public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
From: yanzhang.wang@intel.com
To: gcc-patches@gcc.gnu.org
Cc: juzhe.zhong@rivai.ai, kito.cheng@sifive.com, pan2.li@intel.com,
	yanzhang.wang@intel.com
Subject: [PATCH v2] RISC-V: Fix regression of -fzero-call-used-regs=all
Date: Fri,  7 Apr 2023 14:59:40 +0800	[thread overview]
Message-ID: <20230407065940.2331101-1-yanzhang.wang@intel.com> (raw)
In-Reply-To: <20230406133441.1944365-1-yanzhang.wang@intel.com>

From: Yanzhang Wang <yanzhang.wang@intel.com>

This patch registers a riscv specific function to
TARGET_ZERO_CALL_USED_REGS instead of default in targhooks.cc. It will
clean gpr and vector relevant registers.

	PR 109104

gcc/ChangeLog:

	* config/riscv/riscv-protos.h (GCC_RISCV_PROTOS_H):
	(emit_hard_vlmax_vsetvl):
	(vector_zero_call_used_regs):
	* config/riscv/riscv-v.cc (emit_hard_vlmax_vsetvl):
	(emit_vlmax_vsetvl):
	(vector_zero_call_used_regs):
	* config/riscv/riscv.cc (riscv_zero_call_used_regs):
	(TARGET_ZERO_CALL_USED_REGS):

gcc/testsuite/ChangeLog:

	* gcc.target/riscv/zero-scratch-regs-1.c: New test.
	* gcc.target/riscv/zero-scratch-regs-2.c: New test.
	* gcc.target/riscv/zero-scratch-regs-3.c: New test.

Signed-off-by: Yanzhang Wang <yanzhang.wang@intel.com>
Co-authored-by: Pan Li <pan2.li@intel.com>
Co-authored-by: Ju-Zhe Zhong <juzhe.zhong@rivai.ai>
Co-authored-by: Kito Cheng <kito.cheng@sifive.com>
---
 gcc/config/riscv/riscv-protos.h               |  5 ++
 gcc/config/riscv/riscv-v.cc                   | 67 ++++++++++++++++++-
 gcc/config/riscv/riscv.cc                     | 21 ++++++
 .../gcc.target/riscv/zero-scratch-regs-1.c    |  9 +++
 .../gcc.target/riscv/zero-scratch-regs-2.c    | 24 +++++++
 .../gcc.target/riscv/zero-scratch-regs-3.c    | 57 ++++++++++++++++
 6 files changed, 180 insertions(+), 3 deletions(-)
 create mode 100644 gcc/testsuite/gcc.target/riscv/zero-scratch-regs-1.c
 create mode 100644 gcc/testsuite/gcc.target/riscv/zero-scratch-regs-2.c
 create mode 100644 gcc/testsuite/gcc.target/riscv/zero-scratch-regs-3.c

diff --git a/gcc/config/riscv/riscv-protos.h b/gcc/config/riscv/riscv-protos.h
index 4611447ddde..7ab0ec4b8be 100644
--- a/gcc/config/riscv/riscv-protos.h
+++ b/gcc/config/riscv/riscv-protos.h
@@ -22,6 +22,8 @@ along with GCC; see the file COPYING3.  If not see
 #ifndef GCC_RISCV_PROTOS_H
 #define GCC_RISCV_PROTOS_H
 
+#include "hard-reg-set.h"
+
 /* Symbol types we understand.  The order of this list must match that of
    the unspec enum in riscv.md, subsequent to UNSPEC_ADDRESS_FIRST.  */
 enum riscv_symbol_type {
@@ -159,6 +161,7 @@ bool check_builtin_call (location_t, vec<location_t>, unsigned int,
 bool const_vec_all_same_in_range_p (rtx, HOST_WIDE_INT, HOST_WIDE_INT);
 bool legitimize_move (rtx, rtx, machine_mode);
 void emit_vlmax_vsetvl (machine_mode, rtx);
+void emit_hard_vlmax_vsetvl (machine_mode, rtx);
 void emit_vlmax_op (unsigned, rtx, rtx, machine_mode);
 void emit_vlmax_op (unsigned, rtx, rtx, rtx, machine_mode);
 void emit_nonvlmax_op (unsigned, rtx, rtx, rtx, machine_mode);
@@ -206,6 +209,8 @@ enum vlen_enum
 bool slide1_sew64_helper (int, machine_mode, machine_mode,
 			  machine_mode, rtx *);
 rtx gen_avl_for_scalar_move (rtx);
+
+HARD_REG_SET vector_zero_call_used_regs (HARD_REG_SET);
 }
 
