================
@@ -5617,9 +5648,13 @@ class sme2_fp16_quarter_tile_outer_product<bit M, bit N,
bit S, string mnemonic,
let Constraints = "$ZAda = $_ZAda";
}
-multiclass sme2_fmop4as_fp16_non_widening<bit S, string mnemonic> {
+multiclass sme2_fmop4as_fp16_non_widening<bit S, string mnemonic, string op> {
// Single vectors
- def _MZZ_H : sme2_fp16_quarter_tile_outer_product<0, 0, S, mnemonic,
ZPR16Mul2_Lo, ZPR16Mul2_Hi>;
+ def _MZZ_H : sme2_fp16_quarter_tile_outer_product<0, 0, S, mnemonic,
ZPR16Mul2_Lo, ZPR16Mul2_Hi>, SMEPseudo2Instr<NAME # _MZZ_H, 1>;
+
+ def NAME # _MZZ_H # _PSEUDO :
sme2_quarter_tile_outer_product_pseudo_single_single<ZPR16Mul2_Lo,
ZPR16Mul2_Hi, SMEMatrixTileH>, SMEPseudo2Instr<NAME # _MZZ_H, 0>;
+
+ def : SME2_ZA_Tile_TwoVec_Pat<NAME # _MZZ_H, !cast<SDPatternOperator>(op #
"_1x1"), timm32_0_3, nxv8f16>;
----------------
CarolineConcatto wrote:
I believe that here timm32_0_3 should be timm32_0_1.
Half-precision, single vectors" variants: is the name of the ZA tile ZA0-ZA1,
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
