================
@@ -5763,9 +5806,13 @@ class sme2_fp64_quarter_tile_outer_product<bit M, bit N,
bit S, string mnemonic,
let Constraints = "$ZAda = $_ZAda";
}
-multiclass sme2_fmop4as_fp64_non_widening<bit S, string mnemonic> {
+multiclass sme2_fmop4as_fp64_non_widening<bit S, string mnemonic, string op> {
// Single vectors
- def _MZZ_D : sme2_fp64_quarter_tile_outer_product<0, 0, S, mnemonic,
ZPR64Mul2_Lo, ZPR64Mul2_Hi>;
+ def _MZZ_D : sme2_fp64_quarter_tile_outer_product<0, 0, S, mnemonic,
ZPR64Mul2_Lo, ZPR64Mul2_Hi>, SMEPseudo2Instr<NAME # _MZZ_D, 1>;
+
+ def NAME # _MZZ_D # _PSEUDO :
sme2_quarter_tile_outer_product_pseudo_single_single<ZPR64Mul2_Lo,
ZPR64Mul2_Hi, SMEMatrixTileD>, SMEPseudo2Instr<NAME # _MZZ_D, 0>;
+
+ def : SME2_ZA_Tile_TwoVec_Pat<NAME # _MZZ_D, !cast<SDPatternOperator>(op #
"_1x1"), timm32_0_3, nxv2f64>;
----------------
CarolineConcatto wrote:
I believe that here timm32_0_3 should be timm32_0_7.
"Double-precision, single vectors" variants: is the name of the ZA tile
ZA0-ZA7, encoded in the "ZAda" field.
From:
https://developer.arm.com/documentation/ddi0602/2024-12/SME-Instructions/FMOP4A--non-widening---Floating-point-quarter-tile-outer-products--accumulating-?lang=en
https://github.com/llvm/llvm-project/pull/127797
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
