[llvm] [AArch64][SME] Create separate FORM_TRANSPOSE pseudos for ZPR & ZPRMul classes (PR #123755)
Kerry McLaughlin via llvm-commits
llvm-commits at lists.llvm.org
Tue Jan 21 06:35:49 PST 2025
https://github.com/kmclaughlin-arm created https://github.com/llvm/llvm-project/pull/123755
The `FORM_TRANSPOSED_REG_TUPLE` pseudo nodes use either the `ZPR2Mul2` or
`ZPR4Mul4` register classes for output. To extend these to other multi-vector
intrinsics which instead create a `ZPR2`/`ZPR4` register sequence, a new
pseudo has been added and the existing one renamed.
>From d1c9a3a109c78f3eeaf3e1688c7fa223cbb42f84 Mon Sep 17 00:00:00 2001
From: Kerry McLaughlin <kerry.mclaughlin at arm.com>
Date: Thu, 16 Jan 2025 14:20:56 +0000
Subject: [PATCH 1/2] Add tests for udot (multi, single) using FORM_TRANSPOSED
pseudos
---
.../AArch64/sme2-intrinsics-int-dots.ll | 511 ++++++++++++++++++
1 file changed, 511 insertions(+)
diff --git a/llvm/test/CodeGen/AArch64/sme2-intrinsics-int-dots.ll b/llvm/test/CodeGen/AArch64/sme2-intrinsics-int-dots.ll
index 86ed63d743713c..df422df5c13c81 100644
--- a/llvm/test/CodeGen/AArch64/sme2-intrinsics-int-dots.ll
+++ b/llvm/test/CodeGen/AArch64/sme2-intrinsics-int-dots.ll
@@ -319,6 +319,39 @@ define void @udot_single_za32_u16_vg1x2(i32 %slice, <vscale x 16 x i8> %unused,
ret void
}
+define void @udot_single_za32_u16_vg1x2_tuple(ptr %ptr, i64 %stride, <vscale x 8 x i16> %zn) #0 {
+; CHECK-LABEL: udot_single_za32_u16_vg1x2_tuple:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill
+; CHECK-NEXT: addvl sp, sp, #-1
+; CHECK-NEXT: str p8, [sp, #7, mul vl] // 2-byte Folded Spill
+; CHECK-NEXT: ptrue pn8.b
+; CHECK-NEXT: add x9, x0, x1
+; CHECK-NEXT: ld1h { z2.h, z3.h }, pn8/z, [x0]
+; CHECK-NEXT: ld1h { z6.h, z7.h }, pn8/z, [x9]
+; CHECK-NEXT: mov w8, wzr
+; CHECK-NEXT: mov z5.d, z2.d
+; CHECK-NEXT: mov z4.d, z7.d
+; CHECK-NEXT: udot za.s[w8, 0, vgx2], { z5.h, z6.h }, z0.h
+; CHECK-NEXT: udot za.s[w8, 0, vgx2], { z3.h, z4.h }, z0.h
+; CHECK-NEXT: ldr p8, [sp, #7, mul vl] // 2-byte Folded Reload
+; CHECK-NEXT: addvl sp, sp, #1
+; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload
+; CHECK-NEXT: ret
+entry:
+ %0 = tail call target("aarch64.svcount") @llvm.aarch64.sve.ptrue.c8()
+ %1 = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld1.pn.x2.nxv8i16(target("aarch64.svcount") %0, ptr %ptr)
+ %2 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } %1, 0
+ %3 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } %1, 1
+ %arrayidx2 = getelementptr inbounds i8, ptr %ptr, i64 %stride
+ %4 = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld1.pn.x2.nxv8i16(target("aarch64.svcount") %0, ptr %arrayidx2)
+ %5 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } %4, 0
+ %6 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } %4, 1
+ call void @llvm.aarch64.sme.udot.single.za32.vg1x2.nxv8i16(i32 0, <vscale x 8 x i16> %2, <vscale x 8 x i16> %5, <vscale x 8 x i16> %zn)
+ call void @llvm.aarch64.sme.udot.single.za32.vg1x2.nxv8i16(i32 0, <vscale x 8 x i16> %3, <vscale x 8 x i16> %6, <vscale x 8 x i16> %zn)
+ ret void
+}
+
define void @udot_single_za32_u16_vg1x4(i32 %slice, <vscale x 16 x i8> %unused, <vscale x 8 x i16> %zn0, <vscale x 8 x i16> %zn1, <vscale x 8 x i16> %zn2, <vscale x 8 x i16> %zn3, <vscale x 8 x i16> %zn4) #0 {
; CHECK-LABEL: udot_single_za32_u16_vg1x4:
; CHECK: // %bb.0:
@@ -332,6 +365,102 @@ define void @udot_single_za32_u16_vg1x4(i32 %slice, <vscale x 16 x i8> %unused,
ret void
}
+define void @udot_single_za32_u16_vg1x4_tuple(ptr %ptr, i64 %stride, <vscale x 8 x i16> %zn) #0 {
+; CHECK-LABEL: udot_single_za32_u16_vg1x4_tuple:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill
+; CHECK-NEXT: addvl sp, sp, #-13
+; CHECK-NEXT: add x10, x1, x1, lsl #1
+; CHECK-NEXT: str p8, [sp, #7, mul vl] // 2-byte Folded Spill
+; CHECK-NEXT: ptrue pn8.b
+; CHECK-NEXT: str z19, [sp, #1, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: add x9, x0, x1
+; CHECK-NEXT: mov w8, wzr
