https://gcc.gnu.org/g:e4b8db26de35239bd621aad9c0361f25d957122b
commit r15-2659-ge4b8db26de35239bd621aad9c0361f25d957122b Author: Pengxuan Zheng <quic_pzh...@quicinc.com> Date: Wed Jul 31 17:00:01 2024 -0700 aarch64: Improve Advanced SIMD popcount expansion by using SVE [PR113860] This patch improves the Advanced SIMD popcount expansion by using SVE if available. For example, GCC currently generates the following code sequence for V2DI: cnt v31.16b, v31.16b uaddlp v31.8h, v31.16b uaddlp v31.4s, v31.8h uaddlp v31.2d, v31.4s However, by using SVE, we can generate the following sequence instead: ptrue p7.b, all cnt z31.d, p7/m, z31.d Similar improvements can be made for V4HI, V8HI, V2SI and V4SI too. The scalar popcount expansion can also be improved similarly by using SVE and those changes will be included in a separate patch. PR target/113860 gcc/ChangeLog: * config/aarch64/aarch64-simd.md (popcount<mode>2): Add TARGET_SVE support. * config/aarch64/aarch64-sve.md (@aarch64_pred_<optab><mode>): Use new iterator SVE_VDQ_I. * config/aarch64/iterators.md (SVE_VDQ_I): New mode iterator. (VPRED): Add V8QI, V16QI, V4HI, V8HI and V2SI. gcc/testsuite/ChangeLog: * gcc.target/aarch64/popcnt-sve.c: New test. Signed-off-by: Pengxuan Zheng <quic_pzh...@quicinc.com> Diff: --- gcc/config/aarch64/aarch64-simd.md | 9 +++ gcc/config/aarch64/aarch64-sve.md | 13 ++-- gcc/config/aarch64/iterators.md | 5 ++ gcc/testsuite/gcc.target/aarch64/popcnt-sve.c | 88 +++++++++++++++++++++++++++ 4 files changed, 109 insertions(+), 6 deletions(-) diff --git a/gcc/config/aarch64/aarch64-simd.md b/gcc/config/aarch64/aarch64-simd.md index 459e11b09a19..816f499e9634 100644 --- a/gcc/config/aarch64/aarch64-simd.md +++ b/gcc/config/aarch64/aarch64-simd.md @@ -3508,6 +3508,15 @@ (popcount:VDQHSD (match_operand:VDQHSD 1 "register_operand")))] "TARGET_SIMD" { + if (TARGET_SVE) + { + rtx p = aarch64_ptrue_reg (<VPRED>mode); + emit_insn (gen_aarch64_pred_popcount<mode> (operands[0], + p, + operands[1])); + DONE; + } + /* Generate a byte popcount. */ machine_mode mode = <bitsize> == 64 ? V8QImode : V16QImode; rtx tmp = gen_reg_rtx (mode); diff --git a/gcc/config/aarch64/aarch64-sve.md b/gcc/config/aarch64/aarch64-sve.md index c3ed5075c4ed..a5cd42be9d5c 100644 --- a/gcc/config/aarch64/aarch64-sve.md +++ b/gcc/config/aarch64/aarch64-sve.md @@ -3104,16 +3104,16 @@ ;; Integer unary arithmetic predicated with a PTRUE. (define_insn "@aarch64_pred_<optab><mode>" - [(set (match_operand:SVE_I 0 "register_operand") - (unspec:SVE_I + [(set (match_operand:SVE_VDQ_I 0 "register_operand") + (unspec:SVE_VDQ_I [(match_operand:<VPRED> 1 "register_operand") - (SVE_INT_UNARY:SVE_I - (match_operand:SVE_I 2 "register_operand"))] + (SVE_INT_UNARY:SVE_VDQ_I + (match_operand:SVE_VDQ_I 2 "register_operand"))] UNSPEC_PRED_X))] "TARGET_SVE" {@ [ cons: =0 , 1 , 2 ; attrs: movprfx ] - [ w , Upl , 0 ; * ] <sve_int_op>\t%0.<Vetype>, %1/m, %2.<Vetype> - [ ?&w , Upl , w ; yes ] movprfx\t%0, %2\;<sve_int_op>\t%0.<Vetype>, %1/m, %2.<Vetype> + [ w , Upl , 0 ; * ] <sve_int_op>\t%Z0.<Vetype>, %1/m, %Z2.<Vetype> + [ ?&w , Upl , w ; yes ] movprfx\t%Z0, %Z2\;<sve_int_op>\t%Z0.<Vetype>, %1/m, %Z2.