_num});
         """))
 
 class GprReadWrite(Register, Single, ReadWrite):
@@ -471,8 +468,11 @@ def log_write(self, f, tag):
         f.write(code_fmt(f"""\
             gen_log_reg_write(ctx, {self.reg_num}, {self.reg_tcg()});
         """))
+    def analyze_read(self, f, regno):
+        f.write(code_fmt(f"""\
+            ctx_log_reg_read(ctx, {self.reg_num});
+        """))
     def analyze_write(self, f, tag, regno):
-        self.decl_reg_num(f, regno)
         predicated = "true" if is_predicated(tag) else "false"
         f.write(code_fmt(f"""\
             ctx_log_reg_write(ctx, {self.reg_num}, {predicated});
@@ -493,7 +493,6 @@ def log_write(self, f, tag):
             gen_write_ctrl_reg(ctx, {self.reg_num}, {self.reg_tcg()});
         """))
     def analyze_write(self, f, tag, regno):
-        self.decl_reg_num(f, regno)
         predicated = "true" if is_predicated(tag) else "false"
         f.write(code_fmt(f"""\
             ctx_log_reg_write(ctx, {self.reg_num}, {predicated});
@@ -511,7 +510,6 @@ def decl_tcg(self, f, tag, regno):
             gen_read_ctrl_reg(ctx, {self.reg_num}, {self.reg_tcg()});
         """))
     def analyze_read(self, f, regno):
-        self.decl_reg_num(f, regno)
         f.write(code_fmt(f"""\
             ctx_log_reg_read(ctx, {self.reg_num});
         """))
@@ -532,7 +530,6 @@ def idef_arg(self, declared):
         declared.append(self.reg_tcg())
         declared.append("CS")
     def analyze_read(self, f, regno):
-        self.decl_reg_num(f, regno)
         f.write(code_fmt(f"""\
             ctx_log_reg_read(ctx, {self.reg_num});
         """))
@@ -548,7 +545,6 @@ def log_write(self, f, tag):
             gen_log_pred_write(ctx, {self.reg_num}, {self.reg_tcg()});
         """))
     def analyze_write(self, f, tag, regno):
-        self.decl_reg_num(f, regno)
         f.write(code_fmt(f"""\
             ctx_log_pred_write(ctx, {self.reg_num});
         """))
@@ -560,7 +556,6 @@ def decl_tcg(self, f, tag, regno):
             TCGv {self.reg_tcg()} = hex_pred[{self.reg_num}];
         """))
     def analyze_read(self, f, regno):
-        self.decl_reg_num(f, regno)
         f.write(code_fmt(f"""\
             ctx_log_pred_read(ctx, {self.reg_num});
         """))
@@ -571,9 +566,8 @@ def decl_tcg(self, f, tag, regno):
             TCGv {self.reg_tcg()} = get_result_pred(ctx, insn->regno[{regno}]);
         """))
     def analyze_read(self, f, regno):
-        self.decl_reg_num(f, regno)
         f.write(code_fmt(f"""\
-            ctx_log_pred_read(ctx, {self.reg_num});
+            ctx_log_pred_read_new(ctx, {self.reg_num});
         """))
 