+; CHECK-NEXT: str z18, [sp, #2, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z17, [sp, #3, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z16, [sp, #4, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z15, [sp, #5, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z14, [sp, #6, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z13, [sp, #7, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z12, [sp, #8, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z11, [sp, #9, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z10, [sp, #10, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z9, [sp, #11, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z8, [sp, #12, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: ld1h { z4.h - z7.h }, pn8/z, [x0]
+; CHECK-NEXT: ld1h { z28.h - z31.h }, pn8/z, [x9]
+; CHECK-NEXT: add x9, x0, x10
+; CHECK-NEXT: ld1h { z24.h - z27.h }, pn8/z, [x0, x1, lsl #1]
+; CHECK-NEXT: ld1h { z8.h - z11.h }, pn8/z, [x9]
+; CHECK-NEXT: mov z1.d, z4.d
+; CHECK-NEXT: mov z2.d, z28.d
+; CHECK-NEXT: mov z12.d, z5.d
+; CHECK-NEXT: mov z3.d, z24.d
+; CHECK-NEXT: mov z4.d, z8.d
+; CHECK-NEXT: mov z13.d, z29.d
+; CHECK-NEXT: mov z14.d, z25.d
+; CHECK-NEXT: mov z15.d, z9.d
+; CHECK-NEXT: mov z16.d, z6.d
+; CHECK-NEXT: mov z17.d, z30.d
+; CHECK-NEXT: mov z18.d, z26.d
+; CHECK-NEXT: mov z19.d, z10.d
+; CHECK-NEXT: mov z8.d, z7.d
+; CHECK-NEXT: udot za.s[w8, 0, vgx4], { z1.h - z4.h }, z0.h
+; CHECK-NEXT: mov z9.d, z31.d
+; CHECK-NEXT: mov z10.d, z27.d
+; CHECK-NEXT: udot za.s[w8, 0, vgx4], { z12.h - z15.h }, z0.h
+; CHECK-NEXT: udot za.s[w8, 0, vgx4], { z16.h - z19.h }, z0.h
+; CHECK-NEXT: udot za.s[w8, 0, vgx4], { z8.h - z11.h }, z0.h
+; CHECK-NEXT: ldr z19, [sp, #1, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z18, [sp, #2, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z17, [sp, #3, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z16, [sp, #4, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z15, [sp, #5, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z14, [sp, #6, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z13, [sp, #7, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z12, [sp, #8, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z11, [sp, #9, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z10, [sp, #10, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z9, [sp, #11, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z8, [sp, #12, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr p8, [sp, #7, mul vl] // 2-byte Folded Reload
+; CHECK-NEXT: addvl sp, sp, #13
+; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload
+; CHECK-NEXT: ret
+entry:
+ %0 = tail call target("aarch64.svcount") @llvm.aarch64.sve.ptrue.c8()
+ %1 = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld1.pn.x4.nxv8i16(target("aarch64.svcount") %0, ptr %ptr)
+ %2 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %1, 0
+ %3 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %1, 1
+ %4 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %1, 2
+ %5 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %1, 3
+ %arrayidx2 = getelementptr inbounds i8, ptr %ptr, i64 %stride
+ %6 = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld1.pn.x4.nxv8i16(target("aarch64.svcount") %0, ptr %arrayidx2)
+ %7 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %6, 0
+ %8 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %6, 1
+ %9 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %6, 2
+ %10 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %6, 3
+ %mul3 = shl i64 %stride, 1
+ %arrayidx4 = getelementptr inbounds i8, ptr %ptr, i64 %mul3
+ %11 = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld1.pn.x4.nxv8i16(target("aarch64.svcount") %0, ptr %arrayidx4)
+ %12 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %11, 0
+ %13 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %11, 1
+ %14 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %11, 2
+ %15 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %11, 3
+ %mul5 = mul i64 %stride, 3
+ %arrayidx6 = getelementptr inbounds i8, ptr %ptr, i64 %mul5
+ %16 = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld1.pn.x4.nxv8i16(target("aarch64.svcount") %0, ptr %arrayidx6)
+ %17 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %16, 0
+ %18 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %16, 1
