public inbox for gcc-cvs@sourceware.org
help / color / mirror / Atom feed
* [gcc r14-1993] RISC-V: Set the natural size of constant vector mask modes to one RVV data vector.
@ 2023-06-20 14:14 Pan Li
  0 siblings, 0 replies; only message in thread
From: Pan Li @ 2023-06-20 14:14 UTC (permalink / raw)
  To: gcc-cvs

https://gcc.gnu.org/g:cb421ffff66914e56a270dfb1cedb9bb890b43c2

commit r14-1993-gcb421ffff66914e56a270dfb1cedb9bb890b43c2
Author: Li Xu <xuli1@eswincomputing.com>
Date:   Tue Jun 20 06:46:18 2023 +0000

    RISC-V: Set the natural size of constant vector mask modes to one RVV data vector.
    
    If reinterpret vnx2bi as vnx16qi, vnx16qi must occupy no more of the underlying
    registers than vnx2bi.
    
    Consider this following case:
    void test_vreinterpret_v_b64_i8m1 (uint8_t *in, int8_t *out)
    {
      vbool64_t vmask = __riscv_vlm_v_b64 (in, 2);
      vint8m1_t vout = __riscv_vreinterpret_v_b64_i8m1 (vmask);
      __riscv_vse8_v_i8m1(out, vout, 16);
    }
    
    compiler parameters: -march=rv64gcv -mabi=lp64d --param=riscv-autovec-preference=fixed-vlmax -O3
    Compilation fails with:
    test_vreinterpret_v_b64_i8m1during RTL pass: expand
    
    test.c: In function 'test_vreinterpret_v_b64_i8m1':
    test.c:11:22: internal compiler error: in gen_lowpart_general, at rtlhooks.cc:57
       11 |     vint8m1_t vout = __riscv_vreinterpret_v_b64_i8m1(src);
          |                      ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
    0xf11876 gen_lowpart_general(machine_mode, rtx_def*)
            ../.././riscv-gcc/gcc/rtlhooks.cc:57
    0x191435e gen_vreinterpretvnx16qi(rtx_def*, rtx_def*)
            ../.././riscv-gcc/gcc/config/riscv/vector.md:486
    0xe08858 maybe_expand_insn(insn_code, unsigned int, expand_operand*)
            ../.././riscv-gcc/gcc/optabs.cc:8213
    0x1471209 riscv_vector::function_expander::generate_insn(insn_code)
            ../.././riscv-gcc/gcc/config/riscv/riscv-vector-builtins.cc:3813
    0x147629c riscv_vector::function_expander::expand()
            ../.././riscv-gcc/gcc/config/riscv/riscv-vector-builtins.h:520
    0x147629c riscv_vector::expand_builtin(unsigned int, tree_node*, rtx_def*)
            ../.././riscv-gcc/gcc/config/riscv/riscv-vector-builtins.cc:4103
    0x9868f9 expand_builtin(tree_node*, rtx_def*, rtx_def*, machine_mode, int)
            ../.././riscv-gcc/gcc/builtins.cc:7342
    
    gcc/ChangeLog:
    
            * config/riscv/riscv.cc (riscv_regmode_natural_size): set the natural
            size of vector mask mode to one rvv register.
    
    gcc/testsuite/ChangeLog:
    
            * gcc.target/riscv/rvv/autovec/vreinterpet-fixed.c: New test.

Diff:
---
 gcc/config/riscv/riscv.cc                                     |  5 +++++
 .../gcc.target/riscv/rvv/autovec/vreinterpet-fixed.c          | 11 +++++++++++
 2 files changed, 16 insertions(+)

diff --git a/gcc/config/riscv/riscv.cc b/gcc/config/riscv/riscv.cc
index 9558e28de3f..280aa0b33b9 100644
--- a/gcc/config/riscv/riscv.cc
+++ b/gcc/config/riscv/riscv.cc
@@ -7364,6 +7364,11 @@ riscv_regmode_natural_size (machine_mode mode)
      anything smaller than that.  */
   /* ??? For now, only do this for variable-width RVV registers.
      Doing it for constant-sized registers breaks lower-subreg.c.  */
+
+  /* RVV mask modes always consume a single register.  */
+  if (GET_MODE_CLASS (mode) == MODE_VECTOR_BOOL)
+    return BYTES_PER_RISCV_VECTOR;
+
   if (!riscv_vector_chunks.is_constant () && riscv_v_ext_mode_p (mode))
     {
       if (riscv_v_ext_tuple_mode_p (mode))
diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/vreinterpet-fixed.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/vreinterpet-fixed.c
new file mode 100644
index 00000000000..534d5fe0f0a
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/vreinterpet-fixed.c
@@ -0,0 +1,11 @@
+/* { dg-do compile } */
+/* { dg-options "-march=rv64gcv -mabi=lp64d --param=riscv-autovec-preference=fixed-vlmax -O3" } */
+
+#include "riscv_vector.h"
+
+void test_vreinterpret_v_b64_i8m1 (uint8_t *in, int8_t *out)
+{
+  vbool64_t vmask = __riscv_vlm_v_b64 (in, 2);
+  vint8m1_t vout = __riscv_vreinterpret_v_b64_i8m1 (vmask);
+  __riscv_vse8_v_i8m1(out, vout, 16);
+}

^ permalink raw reply	[flat|nested] only message in thread

only message in thread, other threads:[~2023-06-20 14:14 UTC | newest]

Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-06-20 14:14 [gcc r14-1993] RISC-V: Set the natural size of constant vector mask modes to one RVV data vector Pan Li

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).