Committed, thanks Juzhe. Pan
From: 钟居哲 <juzhe.zh...@rivai.ai> Sent: Saturday, September 16, 2023 7:21 AM To: Li, Pan2 <pan2...@intel.com>; gcc-patches <gcc-patches@gcc.gnu.org> Cc: Li, Pan2 <pan2...@intel.com>; Wang, Yanzhang <yanzhang.w...@intel.com>; kito.cheng <kito.ch...@gmail.com> Subject: Re: [PATCH v1] RISC-V: Support FP SGNJX autovec for VLS mode lgtm ________________________________ juzhe.zh...@rivai.ai<mailto:juzhe.zh...@rivai.ai> From: pan2.li<mailto:pan2...@intel.com> Date: 2023-09-15 21:23 To: gcc-patches<mailto:gcc-patches@gcc.gnu.org> CC: juzhe.zhong<mailto:juzhe.zh...@rivai.ai>; pan2.li<mailto:pan2...@intel.com>; yanzhang.wang<mailto:yanzhang.w...@intel.com>; kito.cheng<mailto:kito.ch...@gmail.com> Subject: [PATCH v1] RISC-V: Support FP SGNJX autovec for VLS mode From: Pan Li <pan2...@intel.com<mailto:pan2...@intel.com>> This patch would like to allow the VLS mode autovec for the floating-point binary operation SGNJX. Give sample code as below: void test (float * restrict out, float * restrict in1, float * restrict in2) { for (int i = 0; i < 128; i++) out[i] = in1[i] * copysignf (1.0, in2[i]); } Before this patch: test: li a5,128 vsetvli zero,a5,e32,m1,ta,ma vle32.v v2,0(a1) lui a4,%hi(.LC0) flw fa5,%lo(.LC0)(a4) vfmv.v.f v1,fa5 vle32.v v3,0(a2) vfsgnj.vv v1,v1,v3 vfmul.vv v1,v1,v2 vse32.v v1,0(a0) ret After this patch: test: li a5,128 vsetvli zero,a5,e32,m1,ta,ma vle32.v v1,0(a1) vle32.v v2,0(a2) vfsgnjx.vv v1,v1,v2 vse32.v v1,0(a0) ret This SGNJX autovec acts on function call copysignf/copysignf in math.h too. And it depends on the option -ffast-math. gcc/ChangeLog: * config/riscv/autovec-vls.md (xorsign<mode>3): New pattern. gcc/testsuite/ChangeLog: * gcc.target/riscv/rvv/autovec/vls/def.h: New macro. * gcc.target/riscv/rvv/autovec/vls/floating-point-sgnjx-1.c: New test. * gcc.target/riscv/rvv/autovec/vls/floating-point-sgnjx-2.c: New test. Signed-off-by: Pan Li <pan2...@intel.com<mailto:pan2...@intel.com>> --- gcc/config/riscv/autovec-vls.md | 21 +++++++++ .../gcc.target/riscv/rvv/autovec/vls/def.h | 8 ++++ .../rvv/autovec/vls/floating-point-sgnjx-1.c | 43 +++++++++++++++++++ .../rvv/autovec/vls/floating-point-sgnjx-2.c | 31 +++++++++++++ 4 files changed, 103 insertions(+) create mode 100644 gcc/testsuite/gcc.target/riscv/rvv/autovec/vls/floating-point-sgnjx-1.c create mode 100644 gcc/testsuite/gcc.target/riscv/rvv/autovec/vls/floating-point-sgnjx-2.c diff --git a/gcc/config/riscv/autovec-vls.md b/gcc/config/riscv/autovec-vls.md index 6f48f7d6232..d4ed2081537 100644 --- a/gcc/config/riscv/autovec-vls.md +++ b/gcc/config/riscv/autovec-vls.md @@ -289,6 +289,27 @@ (define_insn_and_split "copysign<mode>3" [(set_attr "type" "vector")] ) +;; ------------------------------------------------------------------------- +;; Includes: +;; - vfsgnjx.vv +;; - vfsgnjx.vf +;; ------------------------------------------------------------------------- +(define_insn_and_split "xorsign<mode>3" + [(set (match_operand:VLSF 0 "register_operand") + (unspec:VLSF + [(match_operand:VLSF 1 "register_operand") + (match_operand:VLSF 2 "register_operand")] UNSPEC_VXORSIGN))] + "TARGET_VECTOR && can_create_pseudo_p ()" + "#" + "&& 1" + [(const_int 0)] + { + riscv_vector::emit_vlmax_insn (code_for_pred (UNSPEC_VXORSIGN, <MODE>mode), + riscv_vector::BINARY_OP, operands); + DONE; + } +) + ;; ------------------------------------------------------------------------------- ;; ---- [INT] Unary operations ;; ------------------------------------------------------------------------------- diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/vls/def.h b/gcc/testsuite/gcc.target/riscv/rvv/autovec/vls/def.h index 1edc1910920..81c4570836b 100644 --- a/gcc/testsuite/gcc.target/riscv/rvv/autovec/vls/def.h +++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/vls/def.h @@ -258,3 +258,11 @@ typedef double v512df __attribute__ ((vector_size (4096))); for (int i = 0; i < NUM; ++i) \ a[i] = (b[i] > c[i]) OP (d[i] < e[i]); \ } + +#define DEF_SGNJX_VV(PREFIX, NUM, TYPE, CALL) \ + void __attribute__ ((noinline, noclone)) \ + PREFIX##_##TYPE##NUM (TYPE *restrict a, TYPE *restrict b, TYPE *restrict c) \ + { \ + for (int i = 0; i < NUM; ++i) \ + a[i] = b[i] * CALL (1.0, c[i]); \ + } diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/vls/floating-point-sgnjx-1.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/vls/floating-point-sgnjx-1.c new file mode 100644 index 00000000000..86c23ef0436 --- /dev/null +++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/vls/floating-point-sgnjx-1.c @@ -0,0 +1,43 @@ +/* { dg-do compile } */ +/* { dg-options "-march=rv64gcv_zvfh_zvl4096b -mabi=lp64d -O3 -fno-schedule-insns -fno-schedule-insns2 --param=riscv-autovec-lmul=m8 -ffast-math" } */ + +#include "def.h" + +DEF_SGNJX_VV (sgnj, 1, _Float16, __builtin_copysignf16) +DEF_SGNJX_VV (sgnj, 2, _Float16, __builtin_copysignf16) +DEF_SGNJX_VV (sgnj, 4, _Float16, __builtin_copysignf16) +DEF_SGNJX_VV (sgnj, 8, _Float16, __builtin_copysignf16) +DEF_SGNJX_VV (sgnj, 16, _Float16, __builtin_copysignf16) +DEF_SGNJX_VV (sgnj, 32, _Float16, __builtin_copysignf16) +DEF_SGNJX_VV (sgnj, 64, _Float16, __builtin_copysignf16) +DEF_SGNJX_VV (sgnj, 128, _Float16, __builtin_copysignf16) +DEF_SGNJX_VV (sgnj, 256, _Float16, __builtin_copysignf16) +DEF_SGNJX_VV (sgnj, 512, _Float16, __builtin_copysignf16) +DEF_SGNJX_VV (sgnj, 1024, _Float16, __builtin_copysignf16) +DEF_SGNJX_VV (sgnj, 2048, _Float16, __builtin_copysignf16) + +DEF_SGNJX_VV (sgnj, 1, float, __builtin_copysignf) +DEF_SGNJX_VV (sgnj, 