| Message ID | 20240328103138.297823-1-demin.han@starfivetech.com |
|---|---|
| State | New |
| Headers | show |
| Series | RISC-V: Refine the condition for add additional vars in RVV cost model | expand |
Thanks a lot for trying to optimize the dynamic LMUL cost model. The need_additional_vector_vars_p looks good to me. But - = (*program_points_per_bb.get (bb)).length () - 1; + = (*program_points_per_bb.get (bb)).length (); I wonder why you remove - 1? juzhe.zhong@rivai.ai From: demin.han Date: 2024-03-28 18:31 To: gcc-patches CC: juzhe.zhong; kito.cheng; pan2.li; jeffreyalaw; rdapp.gcc Subject: [PATCH] RISC-V: Refine the condition for add additional vars in RVV cost model The adjacent_dr_p is sufficient and unnecessary condition for contiguous access. So unnecessary live-ranges are added and result in spill. This patch uses MEMORY_ACCESS_TYPE as condition and constrains segment load/store. Tested on RV64 and no regression. PR target/114506 gcc/ChangeLog: * config/riscv/riscv-vector-costs.cc (non_contiguous_memory_access_p): Rename (need_additional_vector_vars_p): Rename and refine condition gcc/testsuite/ChangeLog: * gcc.dg/vect/costmodel/riscv/rvv/pr114506.c: New test. Signed-off-by: demin.han <demin.han@starfivetech.com> --- gcc/config/riscv/riscv-vector-costs.cc | 25 ++++++++++++------- .../vect/costmodel/riscv/rvv/pr114506.c | 23 +++++++++++++++++ 2 files changed, 39 insertions(+), 9 deletions(-) create mode 100644 gcc/testsuite/gcc.dg/vect/costmodel/riscv/rvv/pr114506.c diff --git a/gcc/config/riscv/riscv-vector-costs.cc b/gcc/config/riscv/riscv-vector-costs.cc index f462c272a6e..9f7fe936a29 100644 --- a/gcc/config/riscv/riscv-vector-costs.cc +++ b/gcc/config/riscv/riscv-vector-costs.cc @@ -563,14 +563,24 @@ get_store_value (gimple *stmt) return gimple_assign_rhs1 (stmt); } -/* Return true if it is non-contiguous load/store. */ +/* Return true if addtional vector vars needed. */ static bool -non_contiguous_memory_access_p (stmt_vec_info stmt_info) +need_additional_vector_vars_p (stmt_vec_info stmt_info) { enum stmt_vec_info_type type = STMT_VINFO_TYPE (vect_stmt_to_vectorize (stmt_info)); - return ((type == load_vec_info_type || type == store_vec_info_type) - && !adjacent_dr_p (STMT_VINFO_DATA_REF (stmt_info))); + if (type == load_vec_info_type || type == store_vec_info_type) + { + if (STMT_VINFO_GATHER_SCATTER_P (stmt_info) + && STMT_VINFO_MEMORY_ACCESS_TYPE (stmt_info) == VMAT_GATHER_SCATTER) + return true; + + machine_mode mode = TYPE_MODE (STMT_VINFO_VECTYPE (stmt_info)); + int lmul = riscv_get_v_regno_alignment (mode); + if (DR_GROUP_SIZE (stmt_info) * lmul > RVV_M8) + return true; + } + return false; } /* Return the LMUL of the current analysis. */ @@ -739,10 +749,7 @@ update_local_live_ranges ( stmt_vec_info stmt_info = vinfo->lookup_stmt (gsi_stmt (si)); enum stmt_vec_info_type type = STMT_VINFO_TYPE (vect_stmt_to_vectorize (stmt_info)); - if (non_contiguous_memory_access_p (stmt_info) - /* LOAD_LANES/STORE_LANES doesn't need a perm indice. */ - && STMT_VINFO_MEMORY_ACCESS_TYPE (stmt_info) - != VMAT_LOAD_STORE_LANES) + if (need_additional_vector_vars_p (stmt_info)) { /* For non-adjacent load/store STMT, we will potentially convert it into: @@ -752,7 +759,7 @@ update_local_live_ranges ( We will be