 /* We classify builtin types into two classes:
diff --git a/gcc/config/riscv/riscv-v.cc b/gcc/config/riscv/riscv-v.cc
index 2e91d019f6c..aad046240ee 100644
--- a/gcc/config/riscv/riscv-v.cc
+++ b/gcc/config/riscv/riscv-v.cc
@@ -43,6 +43,8 @@
 #include "optabs.h"
 #include "tm-constrs.h"
 #include "rtx-vector-builder.h"
+#include "diagnostic-core.h"
+#include "targhooks.h"
 
 using namespace riscv_vector;
 
@@ -118,6 +120,17 @@ const_vec_all_same_in_range_p (rtx x, HOST_WIDE_INT minval,
 	  && IN_RANGE (INTVAL (elt), minval, maxval));
 }
 
+/* Emit a vlmax vsetvl instruction with side effect, this should be only used
+   when optimization is tune off or emit after vsetvl insertion pass.  */
+void
+emit_hard_vlmax_vsetvl (machine_mode vmode, rtx vl)
+{
+  unsigned int sew = get_sew (vmode);
+  emit_insn (gen_vsetvl (Pmode, vl, RVV_VLMAX, gen_int_mode (sew, Pmode),
+			 gen_int_mode (get_vlmul (vmode), Pmode), const0_rtx,
+			 const0_rtx));
+}
+
 void
 emit_vlmax_vsetvl (machine_mode vmode, rtx vl)
 {
@@ -126,9 +139,7 @@ emit_vlmax_vsetvl (machine_mode vmode, rtx vl)
   unsigned int ratio = calculate_ratio (sew, vlmul);
 
   if (!optimize)
-    emit_insn (gen_vsetvl (Pmode, vl, RVV_VLMAX, gen_int_mode (sew, Pmode),
-			   gen_int_mode (get_vlmul (vmode), Pmode), const0_rtx,
-			   const0_rtx));
+    emit_hard_vlmax_vsetvl (vmode, vl);
   else
     emit_insn (gen_vlmax_avl (Pmode, vl, gen_int_mode (ratio, Pmode)));
 }
@@ -724,4 +735,54 @@ gen_avl_for_scalar_move (rtx avl)
     }
 }
 
