public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
* [PATCH V3 0/2] Fix ICE with vwsll combine on 32bit targets
@ 2024-06-17 18:33 Edwin Lu
  2024-06-17 18:33 ` [PATCH V3 1/2] RISC-V: Fix vwsll combine on rv32 targets Edwin Lu
  2024-06-17 18:33 ` [PATCH V3 2/2] RISC-V: Move mode assertion out of conditional branch in emit_insn Edwin Lu
  0 siblings, 2 replies; 7+ messages in thread
From: Edwin Lu @ 2024-06-17 18:33 UTC (permalink / raw)
  To: gcc-patches; +Cc: gnu-toolchain, Edwin Lu

The following testcases have been failing on rv32 targets since 
r15-953-gaf4bf422a69:
FAIL: gcc.target/riscv/rvv/autovec/binop/vwsll-1.c (internal compiler
error: in maybe_legitimize_operand, at optabs.cc:8056)
FAIL: gcc.target/riscv/rvv/autovec/binop/vwsll-1.c (test for excess
errors)

Fix the bug and also robustify our emit_insn by making an assertion
check unconditional

I'm not sure if this ICE warrants its own separate testcase since it is
already being tested. I do have a minimal testcase on hand if we would
like to add one.

V2: Remove subreg condition and change assert to internal error

V3: Update the _trunc_scalar splitter as well

Edwin Lu (2):
  RISC-V: Fix vwsll combine on rv32 targets
  RISC-V: Move mode assertion out of conditional branch in emit_insn

 gcc/config/riscv/autovec-opt.md |  6 ++----
 gcc/config/riscv/riscv-v.cc     | 25 +++++++++++++++++++------
 2 files changed, 21 insertions(+), 10 deletions(-)

-- 
2.34.1


^ permalink raw reply	[flat|nested] 7+ messages in thread

* [PATCH V3 1/2] RISC-V: Fix vwsll combine on rv32 targets
  2024-06-17 18:33 [PATCH V3 0/2] Fix ICE with vwsll combine on 32bit targets Edwin Lu
@ 2024-06-17 18:33 ` Edwin Lu
  2024-06-17 23:31   ` Jeff Law
  2024-06-17 18:33 ` [PATCH V3 2/2] RISC-V: Move mode assertion out of conditional branch in emit_insn Edwin Lu
  1 sibling, 1 reply; 7+ messages in thread
From: Edwin Lu @ 2024-06-17 18:33 UTC (permalink / raw)
  To: gcc-patches; +Cc: gnu-toolchain, Edwin Lu, Robin Dapp

On rv32 targets, vwsll_zext1_scalar_<mode> would trigger an ice in
maybe_legitimize_instruction when zero extending a uint32 to uint64 due
to a mismatch between the input operand's mode (DI) and the expanded insn
operand's mode (Pmode == SI). Ensure that mode of the operands match

Tested on rv32/64 gcv newlib. Letting CI perform additional testing

gcc/ChangeLog:

	* config/riscv/autovec-opt.md: Fix mode mismatch

Signed-off-by: Edwin Lu <ewlu@rivosinc.com>
Co-authored-by: Robin Dapp <rdapp@ventanamicro.com>
---
V2: Remove subreg check

V3: Update _trunc_scalar splitter as well
---
 gcc/config/riscv/autovec-opt.md | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

diff --git a/gcc/config/riscv/autovec-opt.md b/gcc/config/riscv/autovec-opt.md
index 6a2eabbd854..d7a3cfd4602 100644
--- a/gcc/config/riscv/autovec-opt.md
+++ b/gcc/config/riscv/autovec-opt.md
@@ -1517,8 +1517,7 @@ (define_insn_and_split "*vwsll_zext1_scalar_<mode>"
   "&& 1"
   [(const_int 0)]
   {
-    if (GET_CODE (operands[2]) == SUBREG)
-      operands[2] = SUBREG_REG (operands[2]);
+    operands[2] = gen_lowpart (Pmode, operands[2]);
     insn_code icode = code_for_pred_vwsll_scalar (<MODE>mode);
     riscv_vector::emit_vlmax_insn (icode, riscv_vector::BINARY_OP, operands);
     DONE;
@@ -1584,8 +1583,7 @@ (define_insn_and_split "*vwsll_zext1_trunc_scalar_<mode>"
   "&& 1"
   [(const_int 0)]
   {
-    if (GET_CODE (operands[2]) == SUBREG)
-      operands[2] = SUBREG_REG (operands[2]);
+    operands[2] = gen_lowpart (Pmode, operands[2]);
     insn_code icode = code_for_pred_vwsll_scalar (<V_DOUBLE_TRUNC>mode);
     riscv_vector::emit_vlmax_insn (icode, riscv_vector::BINARY_OP, operands);
     DONE;
-- 
2.34.1