likely using one more vector variable. */ unsigned int max_point - = (*program_points_per_bb.get (bb)).length () - 1; + = (*program_points_per_bb.get (bb)).length (); auto *live_ranges = live_ranges_per_bb.get (bb); bool existed_p = false; tree var = type == load_vec_info_type diff --git a/gcc/testsuite/gcc.dg/vect/costmodel/riscv/rvv/pr114506.c b/gcc/testsuite/gcc.dg/vect/costmodel/riscv/rvv/pr114506.c new file mode 100644 index 00000000000..a88d24b2d2d --- /dev/null +++ b/gcc/testsuite/gcc.dg/vect/costmodel/riscv/rvv/pr114506.c @@ -0,0 +1,23 @@ +/* { dg-do compile } */ +/* { dg-options "-march=rv64gcv -mabi=lp64d -O3 -ftree-vectorize -mrvv-max-lmul=dynamic -fdump-tree-vect-details" } */ + +float a[32000], b[32000], c[32000], d[32000]; +float aa[256][256], bb[256][256], cc[256][256]; + +void +s2275 () +{ + for (int i = 0; i < 256; i++) + { + for (int j = 0; j < 256; j++) + { + aa[j][i] = aa[j][i] + bb[j][i] * cc[j][i]; + } + a[i] = b[i] + c[i] * d[i]; + } +} + +/* { dg-final { scan-assembler-times {e32,m8} 1 } } */ +/* { dg-final { scan-assembler-not {e32,m4} } } */ +/* { dg-final { scan-assembler-not {csrr} } } */ +/* { dg-final { scan-tree-dump-not "Preferring smaller LMUL loop because it has unexpected spills" "vect" } } */
Hi,
the point starts from 1. the max_point should equal to length();
Should I prepare an individual patch for this?
From: juzhe.zhong@rivai.ai <juzhe.zhong@rivai.ai>
Sent: 2024年3月28日 18:45
To: Demin Han <demin.han@starfivetech.com>; gcc-patches <gcc-patches@gcc.gnu.org>
Cc: kito.cheng <kito.cheng@gmail.com>; pan2.li <pan2.li@intel.com>; jeffreyalaw <jeffreyalaw@gmail.com>; Robin Dapp <rdapp.gcc@gmail.com>
Subject: Re: [PATCH] RISC-V: Refine the condition for add additional vars in RVV cost model
Thanks a lot for trying to optimize the dynamic LMUL cost model.
The need_additional_vector_vars_p looks good to me.
But
- = (*program_points_per_bb.get (bb)).length () - 1;
+ = (*program_points_per_bb.get (bb)).length ();
I wonder why you remove - 1?
On 3/28/24 4:31 AM, demin.han wrote: > The adjacent_dr_p is sufficient and unnecessary condition for contiguous access. > So unnecessary live-ranges are added and result in spill. > > This patch uses MEMORY_ACCESS_TYPE as condition and constrains segment > load/store. > > Tested on RV64 and no regression. > > PR target/114506 > > gcc/ChangeLog: > > * config/riscv/riscv-vector-costs.cc (non_contiguous_memory_access_p): Rename > (need_additional_vector_vars_p): Rename and refine condition > > gcc/testsuite/ChangeLog: > > * gcc.dg/vect/costmodel/riscv/rvv/pr114506.c: New test. Note I think this should defer to gcc-15. It doesn't affect code correctness AFAICT and it's not a regression relative to gcc-13. jeff
diff --git a/gcc/config/riscv/riscv-vector-costs.cc b/gcc/config/riscv/riscv-vector-costs.cc index f462c272a6e..9f7fe936a29 100644 --- a/gcc/config/riscv/riscv-vector-costs.cc +++ b/gcc/config/riscv/riscv-vector-costs.cc @@ -563,14 +563,24 @@ get_store_value (gimple *stmt) return gimple_assign_rhs1 (stmt); } -/* Return true if it is non-contiguous load/store. */ +/* Return true if addtional vector vars needed. */ static bool -non_contiguous_memory_access_p (stmt_vec_info stmt_info) +need_additional_vector_vars_p (stmt_vec_info stmt_info) { enum stmt_vec_info_type