+HARD_REG_SET
+vector_zero_call_used_regs (HARD_REG_SET need_zeroed_hardregs)
+{
+  HARD_REG_SET zeroed_hardregs;
+  CLEAR_HARD_REG_SET (zeroed_hardregs);
+
+  /* Find a register to hold vl.  */
+  unsigned vl_regno = INVALID_REGNUM;
+  /* Skip the first GPR, otherwise the existing vl is kept due to the same
+     between vl and avl.  */
+  for (unsigned regno = GP_REG_FIRST + 1; regno <= GP_REG_LAST; regno++)
+    {
+      if (TEST_HARD_REG_BIT (need_zeroed_hardregs, regno))
+	{
+	  vl_regno = regno;
+	  break;
+	}
+    }
+
+  if (vl_regno > GP_REG_LAST)
+    sorry ("can't allocate vl register for %qs on this target",
+	   "-fzero-call-used-regs");
+
+  bool emitted_vlmax_vsetvl = false;
+  rtx vl = gen_rtx_REG (Pmode, vl_regno); /* vl is VLMAX.  */
+  for (unsigned regno = V_REG_FIRST; regno <= V_REG_LAST; ++regno)
+    {
+      if (TEST_HARD_REG_BIT (need_zeroed_hardregs, regno))
+	{
+	  rtx target = regno_reg_rtx[regno];
+	  machine_mode mode = GET_MODE (target);
+	  poly_uint16 nunits = GET_MODE_NUNITS (mode);
+	  machine_mode mask_mode = get_vector_mode (BImode, nunits).require ();
+
+	  if (!emitted_vlmax_vsetvl)
+	    {
+	      emit_hard_vlmax_vsetvl (mode, vl);
+	      emitted_vlmax_vsetvl = true;
+	    }
+
+	  emit_vlmax_op (code_for_pred_mov (mode), target, CONST0_RTX (mode),
+			 vl, mask_mode);
+
+	  SET_HARD_REG_BIT (zeroed_hardregs, regno);
+	}
+    }
+
+  return zeroed_hardregs;
+}
+
 } // namespace riscv_vector
diff --git a/gcc/config/riscv/riscv.cc b/gcc/config/riscv/riscv.cc
index 5f542932d13..c3ae9571766 100644
--- a/gcc/config/riscv/riscv.cc
+++ b/gcc/config/riscv/riscv.cc
@@ -7066,6 +7066,24 @@ riscv_shamt_matches_mask_p (int shamt, HOST_WIDE_INT mask)
   return shamt == ctz_hwi (mask);
 }
 
+/* Generate a sequence of instructions that zero registers specified by
+   NEED_ZEROED_HARDREGS.  Return the ZEROED_HARDREGS that are actually
+   zeroed.  */
+HARD_REG_SET
+riscv_zero_call_used_regs (HARD_REG_SET need_zeroed_hardregs)
+{
+  HARD_REG_SET zeroed_hardregs;
+  CLEAR_HARD_REG_SET (zeroed_hardregs);
+
+  if (TARGET_VECTOR)
+    zeroed_hardregs
+      |= riscv_vector::vector_zero_call_used_regs (need_zeroed_hardregs);
+
+  return zeroed_hardregs | default_zero_call_used_regs (need_zeroed_hardregs
+							& ~zeroed_hardregs);
+}
+
+
 /* Initialize the GCC target structure.  */
 #undef TARGET_ASM_ALIGNED_HI_OP
 #define TARGET_ASM_ALIGNED_HI_OP "\t.half\t"
@@ -7317,6 +7335,9 @@ riscv_shamt_matches_mask_p (int shamt, HOST_WIDE_INT mask)
 #undef TARGET_DWARF_POLY_INDETERMINATE_VALUE
 #define TARGET_DWARF_POLY_INDETERMINATE_VALUE riscv_dwarf_poly_indeterminate_value
 
+#undef TARGET_ZERO_CALL_USED_REGS
+#define TARGET_ZERO_CALL_USED_REGS riscv_zero_call_used_regs
+
 struct gcc_target targetm = TARGET_INITIALIZER;
 