2, float, __builtin_copysignf) +DEF_SGNJX_VV (sgnj, 4, float, __builtin_copysignf) +DEF_SGNJX_VV (sgnj, 8, float, __builtin_copysignf) +DEF_SGNJX_VV (sgnj, 16, float, __builtin_copysignf) +DEF_SGNJX_VV (sgnj, 32, float, __builtin_copysignf) +DEF_SGNJX_VV (sgnj, 64, float, __builtin_copysignf) +DEF_SGNJX_VV (sgnj, 128, float, __builtin_copysignf) +DEF_SGNJX_VV (sgnj, 256, float, __builtin_copysignf) +DEF_SGNJX_VV (sgnj, 512, float, __builtin_copysignf) +DEF_SGNJX_VV (sgnj, 1024, float, __builtin_copysignf) + +DEF_SGNJX_VV (sgnj, 1, double, __builtin_copysign) +DEF_SGNJX_VV (sgnj, 2, double, __builtin_copysign) +DEF_SGNJX_VV (sgnj, 4, double, __builtin_copysign) +DEF_SGNJX_VV (sgnj, 8, double, __builtin_copysign) +DEF_SGNJX_VV (sgnj, 16, double, __builtin_copysign) +DEF_SGNJX_VV (sgnj, 32, double, __builtin_copysign) +DEF_SGNJX_VV (sgnj, 64, double, __builtin_copysign) +DEF_SGNJX_VV (sgnj, 128, double, __builtin_copysign) +DEF_SGNJX_VV (sgnj, 256, double, __builtin_copysign) +DEF_SGNJX_VV (sgnj, 512, double, __builtin_copysign) + +/* { dg-final { scan-assembler-times {vfsgnjx\.vv\s+v[0-9]+,\s*v[0-9]+,\s*v[0-9]+} 30 } } */ +/* { dg-final { scan-assembler-not {csrr} } } */ diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/vls/floating-point-sgnjx-2.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/vls/floating-point-sgnjx-2.c new file mode 100644 index 00000000000..7e017de6a25 --- /dev/null +++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/vls/floating-point-sgnjx-2.c @@ -0,0 +1,31 @@ +/* { dg-do compile } */ +/* { dg-options "-march=rv64gcv_zvfh_zvl4096b -mabi=lp64d -O3 -fno-schedule-insns -fno-schedule-insns2 --param=riscv-autovec-lmul=m8 -ffast-math" } */ + +#include "def.h" +#include <math.h> + +DEF_SGNJX_VV (sgnj, 1, float, copysignf) +DEF_SGNJX_VV (sgnj, 2, float, copysignf) +DEF_SGNJX_VV (sgnj, 4, float, copysignf) +DEF_SGNJX_VV (sgnj, 8, float, copysignf) +DEF_SGNJX_VV (sgnj, 16, float, copysignf) +DEF_SGNJX_VV (sgnj, 32, float, copysignf) +DEF_SGNJX_VV (sgnj, 64, float, copysignf) +DEF_SGNJX_VV (sgnj, 128, float, copysignf) +DEF_SGNJX_VV (sgnj, 256, float, copysignf) +DEF_SGNJX_VV (sgnj, 512, float, copysignf) +DEF_SGNJX_VV (sgnj, 1024, float, copysignf) + +DEF_SGNJX_VV (sgnj, 1, double, copysign) +DEF_SGNJX_VV (sgnj, 2, double, copysign) +DEF_SGNJX_VV (sgnj, 4, double, copysign) +DEF_SGNJX_VV (sgnj, 8, double, copysign) +DEF_SGNJX_VV (sgnj, 16, double, copysign) +DEF_SGNJX_VV (sgnj, 32, double, copysign) +DEF_SGNJX_VV (sgnj, 64, double, copysign) +DEF_SGNJX_VV (sgnj, 128, double, copysign) +DEF_SGNJX_VV (sgnj, 256, double, copysign) +DEF_SGNJX_VV (sgnj, 512, double, copysign) + +/* { dg-final { scan-assembler-times {vfsgnjx\.vv\s+v[0-9]+,\s*v[0-9]+,\s*v[0-9]+} 19 } } */ +/* { dg-final { scan-assembler-not {csrr} } } */ -- 2.34.1