[llvm][SVE] Reg + reg addressing mode for LD1RO.

Reviewers: efriedma, sdesmalen

Reviewed By: efriedma

Subscribers: tschuett, hiraditya, rkruppe, psnobl, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D80741

diff --git a/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td b/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td
index d6e52f9..9bab0e4 100644 (file)
--- a/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td
+++ b/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td
@@ -1960,10 +1960,10 @@ let Predicates = [HasSVE, HasMatMulFP64] in {
   defm LD1RO_H_IMM : sve_mem_ldor_si<0b01, "ld1roh", Z_h, ZPR16, nxv8i16, nxv8i1,  AArch64ld1ro>;
   defm LD1RO_W_IMM : sve_mem_ldor_si<0b10, "ld1row", Z_s, ZPR32, nxv4i32, nxv4i1,  AArch64ld1ro>;
   defm LD1RO_D_IMM : sve_mem_ldor_si<0b11, "ld1rod", Z_d, ZPR64, nxv2i64, nxv2i1,  AArch64ld1ro>;
-  defm LD1RO_B     : sve_mem_ldor_ss<0b00, "ld1rob", Z_b, ZPR8,  GPR64NoXZRshifted8>;
-  defm LD1RO_H     : sve_mem_ldor_ss<0b01, "ld1roh", Z_h, ZPR16, GPR64NoXZRshifted16>;
-  defm LD1RO_W     : sve_mem_ldor_ss<0b10, "ld1row", Z_s, ZPR32, GPR64NoXZRshifted32>;
-  defm LD1RO_D     : sve_mem_ldor_ss<0b11, "ld1rod", Z_d, ZPR64, GPR64NoXZRshifted64>;
+  defm LD1RO_B     : sve_mem_ldor_ss<0b00, "ld1rob", Z_b, ZPR8,  GPR64NoXZRshifted8,  nxv16i8, nxv16i1, AArch64ld1ro, am_sve_regreg_lsl0>;
+  defm LD1RO_H     : sve_mem_ldor_ss<0b01, "ld1roh", Z_h, ZPR16, GPR64NoXZRshifted16, nxv8i16, nxv8i1,  AArch64ld1ro, am_sve_regreg_lsl1>;
+  defm LD1RO_W     : sve_mem_ldor_ss<0b10, "ld1row", Z_s, ZPR32, GPR64NoXZRshifted32, nxv4i32, nxv4i1,  AArch64ld1ro, am_sve_regreg_lsl2>;
+  defm LD1RO_D     : sve_mem_ldor_ss<0b11, "ld1rod", Z_d, ZPR64, GPR64NoXZRshifted64, nxv2i64, nxv2i1,  AArch64ld1ro, am_sve_regreg_lsl3>;
   defm ZIP1_ZZZ_Q  : sve_int_perm_bin_perm_128_zz<0b00, 0, "zip1", int_aarch64_sve_zip1q>;
   defm ZIP2_ZZZ_Q  : sve_int_perm_bin_perm_128_zz<0b00, 1, "zip2", int_aarch64_sve_zip2q>;
   defm UZP1_ZZZ_Q  : sve_int_perm_bin_perm_128_zz<0b01, 0, "uzp1", int_aarch64_sve_uzp1q>;

diff --git a/llvm/lib/Target/AArch64/SVEInstrFormats.td b/llvm/lib/Target/AArch64/SVEInstrFormats.td
index bde17b3..01e6696 100644 (file)
--- a/llvm/lib/Target/AArch64/SVEInstrFormats.td
+++ b/llvm/lib/Target/AArch64/SVEInstrFormats.td
@@ -7699,11 +7699,15 @@ class sve_mem_ldor_ss<bits<2> sz, string asm, RegisterOperand VecList,
 }
 