 #include "gt-riscv.h"
diff --git a/gcc/testsuite/gcc.target/riscv/zero-scratch-regs-1.c b/gcc/testsuite/gcc.target/riscv/zero-scratch-regs-1.c
new file mode 100644
index 00000000000..41d94ab921a
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/zero-scratch-regs-1.c
@@ -0,0 +1,9 @@
+/* { dg-do compile } */
+/* { dg-options "-march=rv64gc -mabi=lp64 -O2 -fzero-call-used-regs=used -fno-stack-protector -fno-PIC" } */
+
+void
+foo (void)
+{
+}
+
+/* { dg-final { scan-assembler-not "li\t" } } */
diff --git a/gcc/testsuite/gcc.target/riscv/zero-scratch-regs-2.c b/gcc/testsuite/gcc.target/riscv/zero-scratch-regs-2.c
new file mode 100644
index 00000000000..9161dd3d4ec
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/zero-scratch-regs-2.c
@@ -0,0 +1,24 @@
+/* { dg-do compile } */
+/* { dg-options "-march=rv64gc -mabi=lp64 -O2 -fzero-call-used-regs=all-gpr" } */
+
+void
+foo (void)
+{
+}
+
+/* { dg-final { scan-assembler-not "vsetvli" } } */
+/* { dg-final { scan-assembler "li\[ \t\]*t0,0" } } */
+/* { dg-final { scan-assembler "li\[ \t\]*t1,0" } } */
+/* { dg-final { scan-assembler "li\[ \t\]*t2,0" } } */
+/* { dg-final { scan-assembler "li\[ \t\]*a0,0" } } */
+/* { dg-final { scan-assembler "li\[ \t\]*a1,0" } } */
+/* { dg-final { scan-assembler "li\[ \t\]*a2,0" } } */
+/* { dg-final { scan-assembler "li\[ \t\]*a3,0" } } */
+/* { dg-final { scan-assembler "li\[ \t\]*a4,0" } } */
+/* { dg-final { scan-assembler "li\[ \t\]*a5,0" } } */
+/* { dg-final { scan-assembler "li\[ \t\]*a6,0" } } */
+/* { dg-final { scan-assembler "li\[ \t\]*a7,0" } } */
+/* { dg-final { scan-assembler "li\[ \t\]*t3,0" } } */
+/* { dg-final { scan-assembler "li\[ \t\]*t4,0" } } */
+/* { dg-final { scan-assembler "li\[ \t\]*t5,0" } } */
+/* { dg-final { scan-assembler "li\[ \t\]*t6,0" } } */
diff --git a/gcc/testsuite/gcc.target/riscv/zero-scratch-regs-3.c b/gcc/testsuite/gcc.target/riscv/zero-scratch-regs-3.c
new file mode 100644
index 00000000000..824fe9e548f
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/zero-scratch-regs-3.c
@@ -0,0 +1,57 @@
+/* { dg-do compile } */
+/* { dg-options "-march=rv64gcv -mabi=lp64 -O2 -fzero-call-used-regs=all" } */
+
+void
+foo (void)
+{
+}
+
+/* { dg-final { scan-assembler "vsetvli\[ \t\]*t0,zero,e32,m1,tu,mu" } } */
+/* { dg-final { scan-assembler "vmv.v.i\[ \t\]*v0,0" } } */
+/* { dg-final { scan-assembler "vmv.v.i\[ \t\]*v1,0" } } */
+/* { dg-final { scan-assembler "vmv.v.i\[ \t\]*v2,0" } } */
+/* { dg-final { scan-assembler "vmv.v.i\[ \t\]*v3,0" } } */
+/* { dg-final { scan-assembler "vmv.v.i\[ \t\]*v4,0" } } */
+/* { dg-final { scan-assembler "vmv.v.i\[ \t\]*v5,0" } } */
+/* { dg-final { scan-assembler "vmv.v.i\[ \t\]*v6,0" } } */
+/* { dg-final { scan-assembler "vmv.v.i\[ \t\]*v7,0" } } */
+/* { dg-final { scan-assembler "vmv.v.i\[ \t\]*v8,0" } } */
+/* { dg-final { scan-assembler "vmv.v.i\[ \t\]*v9,0" } } */
+/* { dg-final { scan-assembler "vmv.v.i\[ \t\]*v10,0" } } */
+/* { dg-final { scan-assembler "vmv.v.i\[ \t\]*v11,0" } } */
+/* { dg-final { scan-assembler "vmv.v.i\[ \t\]*v12,0" } } */
+/* { dg-final { scan-assembler "vmv.v.i\[ \t\]*v13,0" } } */
+/* { dg-final { scan-assembler "vmv.v.i\[ \t\]*v14,0" } } */
+/* { dg-final { scan-assembler "vmv.v.i\[ \t\]*v15,0" } } */
+/* { dg-final { scan-assembler "vmv.v.i\[ \t\]*v16,0" } } */
+/* { dg-final { scan-assembler "vmv.v.i\[ \t\]*v17,0" } } */
+/* { dg-final { scan-assembler "vmv.v.i\[ \t\]*v18,0" } } */
+/* { dg-final { scan-assembler "vmv.v.i\[ \t\]*v19,0" } } */
+/* { dg-final { scan-assembler "vmv.v.i\[ \t\]*v20,0" } } */
+/* { dg-final { scan-assembler "vmv.v.i\[ \t\]*v21,0" } } */
+/* { dg-final { scan-assembler "vmv.v.i\[ \t\]*v22,0" } } */
+/* { dg-final { scan-assembler "vmv.v.i\[ \t\]*v23,0" } } */
+/* { dg-final { scan-assembler "vmv.v.i\[ \t\]*v24,0" } } */
+/* { dg-final { scan-assembler "vmv.v.i\[ \t\]*v25,0" } } */
+/* { dg-final { scan-assembler "vmv.v.i\[ \t\]*v26,0" } } */
+/* { dg-final { scan-assembler "vmv.v.i\[ \t\]*v27,0" } } */
+/* { dg-final { scan-assembler "vmv.v.i\[ \t\]*v28,0" } } */
+/* { dg-final { scan-assembler "vmv.v.i\[ \t\]*v29,0" } } */
+/* { dg-final { scan-assembler "vmv.v.i\[ \t\]*v30,0" } } */
+/* { dg-final { scan-assembler "vmv.v.i\[ \t\]*v31,0" } } */
+/* { dg-final { scan-assembler "li\[ \t\]*t0,0" } } */
+/* { dg-final { scan-assembler "li\[ \t\]*t1,0" } } */
+/* { dg-final { scan-assembler "li\[ \t\]*t2,0" } } */
+/* { dg-final { scan-assembler "li\[ \t\]*a0,0" } } */
+/* { dg-final { scan-assembler "li\[ \t\]*a1,0" } } */
+/* { dg-final { scan-assembler "li\[ \t\]*a2,0" } } */
+/* { dg-final { scan-assembler "li\[ \t\]*a3,0" } } */
+/* { dg-final { scan-assembler "li\[ \t\]*a4,0" } } */
+/* { dg-final { scan-assembler "li\[ \t\]*a5,0" } } */
+/* { dg-final { scan-assembler "li\[ \t\]*a6,0" } } */
+/* { dg-final { scan-assembler "li\[ \t\]*a7,0" } } */
+/* { dg-final { scan-assembler "li\[ \t\]*t3,0" } } */
+/* { dg-final { scan-assembler "li\[ \t\]*t4,0" } } */
+/* { dg-final { scan-assembler "li\[ \t\]*t5,0" } } */
+/* { dg-final { scan-assembler "li\[ \t\]*t6,0" } } */
+/* { dg-final { scan-assembler "fmv.d.x\[ \t\]*ft0,zero" } } */
-- 
2.39.2