 class PredReadWrite(Register, Single, ReadWrite):
@@ -587,8 +581,11 @@ def log_write(self, f, tag):
         f.write(code_fmt(f"""\
             gen_log_pred_write(ctx, {self.reg_num}, {self.reg_tcg()});
         """))
+    def analyze_read(self, f, regno):
+        f.write(code_fmt(f"""\
+            ctx_log_pred_read(ctx, {self.reg_num});
+        """))
     def analyze_write(self, f, tag, regno):
-        self.decl_reg_num(f, regno)
         f.write(code_fmt(f"""\
             ctx_log_pred_write(ctx, {self.reg_num});
         """))
@@ -605,7 +602,6 @@ def log_write(self, f, tag):
             gen_log_reg_write_pair(ctx, {self.reg_num}, {self.reg_tcg()});
         """))
     def analyze_write(self, f, tag, regno):
-        self.decl_reg_num(f, regno)
         predicated = "true" if is_predicated(tag) else "false"
         f.write(code_fmt(f"""\
             ctx_log_reg_write_pair(ctx, {self.reg_num}, {predicated});
@@ -621,7 +617,6 @@ def decl_tcg(self, f, tag, regno):
                                     hex_gpr[{self.reg_num} + 1]);
         """))
     def analyze_read(self, f, regno):
-        self.decl_reg_num(f, regno)
         f.write(code_fmt(f"""\
             ctx_log_reg_read_pair(ctx, {self.reg_num});
         """))
@@ -640,8 +635,11 @@ def log_write(self, f, tag):
         f.write(code_fmt(f"""\
             gen_log_reg_write_pair(ctx, {self.reg_num}, {self.reg_tcg()});
         """))
+    def analyze_read(self, f, regno):
+        f.write(code_fmt(f"""\
+            ctx_log_reg_read_pair(ctx, {self.reg_num});
+        """))
     def analyze_write(self, f, tag, regno):
-        self.decl_reg_num(f, regno)
         predicated = "true" if is_predicated(tag) else "false"
         f.write(code_fmt(f"""\
             ctx_log_reg_write_pair(ctx, {self.reg_num}, {predicated});
@@ -663,7 +661,6 @@ def log_write(self, f, tag):
             gen_write_ctrl_reg_pair(ctx, {self.reg_num}, {self.reg_tcg()});
         """))
     def analyze_write(self, f, tag, regno):
-        self.decl_reg_num(f, regno)
         predicated = "true" if is_predicated(tag) else "false"
         f.write(code_fmt(f"""\
             ctx_log_reg_write_pair(ctx, {self.reg_num}, {predicated});
@@ -681,7 +678,6 @@ def decl_tcg(self, f, tag, regno):
             gen_read_ctrl_reg_pair(ctx, {self.reg_num}, {self.reg_tcg()});
         """))
     def analyze_read(self, f, regno):
-        self.decl_reg_num(f, regno)
         f.write(code_fmt(f"""\
             ctx_log_reg_read_pair(ctx, {self.reg_num});
         """))
@@ -705,7 +701,6 @@ def helper_hvx_desc(self, f):
             /* {self.reg_tcg()} is *(MMVector *)({self.helper_arg_name()}) */
         """))
     def analyze_write(self, f, tag, regno):
-        self.decl_reg_num(f, regno)
         newv = hvx_newv(tag)
         predicated = "true" if is_predicated(tag) else "false"
         f.write(code_fmt(f"""\
@@ -728,7 +723,6 @@ def helper_hvx_desc(self, f):
             /* {self.reg_tcg()} is *(MMVector *)({self.helper_arg_name()}) */
         """))
     def analyze_read(self, f, regno):
-        self.decl_reg_num(f, regno)
         f.write(code_fmt(f"""\
             ctx_log_vreg_read(ctx, {self.reg_num});
         """))
@@ -746,9 +740,8 @@ def helper_hvx_desc(self, f):
             /* {self.reg_tcg()} is *(MMVector *)({self.helper_arg_name()}) */
         """))
     def analyze_read(self, f, regno):
-        self.decl_reg_num(f, regno)
         f.write(code_fmt(f"""\
-            ctx_log_vreg_read(ctx, {self.reg_num});
+            ctx_log_vreg_read_new(ctx, {self.reg_num});
         """))
 
 class VRegReadWrite(Register, Hvx, ReadWrite):
@@ -772,8 +765,11 @@ def helper_hvx_desc(self, f):
         f.write(code_fmt(f"""\
             /* {self.reg_tcg()} is *(MMVector *)({self.helper_arg_name()}) */
         """))
+    def analyze_read(self, f, regno):
+        f.write(code_fmt(f"""\
+            ctx_log_vreg_read(ctx, {self.reg_num});
+        """))
     def analyze_write(self, f, tag, regno):
-        self.decl_reg_num(f, regno)
         newv = hvx_newv(tag)
         predicated = "true" if is_predicated(tag) else "false"
         f.write(code_fmt(f"""\
@@ -803,8 +799,11 @@ def helper_hvx_desc(self, f):
         f.write(code_fmt(f"""\
             /* {self.reg_tcg()} is *(MMVector *)({self.helper_arg_name()}) */
         """))
+    def analyze_read(self, f, regno):
+        f.write(code_fmt(f"""\
+            ctx_log_vreg_read(ctx, {self.reg_num});
+        """))
     def analyze_write(self, f, tag, regno):
-        self.decl_reg_num(f, regno)
         newv = hvx_newv(tag)
         predicated = "true" if is_predicated(tag) else "false"
         f.write(code_fmt(f"""\
@@ -830,7 +829,6 @@ def helper_hvx_desc(self, f):
             /* {self.reg_tcg()} is *(MMVectorPair *)({self.helper_arg_name()}) */
         """))
     def analyze_write(self, f, tag, regno):
-        self.decl_reg_num(f, regno)
         newv = hvx_newv(tag)
         predicated = "true" if is_predicated(tag) else "false"
         f.write(code_fmt(f"""\
@@ -860,7 +858,6 @@ def helper_hvx_desc(self, f):
             /* {self.reg_tcg()} is *(MMVectorPair *)({self.helper_arg_name()}) */
         """))
     def analyze_read(self, f, regno):
-        self.decl_reg_num(f, regno)
         f.write(code_fmt(f"""\
             ctx_log_vreg_read_pair(ctx, {self.reg_num});
         """))
@@ -892,8 +889,11 @@ def helper_hvx_desc(self, f):
         f.write(code_fmt(f"""\
             /* {self.reg_tcg()} is *(MMVectorPair *)({self.helper_arg_name()}) */
         """))
+    def analyze_read(self, f, regno):
+        f.write(code_fmt(f"""\
+            ctx_log_vreg_read_pair(ctx, {self.reg_num});
+        """))
     def analyze_write(self, f, tag, regno):
-        self.decl_reg_num(f, regno)
         newv = hvx_newv(tag)
         predicated = "true" if is_predicated(tag) else "false"
         f.write(code_fmt(f"""\
@@ -919,7 +919,6 @@ def helper_hvx_desc(self, f):
             /* {self.reg_tcg()} is *(MMQReg *)({self.helper_arg_name()}) */
         """))
     def analyze_write(self, f, tag, regno):
-        self.decl_reg_num(f, regno)
         f.write(code_fmt(f"""\
             ctx_log_qreg_write(ctx, {self.reg_num});
         """))
@@ -941,7 +940,6 @@ def helper_hvx_desc(self, f):
             /* {self.reg_tcg()} is *(MMQReg *)({self.helper_arg_name()}) */
         """))
     def analyze_read(self, f, regno):
-        self.decl_reg_num(f, regno)
         f.write(code_fmt(f"""\
             ctx_log_qreg_read(ctx, {self.reg_num});
         """))
@@ -967,8 +965,11 @@ def helper_hvx_desc(self, f):
         f.write(code_fmt(f"""\
             /* {self.reg_tcg()} is *(MMQReg *)({self.helper_arg_name()}) */
         """))
+    def analyze_read(self, f, regno):
+        f.write(code_fmt(f"""\
+            ctx_log_qreg_read(ctx, {self.reg_num});
+        """))
     def analyze_write(self, f, tag, regno):
-        self.decl_reg_num(f, regno)
         f.write(code_fmt(f"""\
             ctx_log_qreg_write(ctx, {self.reg_num});
         """))
diff --git a/target/hexagon/translate.h b/target/hexagon/translate.h
index 4dd59c6726..f06d71fc53 100644
--- a/target/hexagon/translate.h
+++ b/target/hexagon/translate.h
@@ -1,5 +1,5 @@
 /*
- *  Copyright(c) 2019-2023 Qualcomm Innovation Center, Inc. All Rights Reserved.
+ *  Copyright(c) 2019-2024 Qualcomm Innovation Center, Inc. All Rights Reserved.
  *
  *  This program is free software; you can redistribute it and/or modify
  *  it under the terms of the GNU General Public License as published by
@@ -75,6 +75,8 @@ typedef struct DisasContext {
     TCGv dczero_addr;
 } DisasContext;
 