^ permalink raw reply	[flat|nested] 7+ messages in thread

* [PATCH V3 2/2] RISC-V: Move mode assertion out of conditional branch in emit_insn
  2024-06-17 18:33 [PATCH V3 0/2] Fix ICE with vwsll combine on 32bit targets Edwin Lu
  2024-06-17 18:33 ` [PATCH V3 1/2] RISC-V: Fix vwsll combine on rv32 targets Edwin Lu
@ 2024-06-17 18:33 ` Edwin Lu
  2024-06-17 23:33   ` Jeff Law
  1 sibling, 1 reply; 7+ messages in thread
From: Edwin Lu @ 2024-06-17 18:33 UTC (permalink / raw)
  To: gcc-patches; +Cc: gnu-toolchain, Edwin Lu, Robin Dapp

When emitting insns, we have an early assertion to ensure the input
operand's mode and the expanded operand's mode are the same; however, it
does not perform this check if the pattern does not have an explicit
machine mode specifying the operand. In this scenario, it will always
assume that mode = Pmode to correctly satisfy the
maybe_legitimize_operand check, however, there may be problems when
working in 32 bit environments.

Make the assert unconditional and replace it with an internal error for
more descriptive logging

gcc/ChangeLog:

	* config/riscv/riscv-v.cc: Move assert out of conditional block

Signed-off-by: Edwin Lu <ewlu@rivosinc.com>
Co-authored-by: Robin Dapp <rdapp@ventanamicro.com>
---
V2: change assert to internal error

V3: No change
---
 gcc/config/riscv/riscv-v.cc | 25 +++++++++++++++++++------
 1 file changed, 19 insertions(+), 6 deletions(-)

diff --git a/gcc/config/riscv/riscv-v.cc b/gcc/config/riscv/riscv-v.cc
index 8911f5783c8..5306711c1b7 100644
--- a/gcc/config/riscv/riscv-v.cc
+++ b/gcc/config/riscv/riscv-v.cc
@@ -50,6 +50,7 @@
 #include "rtx-vector-builder.h"
 #include "targhooks.h"
 #include "predict.h"
+#include "errors.h"
 
 using namespace riscv_vector;
 
@@ -290,11 +291,17 @@ public:
 	   always Pmode.  */
 	if (mode == VOIDmode)
 	  mode = Pmode;
-	else
-	  /* Early assertion ensures same mode since maybe_legitimize_operand
-	     will check this.  */
-	  gcc_assert (GET_MODE (ops[opno]) == VOIDmode
-		      || GET_MODE (ops[opno]) == mode);
+
+	/* Early assertion ensures same mode since maybe_legitimize_operand
+	   will check this.  */
+	machine_mode required_mode = GET_MODE (ops[opno]);
+	if (required_mode != VOIDmode && required_mode != mode)
+	  internal_error ("expected mode %s for operand %d of "
+			  "insn %s but got mode %s.\n",
+			  GET_MODE_NAME (mode),
+			  opno,
+			  insn_data[(int) icode].name,
+			  GET_MODE_NAME (required_mode));
 
 	add_input_operand (ops[opno], mode);
       }
@@ -346,7 +353,13 @@ public:
     else if (m_insn_flags & VXRM_RDN_P)
       add_rounding_mode_operand (VXRM_RDN);
 
-    gcc_assert (insn_data[(int) icode].n_operands == m_opno);
+
+    if (insn_data[(int) icode].n_operands != m_opno)
+      internal_error ("invalid number of operands for insn %s, "
+		      "expected %d but got %d.\n",
+		      insn_data[(int) icode].name,
+		      insn_data[(int) icode].n_operands, m_opno);
+
     expand (icode, any_mem_p);
   }
 
-- 
2.34.1


^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: [PATCH V3 1/2] RISC-V: Fix vwsll combine on rv32 targets
  2024-06-17 18:33 ` [PATCH V3 1/2] RISC-V: Fix vwsll combine on rv32 targets Edwin Lu
@ 2024-06-17 23:31   ` Jeff Law
  2024-06-18 22:34     ` [Committed " Edwin Lu
  0 siblings, 1 reply; 7+ messages in thread
From: Jeff Law @ 2024-06-17 23:31 UTC (permalink / raw)
  To: Edwin Lu, gcc-patches; +Cc: gnu-toolchain, Robin Dapp



On 6/17/24 12:33 PM, Edwin Lu wrote:
> On rv32 targets, vwsll_zext1_scalar_<mode> would trigger an ice in
> maybe_legitimize_instruction when zero extending a uint32 to uint64 due
> to a mismatch between the input operand's mode (DI) and the expanded insn
> operand's mode (Pmode == SI). Ensure that mode of the operands match
> 
> Tested on rv32/64 gcv newlib. Letting CI perform additional testing
> 
> gcc/ChangeLog:
> 
> 	* config/riscv/autovec-opt.md: Fix mode mismatch
OK
jeff
> 