<Vetype> } ) @@ -3168,6 +3168,7 @@ } ) + ;; ------------------------------------------------------------------------- ;; ---- [INT] General unary arithmetic corresponding to unspecs ;; ------------------------------------------------------------------------- diff --git a/gcc/config/aarch64/iterators.md b/gcc/config/aarch64/iterators.md index 95fe8f070f4c..aaa4afefe2ce 100644 --- a/gcc/config/aarch64/iterators.md +++ b/gcc/config/aarch64/iterators.md @@ -559,6 +559,9 @@ ;; element modes (define_mode_iterator SVE_I_SIMD_DI [SVE_I V2DI]) +;; All SVE and Advanced SIMD integer vector modes. +(define_mode_iterator SVE_VDQ_I [SVE_I VDQ_I]) + ;; SVE integer vector modes whose elements are 16 bits or wider. (define_mode_iterator SVE_HSDI [VNx8HI VNx4HI VNx2HI VNx4SI VNx2SI @@ -2278,6 +2281,8 @@ (VNx32BF "VNx8BI") (VNx16SI "VNx4BI") (VNx16SF "VNx4BI") (VNx8DI "VNx2BI") (VNx8DF "VNx2BI") + (V8QI "VNx8BI") (V16QI "VNx16BI") + (V4HI "VNx4BI") (V8HI "VNx8BI") (V2SI "VNx2BI") (V4SI "VNx4BI") (V2DI "VNx2BI")]) ;; ...and again in lower case. diff --git a/gcc/testsuite/gcc.target/aarch64/popcnt-sve.c b/gcc/testsuite/gcc.target/aarch64/popcnt-sve.c new file mode 100644 index 000000000000..8e349efe3907 --- /dev/null +++ b/gcc/testsuite/gcc.target/aarch64/popcnt-sve.c @@ -0,0 +1,88 @@ +/* { dg-do compile } */ +/* { dg-options "-O2 -march=armv8.2-a+sve -fno-vect-cost-model -fno-schedule-insns -fno-schedule-insns2" } */ +/* { dg-final { check-function-bodies "**" "" "" } } */ + +/* +** f_v4hi: +** ptrue (p[0-7]).b, all +** ldr d([0-9]+), \[x0\] +** cnt z\2.h, \1/m, z\2.h +** str d\2, \[x1\] +** ret +*/ +void +f_v4hi (unsigned short *__restrict b, unsigned short *__restrict d) +{ + d[0] = __builtin_popcount (b[0]); + d[1] = __builtin_popcount (b[1]); + d[2] = __builtin_popcount (b[2]); + d[3] = __builtin_popcount (b[3]); +} + +/* +** f_v8hi: +** ptrue (p[0-7]).b, all +** ldr q([0-9]+), \[x0\] +** cnt z\2.h, \1/m, z\2.h +** str q\2, \[x1\] +** ret +*/ +void +f_v8hi (unsigned short *__restrict b, unsigned short *__restrict d) +{ + d[0] = __builtin_popcount (b[0]); + d[1] = __builtin_popcount (b[1]); + d[2] = __builtin_popcount (b[2]); + d[3] = __builtin_popcount (b[3]); + d[4] = __builtin_popcount (b[4]); + d[5] = __builtin_popcount (b[5]); + d[6] = __builtin_popcount (b[6]); + d[7] = __builtin_popcount (b[7]); +} + +/* +** f_v2si: +** ptrue (p[0-7]).b, all +** ldr d([0-9]+), \[x0\] +** cnt z\2.s, \1/m, z\2.s +** str d\2, \[x1\] +** ret +*/ +void +f_v2si (unsigned int *__restrict b, unsigned int *__restrict d) +{ + d[0] = __builtin_popcount (b[0]); + d[1] = __builtin_popcount (b[1]); +} + +/* +** f_v4si: +** ptrue (p[0-7]).b, all +** ldr q([0-9]+), \[x0\] +** cnt z\2.s, \1/m, z\2.s +** str q\2, \[x1\] +** ret +*/ +void +f_v4si (unsigned int *__restrict b, unsigned int *__restrict d) +{ + d[0] = __builtin_popcount (b[0]); + d[1] = __builtin_popcount (b[1]); + d[2] = __builtin_popcount (b[2]); + d[3] = __builtin_popcount (b[3]); +} + +/* +** f_v2di: +** ptrue (p[0-7]).b, all +** ldr q([0-9]+), \[x0\] +** cnt z\2.d, \1/m, z\2.d +** str q\2, \[x1\] +** ret +*/ +void +f_v2di (unsigned long *__restrict b, unsigned long *__restrict d) +{ + d[0] = __builtin_popcountll (b[0]); + d[1] = __builtin_popcountll (b[1]); +}