+bool is_gather_store_insn(DisasContext *ctx);
+
 static inline void ctx_log_pred_write(DisasContext *ctx, int pnum)
 {
     if (!test_bit(pnum, ctx->pregs_written)) {
@@ -89,6 +91,12 @@ static inline void ctx_log_pred_read(DisasContext *ctx, int pnum)
     set_bit(pnum, ctx->pregs_read);
 }
 
+static inline void ctx_log_pred_read_new(DisasContext *ctx, int pnum)
+{
+    g_assert(test_bit(pnum, ctx->pregs_written));
+    set_bit(pnum, ctx->pregs_read);
+}
+
 static inline void ctx_log_reg_write(DisasContext *ctx, int rnum,
                                      bool is_predicated)
 {
@@ -120,6 +128,12 @@ static inline void ctx_log_reg_read(DisasContext *ctx, int rnum)
     set_bit(rnum, ctx->regs_read);
 }
 
+static inline void ctx_log_reg_read_new(DisasContext *ctx, int rnum)
+{
+    g_assert(test_bit(rnum, ctx->regs_written));
+    set_bit(rnum, ctx->regs_read);
+}
+
 static inline void ctx_log_reg_read_pair(DisasContext *ctx, int rnum)
 {
     ctx_log_reg_read(ctx, rnum);
@@ -171,6 +185,15 @@ static inline void ctx_log_vreg_read(DisasContext *ctx, int rnum)
     set_bit(rnum, ctx->vregs_read);
 }
 
+static inline void ctx_log_vreg_read_new(DisasContext *ctx, int rnum)
+{
+    g_assert(is_gather_store_insn(ctx) ||
+             test_bit(rnum, ctx->vregs_updated) ||
+             test_bit(rnum, ctx->vregs_select) ||
+             test_bit(rnum, ctx->vregs_updated_tmp));
+    set_bit(rnum, ctx->vregs_read);
+}
+
 static inline void ctx_log_vreg_read_pair(DisasContext *ctx, int rnum)
 {
     ctx_log_vreg_read(ctx, rnum ^ 0);
@@ -205,7 +228,6 @@ extern TCGv hex_vstore_addr[VSTORES_MAX];
 extern TCGv hex_vstore_size[VSTORES_MAX];
 extern TCGv hex_vstore_pending[VSTORES_MAX];
 
-bool is_gather_store_insn(DisasContext *ctx);
 void process_store(DisasContext *ctx, int slot_num);
 
 FIELD(PROBE_PKT_SCALAR_STORE_S0, MMU_IDX,       0, 2)
-- 
2.25.1