^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: [PATCH V3 2/2] RISC-V: Move mode assertion out of conditional branch in emit_insn
  2024-06-17 18:33 ` [PATCH V3 2/2] RISC-V: Move mode assertion out of conditional branch in emit_insn Edwin Lu
@ 2024-06-17 23:33   ` Jeff Law
  2024-06-18 22:34     ` [Committed " Edwin Lu
  0 siblings, 1 reply; 7+ messages in thread
From: Jeff Law @ 2024-06-17 23:33 UTC (permalink / raw)
  To: Edwin Lu, gcc-patches; +Cc: gnu-toolchain, Robin Dapp



On 6/17/24 12:33 PM, Edwin Lu wrote:
> When emitting insns, we have an early assertion to ensure the input
> operand's mode and the expanded operand's mode are the same; however, it
> does not perform this check if the pattern does not have an explicit
> machine mode specifying the operand. In this scenario, it will always
> assume that mode = Pmode to correctly satisfy the
> maybe_legitimize_operand check, however, there may be problems when
> working in 32 bit environments.
> 
> Make the assert unconditional and replace it with an internal error for
> more descriptive logging
> 
> gcc/ChangeLog:
> 
> 	* config/riscv/riscv-v.cc: Move assert out of conditional block
OK.

Jeff


^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: [Committed V3 1/2] RISC-V: Fix vwsll combine on rv32 targets
  2024-06-17 23:31   ` Jeff Law
@ 2024-06-18 22:34     ` Edwin Lu
  0 siblings, 0 replies; 7+ messages in thread
From: Edwin Lu @ 2024-06-18 22:34 UTC (permalink / raw)
  To: Jeff Law, gcc-patches; +Cc: gnu-toolchain, Robin Dapp

Committed. Thanks!

Edwin

On 6/17/2024 5:31 PM, Jeff Law wrote:
>
>
> On 6/17/24 12:33 PM, Edwin Lu wrote:
>> On rv32 targets, vwsll_zext1_scalar_<mode> would trigger an ice in
>> maybe_legitimize_instruction when zero extending a uint32 to uint64 due
>> to a mismatch between the input operand's mode (DI) and the expanded 
>> insn
>> operand's mode (Pmode == SI). Ensure that mode of the operands match
>>
>> Tested on rv32/64 gcv newlib. Letting CI perform additional testing
>>
>> gcc/ChangeLog:
>>
>>     * config/riscv/autovec-opt.md: Fix mode mismatch
> OK
> jeff
>>

^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: [Committed V3 2/2] RISC-V: Move mode assertion out of conditional branch in emit_insn
  2024-06-17 23:33   ` Jeff Law
@ 2024-06-18 22:34     ` Edwin Lu
  0 siblings, 0 replies; 7+ messages in thread
From: Edwin Lu @ 2024-06-18 22:34 UTC (permalink / raw)
  To: Jeff Law, gcc-patches; +Cc: gnu-toolchain, Robin Dapp

Thanks!

Edwin

On 6/17/2024 5:33 PM, Jeff Law wrote:
>
>
> On 6/17/24 12:33 PM, Edwin Lu wrote:
>> When emitting insns, we have an early assertion to ensure the input
>> operand's mode and the expanded operand's mode are the same; however, it
>> does not perform this check if the pattern does not have an explicit
>> machine mode specifying the operand. In this scenario, it will always
>> assume that mode = Pmode to correctly satisfy the
>> maybe_legitimize_operand check, however, there may be problems when
>> working in 32 bit environments.
>>
>> Make the assert unconditional and replace it with an internal error for
>> more descriptive logging
>>
>> gcc/ChangeLog:
>>
>>     * config/riscv/riscv-v.cc: Move assert out of conditional block
> OK.
>
> Jeff
>

^ permalink raw reply	[flat|nested] 7+ messages in thread

end of thread, other threads:[~2024-06-18 22:34 UTC | newest]

Thread overview: 7+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2024-06-17 18:33 [PATCH V3 0/2] Fix ICE with vwsll combine on 32bit targets Edwin Lu
2024-06-17 18:33 ` [PATCH V3 1/2] RISC-V: Fix vwsll combine on rv32 targets Edwin Lu
2024-06-17 23:31   ` Jeff Law
2024-06-18 22:34     ` [Committed " Edwin Lu
2024-06-17 18:33 ` [PATCH V3 2/2] RISC-V: Move mode assertion out of conditional branch in emit_insn Edwin Lu
2024-06-17 23:33   ` Jeff Law
2024-06-18 22:34     ` [Committed " Edwin Lu

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).