 multiclass sve_mem_ldor_ss<bits<2> sz, string asm, RegisterOperand listty,
-                           ZPRRegOp zprty, RegisterOperand gprty> {
+                           ZPRRegOp zprty, RegisterOperand gprty, ValueType Ty,
+                           ValueType PredTy, SDNode Ld1ro, ComplexPattern AddrCP> {
   def NAME : sve_mem_ldor_ss<sz, asm, listty, gprty>;
 
   def : InstAlias<asm # "\t$Zt, $Pg/z, [$Rn, $Rm]",
                   (!cast<Instruction>(NAME) zprty:$Zt, PPR3bAny:$Pg, GPR64sp:$Rn, gprty:$Rm), 0>;
+
+    def : Pat<(Ty (Ld1ro (PredTy PPR3bAny:$gp), (AddrCP GPR64sp:$base, gprty:$offset))),
+              (!cast<Instruction>(NAME) PPR3bAny:$gp, GPR64sp:$base, gprty:$offset)>;
 }
 
 //===----------------------------------------------------------------------===//

diff --git a/llvm/test/CodeGen/AArch64/sve-intrinsics-ld1ro-addressing-mode-reg-reg.ll b/llvm/test/CodeGen/AArch64/sve-intrinsics-ld1ro-addressing-mode-reg-reg.ll
new file mode 100644 (file)
index 0000000..b4ac587
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/sve-intrinsics-ld1ro-addressing-mode-reg-reg.ll
@@ -0,0 +1,102 @@
+; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve,+f64mm,+bf16 -asm-verbose=0 < %s | FileCheck %s
+
+;
+; LD1ROB
+;
+
+define <vscale x 16 x i8> @ld1rob_i8(<vscale x 16 x i1> %pg, i8* %a, i64 %index) nounwind {
+; CHECK-LABEL: ld1rob_i8:
+; CHECK-NEXT:  ld1rob { z0.b }, p0/z, [x0, x1]
+; CHECK-NEXT:  ret
+  %base = getelementptr i8, i8* %a, i64 %index
+  %load = call <vscale x 16 x i8> @llvm.aarch64.sve.ld1ro.nxv16i8(<vscale x 16 x i1> %pg, i8* %base)
+  ret <vscale x 16 x i8> %load
+}
+
+;
+; LD1ROH
+;
+
+define <vscale x 8 x i16> @ld1roh_i16(<vscale x 8 x i1> %pg, i16* %a, i64 %index) nounwind {
+; CHECK-LABEL: ld1roh_i16:
+; CHECK-NEXT:  ld1roh { z0.h }, p0/z, [x0, x1, lsl #1]
+; CHECK-NEXT:  ret
+  %base = getelementptr i16, i16* %a, i64 %index
+  %load = call <vscale x 8 x i16> @llvm.aarch64.sve.ld1ro.nxv8i16(<vscale x 8 x i1> %pg, i16* %base)
+  ret <vscale x 8 x i16> %load
+}
+
+define <vscale x 8 x half> @ld1roh_f16(<vscale x 8 x i1> %pg, half* %a, i64 %index) nounwind {
+; CHECK-LABEL: ld1roh_f16:
+; CHECK-NEXT:  ld1roh { z0.h }, p0/z, [x0, x1, lsl #1]
+; CHECK-NEXT:  ret
+  %base = getelementptr half, half* %a, i64 %index
+  %load = call <vscale x 8 x half> @llvm.aarch64.sve.ld1ro.nxv8f16(<vscale x 8 x i1> %pg, half* %base)
+  ret <vscale x 8 x half> %load
+}
+
+; bfloat - requires -mattr=+bf16
+define <vscale x 8 x bfloat> @ld1roh_bf16(<vscale x 8 x i1> %pg, bfloat* %a, i64 %index) nounwind {
+; CHECK-LABEL: ld1roh_bf16:
+; CHECK-NEXT:  ld1roh { z0.h }, p0/z, [x0, x1, lsl #1]
+; CHECK-NEXT:  ret
+  %base = getelementptr bfloat, bfloat* %a, i64 %index
+  %load = call <vscale x 8 x bfloat> @llvm.aarch64.sve.ld1ro.nxv8bf16(<vscale x 8 x i1> %pg, bfloat* %base)
+  ret <vscale x 8 x bfloat> %load
+}
+
+;
+; LD1ROW
+;
+
+define<vscale x 4 x i32> @ld1row_i32(<vscale x 4 x i1> %pg, i32* %a, i64 %index) nounwind {
+; CHECK-LABEL: ld1row_i32:
+; CHECK-NEXT:  ld1row { z0.s }, p0/z, [x0, x1, lsl #2]
+; CHECK-NEXT:  ret
+  %base = getelementptr i32, i32* %a, i64 %index
+  %load = call <vscale x 4 x i32> @llvm.aarch64.sve.ld1ro.nxv4i32(<vscale x 4 x i1> %pg, i32* %base)
+  ret <vscale x 4 x i32> %load
+}
+
+define<vscale x 4 x float> @ld1row_f32(<vscale x 4 x i1> %pg, float* %a, i64 %index) nounwind {
+; CHECK-LABEL: ld1row_f32:
+; CHECK-NEXT:  ld1row { z0.s }, p0/z, [x0, x1, lsl #2]
+; CHECK-NEXT:  ret
+  %base = getelementptr float, float* %a, i64 %index
+  %load = call <vscale x 4 x float> @llvm.aarch64.sve.ld1ro.nxv4f32(<vscale x 4 x i1> %pg, float* %base)
+  ret <vscale x 4 x float> %load
+}
+
+;
+; LD1ROD
+;
+
+define <vscale x 2 x i64> @ld1rod_i64(<vscale x 2 x i1> %pg, i64* %a, i64 %index) nounwind {
+; CHECK-LABEL: ld1rod_i64:
+; CHECK-NEXT:  ld1rod { z0.d }, p0/z, [x0, x1, lsl #3]
+; CHECK-NEXT:  ret
+  %base = getelementptr i64, i64* %a, i64 %index
+  %load = call <vscale x 2 x i64> @llvm.aarch64.sve.ld1ro.nxv2i64(<vscale x 2 x i1> %pg, i64* %base)
+  ret <vscale x 2 x i64> %load
+}
+
+define <vscale x 2 x double> @ld1rod_f64(<vscale x 2 x i1> %pg, double* %a, i64 %index) nounwind {
+; CHECK-LABEL: ld1rod_f64:
+; CHECK-NEXT:  ld1rod { z0.d }, p0/z, [x0, x1, lsl #3]
+; CHECK-NEXT:  ret
+  %base = getelementptr double, double* %a, i64 %index
+  %load = call <vscale x 2 x double> @llvm.aarch64.sve.ld1ro.nxv2f64(<vscale x 2 x i1> %pg, double* %base)
+  ret <vscale x 2 x double> %load
+}
+
+declare <vscale x 16 x i8> @llvm.aarch64.sve.ld1ro.nxv16i8(<vscale x 16 x i1>, i8*)
+
+declare <vscale x 8 x i16> @llvm.aarch64.sve.ld1ro.nxv8i16(<vscale x 8 x i1>, i16*)
+declare <vscale x 8 x half> @llvm.aarch64.sve.ld1ro.nxv8f16(<vscale x 8 x i1>, half*)
+declare <vscale x 8 x bfloat> @llvm.aarch64.sve.ld1ro.nxv8bf16(<vscale x 8 x i1>, bfloat*)
+
+declare <vscale x 4 x i32> @llvm.aarch64.sve.ld1ro.nxv4i32(<vscale x 4 x i1>, i32*)
+declare <vscale x 4 x float> @llvm.aarch64.sve.ld1ro.nxv4f32(<vscale x 4 x i1>, float*)
+
+declare <vscale x 2 x i64> @llvm.aarch64.sve.ld1ro.nxv2i64(<vscale x 2 x i1>, i64*)
+declare <vscale x 2 x double> @llvm.aarch64.sve.ld1ro.nxv2f64(<vscale x 2 x i1>, double*)