  parent reply	other threads:[~2023-04-07  7:00 UTC|newest]

Thread overview: 14+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2023-04-06 13:34 [PATCH] " yanzhang.wang
2023-04-06 13:47 ` juzhe.zhong
2023-04-06 14:59   ` Kito Cheng
2023-04-07  6:59 ` yanzhang.wang [this message]
2023-04-07  7:07   ` [PATCH v2] " Kito Cheng
2023-04-07 12:32 ` [PATCH v3] " yanzhang.wang
2023-04-08 18:39   ` Jeff Law
2023-04-10  2:21     ` Wang, Yanzhang
2023-04-10  3:11       ` Kito Cheng
2023-04-10 20:57         ` Jeff Law
2023-04-10  3:00 ` [PATCH v4] " yanzhang.wang
2023-04-11 11:37 ` [PATCH v5] " yanzhang.wang
2023-04-11 12:00   ` Wang, Yanzhang
2023-04-11 14:11     ` Kito Cheng

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20230407065940.2331101-1-yanzhang.wang@intel.com \
    --to=yanzhang.wang@intel.com \
    --cc=gcc-patches@gcc.gnu.org \
    --cc=juzhe.zhong@rivai.ai \
    --cc=kito.cheng@sifive.com \
    --cc=pan2.li@intel.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).