[NFC][LLVM][CodeGen][SVE] Restructure urshr related PatFrags.

[paulwalker-arm]

The current structure is not broken but doesn't follow the expected pattern and thus could be used erroneously. To fix this I have separated the intrinsic (i.e. merging) operation from the ISD (i.e. pred) operation. I would have liked AArch64ursra to include the intrinsic (likewise to have created AArch64srsra) but currently we cannot mix intrinsic references and anonymous SVEAllActive expressions within a PatFrag because the predicate type cannot be inferred.

For the real instruction's PatFrag I have renamed (dropped the i) to match the existing naming convention. If the preference is to keep it then we'll clash with an existing NEON node, which is fine because then I'll follow the style of adding _x to signify unpredicted instructions that clash with NEON. I'm happy either way so just shout if you prefer the latter.

NOTE: AArch64ursra could use SVEAnyPredicate but there are no non-all-active uses and so to maintain the patch's NFCness I stuck with using SVEAllActive.

[llvmbot]

@llvm/pr-subscribers-backend-aarch64

Author: Paul Walker (paulwalker-arm)

Changes

The current structure is not broken but doesn't follow the expected pattern and thus could be used erroneously. To fix this I have separated the intrinsic (i.e. merging) operation from the ISD (i.e. pred) operation. I would have liked AArch64ursra to include the intrinsic (likewise to have created AArch64srsra) but currently we cannot mix intrinsic references and anonymous SVEAllActive expressions within a PatFrag because the predicate type cannot be inferred.

For the real instruction's PatFrag I have renamed (dropped the i) to match the existing naming convention. If the preference is to keep it then we'll clash with an existing NEON node, which is fine because then I'll follow the style of adding _x to signify unpredicted instructions that clash with NEON. I'm happy either way so just shout if you prefer the latter.

NOTE: AArch64ursra could use SVEAnyPredicate but there are no non-all-active uses and so to maintain the patch's NFCness I stuck with using SVEAllActive.

---

Full diff: https://github.com/llvm/llvm-project/pull/170521.diff

1 Files Affected:

• (modified) llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td (+11-7)

diff --git a/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td b/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td
index e99b3f8ff07e0..37de22940afc2 100644
--- a/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td
+++ b/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td
@@ -262,15 +262,15 @@ def AArch64fmaxnm_p_nnan : PatFrag<(ops node:$op1, node:$op2, node:$op3),
 
 def SDT_AArch64Arith_Imm : SDTypeProfile<1, 3, [
   SDTCisVec<0>, SDTCisVec<1>, SDTCisVec<2>, SDTCisVT<3,i32>,
-  SDTCVecEltisVT<1,i1>, SDTCisSameAs<0,2>
+  SDTCVecEltisVT<1,i1>, SDTCisSameNumEltsAs<0,1>, SDTCisSameAs<0,2>
 ]>;
 
 def AArch64asrd_m1 : SDNode<"AArch64ISD::ASRD_MERGE_OP1", SDT_AArch64Arith_Imm>;
-def AArch64urshri_p_node : SDNode<"AArch64ISD::URSHR_I_PRED", SDT_AArch64Arith_Imm>;
+def AArch64urshri_p : SDNode<"AArch64ISD::URSHR_I_PRED", SDT_AArch64Arith_Imm>;
 
-def AArch64urshri_p : PatFrags<(ops node:$op1, node:$op2, node:$op3),
-                           [(int_aarch64_sve_urshr node:$op1, node:$op2, node:$op3),
-                            (AArch64urshri_p_node node:$op1, node:$op2, node:$op3)]>;
+def AArch64urshr : PatFrags<(ops node:$op1, node:$op2, node:$op3),
+                            [(int_aarch64_sve_urshr node:$op1, node:$op2, node:$op3),
+                             (AArch64urshri_p node:$op1, node:$op2, node:$op3)]>;
 
 def SDT_AArch64IntExtend : SDTypeProfile<1, 4, [
   SDTCisVec<0>, SDTCisVec<1>, SDTCisVec<2>, SDTCisVT<3, OtherVT>, SDTCisVec<4>,
@@ -355,6 +355,10 @@ def AArch64ssra : PatFrags<(ops node:$op1, node:$op2, node:$op3),
                            [(int_aarch64_sve_ssra node:$op1, node:$op2, node:$op3),
                             (add node:$op1, (AArch64asr_p (SVEAnyPredicate), node:$op2, (SVEShiftSplatImmR (i32 node:$op3))))]>;
 
+def AArch64ursra : PatFrags<(ops node:$op1, node:$op2, node:$op3),
+                            [(int_aarch64_sve_ursra node:$op1, node:$op2, node:$op3),
+                             (add node:$op1, (AArch64urshri_p (SVEAllActive), node:$op2, node:$op3))]>;
+
 // Replace pattern min(max(v1,v2),v3) by clamp
 def AArch64sclamp : PatFrags<(ops node:$Zd, node:$Zn, node:$Zm),
                               [(int_aarch64_sve_sclamp node:$Zd, node:$Zn, node:$Zm),
@@ -3906,7 +3910,7 @@ let Predicates = [HasSVE2_or_SME] in {
   defm SQSHL_ZPmI  : sve_int_bin_pred_shift_imm_left_dup<0b0110, "sqshl",  "SQSHL_ZPZI",  int_aarch64_sve_sqshl>;
   defm UQSHL_ZPmI  : sve_int_bin_pred_shift_imm_left_dup<0b0111, "uqshl",  "UQSHL_ZPZI",  int_aarch64_sve_uqshl>;
   defm SRSHR_ZPmI  : sve_int_bin_pred_shift_imm_right<   0b1100, "srshr",  "SRSHR_ZPZI",  int_aarch64_sve_srshr>;
-  defm URSHR_ZPmI  : sve_int_bin_pred_shift_imm_right<   0b1101, "urshr",  "URSHR_ZPZI",  AArch64urshri_p>;
+  defm URSHR_ZPmI  : sve_int_bin_pred_shift_imm_right<   0b1101, "urshr",  "URSHR_ZPZI",  AArch64urshr>;
   defm SQSHLU_ZPmI : sve_int_bin_pred_shift_imm_left<    0b1111, "sqshlu", "SQSHLU_ZPZI", int_aarch64_sve_sqshlu>;
 
   // SVE2 integer add/subtract long
@@ -3964,7 +3968,7 @@ let Predicates = [HasSVE2_or_SME] in {
   defm SSRA_ZZI  : sve2_int_bin_accum_shift_imm_right<0b00, "ssra",  AArch64ssra>;
   defm USRA_ZZI  : sve2_int_bin_accum_shift_imm_right<0b01, "usra",  AArch64usra>;
   defm SRSRA_ZZI : sve2_int_bin_accum_shift_imm_right<0b10, "srsra", int_aarch64_sve_srsra, int_aarch64_sve_srshr>;
-  defm URSRA_ZZI : sve2_int_bin_accum_shift_imm_right<0b11, "ursra", int_aarch64_sve_ursra, AArch64urshri_p>;
+  defm URSRA_ZZI : sve2_int_bin_accum_shift_imm_right<0b11, "ursra", AArch64ursra, int_aarch64_sve_urshr>;
 
   // SVE2 complex integer add
   defm CADD_ZZI   : sve2_int_cadd<0b0, "cadd",   int_aarch64_sve_cadd_x>;