type = STMT_VINFO_TYPE (vect_stmt_to_vectorize (stmt_info)); - return ((type == load_vec_info_type || type == store_vec_info_type) - && !adjacent_dr_p (STMT_VINFO_DATA_REF (stmt_info))); + if (type == load_vec_info_type || type == store_vec_info_type) + { + if (STMT_VINFO_GATHER_SCATTER_P (stmt_info) + && STMT_VINFO_MEMORY_ACCESS_TYPE (stmt_info) == VMAT_GATHER_SCATTER) + return true; + + machine_mode mode = TYPE_MODE (STMT_VINFO_VECTYPE (stmt_info)); + int lmul = riscv_get_v_regno_alignment (mode); + if (DR_GROUP_SIZE (stmt_info) * lmul > RVV_M8) + return true; + } + return false; } /* Return the LMUL of the current analysis. */ @@ -739,10 +749,7 @@ update_local_live_ranges ( stmt_vec_info stmt_info = vinfo->lookup_stmt (gsi_stmt (si)); enum stmt_vec_info_type type = STMT_VINFO_TYPE (vect_stmt_to_vectorize (stmt_info)); - if (non_contiguous_memory_access_p (stmt_info) - /* LOAD_LANES/STORE_LANES doesn't need a perm indice. */ - && STMT_VINFO_MEMORY_ACCESS_TYPE (stmt_info) - != VMAT_LOAD_STORE_LANES) + if (need_additional_vector_vars_p (stmt_info)) { /* For non-adjacent load/store STMT, we will potentially convert it into: @@ -752,7 +759,7 @@ update_local_live_ranges ( We will be likely using one more vector variable. */ unsigned int max_point - = (*program_points_per_bb.get (bb)).length () - 1; + = (*program_points_per_bb.get (bb)).length (); auto *live_ranges = live_ranges_per_bb.get (bb); bool existed_p = false; tree var = type == load_vec_info_type diff --git a/gcc/testsuite/gcc.dg/vect/costmodel/riscv/rvv/pr114506.c b/gcc/testsuite/gcc.dg/vect/costmodel/riscv/rvv/pr114506.c new file mode 100644 index 00000000000..a88d24b2d2d --- /dev/null +++ b/gcc/testsuite/gcc.dg/vect/costmodel/riscv/rvv/pr114506.c @@ -0,0 +1,23 @@ +/* { dg-do compile } */ +/* { dg-options "-march=rv64gcv -mabi=lp64d -O3 -ftree-vectorize -mrvv-max-lmul=dynamic -fdump-tree-vect-details" } */ + +float a[32000], b[32000], c[32000], d[32000]; +float aa[256][256], bb[256][256], cc[256][256]; + +void +s2275 () +{ + for (int i = 0; i < 256; i++) + { + for (int j = 0; j < 256; j++) + { + aa[j][i] = aa[j][i] + bb[j][i] * cc[j][i]; + } + a[i] = b[i] + c[i] * d[i]; + } +} + +/* { dg-final { scan-assembler-times {e32,m8} 1 } } */ +/* { dg-final { scan-assembler-not {e32,m4} } } */ +/* { dg-final { scan-assembler-not {csrr} } } */ +/* { dg-final { scan-tree-dump-not "Preferring smaller LMUL loop because it has unexpected spills" "vect" } } */
The adjacent_dr_p is sufficient and unnecessary condition for contiguous access. So unnecessary live-ranges are added and result in spill. This patch uses MEMORY_ACCESS_TYPE as condition and constrains segment load/store. Tested on RV64 and no regression. PR target/114506 gcc/ChangeLog: * config/riscv/riscv-vector-costs.cc (non_contiguous_memory_access_p): Rename (need_additional_vector_vars_p): Rename and refine condition gcc/testsuite/ChangeLog: * gcc.dg/vect/costmodel/riscv/rvv/pr114506.c: New test. Signed-off-by: demin.han <demin.han@starfivetech.com> --- gcc/config/riscv/riscv-vector-costs.cc | 25 ++++++++++++------- .../vect/costmodel/riscv/rvv/pr114506.c | 23 +++++++++++++++++ 2 files changed, 39 insertions(+), 9 deletions(-) create mode 100644 gcc/testsuite/gcc.dg/vect/costmodel/riscv/rvv/pr114506.c