+ %19 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %16, 2
+ %20 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %16, 3
+ call void @llvm.aarch64.sme.udot.single.za32.vg1x4.nxv8i16(i32 0, <vscale x 8 x i16> %2, <vscale x 8 x i16> %7, <vscale x 8 x i16> %12, <vscale x 8 x i16> %17, <vscale x 8 x i16> %zn)
+ call void @llvm.aarch64.sme.udot.single.za32.vg1x4.nxv8i16(i32 0, <vscale x 8 x i16> %3, <vscale x 8 x i16> %8, <vscale x 8 x i16> %13, <vscale x 8 x i16> %18, <vscale x 8 x i16> %zn)
+ call void @llvm.aarch64.sme.udot.single.za32.vg1x4.nxv8i16(i32 0, <vscale x 8 x i16> %4, <vscale x 8 x i16> %9, <vscale x 8 x i16> %14, <vscale x 8 x i16> %19, <vscale x 8 x i16> %zn)
+ call void @llvm.aarch64.sme.udot.single.za32.vg1x4.nxv8i16(i32 0, <vscale x 8 x i16> %5, <vscale x 8 x i16> %10, <vscale x 8 x i16> %15, <vscale x 8 x i16> %20, <vscale x 8 x i16> %zn)
+ ret void
+}
+
define void @udot_single_za32_u8_vg1x2(i32 %slice, <vscale x 16 x i8> %unused, <vscale x 16 x i8> %zn0, <vscale x 16 x i8> %zn1, <vscale x 16 x i8> %zn2) #0 {
; CHECK-LABEL: udot_single_za32_u8_vg1x2:
; CHECK: // %bb.0:
@@ -397,6 +526,38 @@ define void @usdot_single_za32_u8_vg1x2(i32 %slice, <vscale x 16 x i8> %unused,
ret void
}
+define void @usdot_single_za32_u16_vg1x2_tuple(ptr %ptr, i64 %stride, <vscale x 16 x i8> %zn) #0 {
+; CHECK-LABEL: usdot_single_za32_u16_vg1x2_tuple:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill
+; CHECK-NEXT: addvl sp, sp, #-1
+; CHECK-NEXT: str p8, [sp, #7, mul vl] // 2-byte Folded Spill
+; CHECK-NEXT: ptrue pn8.b
+; CHECK-NEXT: mov w8, wzr
+; CHECK-NEXT: ld1b { z2.b, z3.b }, pn8/z, [x0]
+; CHECK-NEXT: ld1b { z6.b, z7.b }, pn8/z, [x0, x1]
+; CHECK-NEXT: mov z5.d, z2.d
+; CHECK-NEXT: mov z4.d, z7.d
+; CHECK-NEXT: usdot za.s[w8, 0, vgx2], { z5.b, z6.b }, z0.b
+; CHECK-NEXT: usdot za.s[w8, 0, vgx2], { z3.b, z4.b }, z0.b
+; CHECK-NEXT: ldr p8, [sp, #7, mul vl] // 2-byte Folded Reload
+; CHECK-NEXT: addvl sp, sp, #1
+; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload
+; CHECK-NEXT: ret
+entry:
+ %0 = tail call target("aarch64.svcount") @llvm.aarch64.sve.ptrue.c8()
+ %1 = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld1.pn.x2.nxv16i8(target("aarch64.svcount") %0, ptr %ptr)
+ %2 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } %1, 0
+ %3 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } %1, 1
+ %arrayidx2 = getelementptr inbounds i8, ptr %ptr, i64 %stride
+ %4 = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld1.pn.x2.nxv16i8(target("aarch64.svcount") %0, ptr %arrayidx2)
+ %5 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } %4, 0
+ %6 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } %4, 1
+ call void @llvm.aarch64.sme.usdot.single.za32.vg1x2.nxv16i8(i32 0, <vscale x 16 x i8> %2, <vscale x 16 x i8> %5, <vscale x 16 x i8> %zn)
+ call void @llvm.aarch64.sme.usdot.single.za32.vg1x2.nxv16i8(i32 0, <vscale x 16 x i8> %3, <vscale x 16 x i8> %6, <vscale x 16 x i8> %zn)
+ ret void
+}
+
define void @usdot_single_za32_u8_vg1x4(i32 %slice, <vscale x 16 x i8> %unused, <vscale x 16 x i8> %zn0, <vscale x 16 x i8> %zn1, <vscale x 16 x i8> %zn2, <vscale x 16 x i8> %zn3, <vscale x 16 x i8> %zn4) #0 {
; CHECK-LABEL: usdot_single_za32_u8_vg1x4:
; CHECK: // %bb.0:
@@ -410,6 +571,100 @@ define void @usdot_single_za32_u8_vg1x4(i32 %slice, <vscale x 16 x i8> %unused,
ret void
}
+define void @usdot_single_za32_u16_vg1x4_tuple(ptr %ptr, i64 %stride, <vscale x 16 x i8> %zn) #0 {
+; CHECK-LABEL: usdot_single_za32_u16_vg1x4_tuple:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill
+; CHECK-NEXT: addvl sp, sp, #-13
+; CHECK-NEXT: lsl x9, x1, #1
+; CHECK-NEXT: str p8, [sp, #7, mul vl] // 2-byte Folded Spill
+; CHECK-NEXT: ptrue pn8.b
+; CHECK-NEXT: str z19, [sp, #1, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: mov w8, wzr
+; CHECK-NEXT: str z18, [sp, #2, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z17, [sp, #3, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z16, [sp, #4, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z15, [sp, #5, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z14, [sp, #6, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z13, [sp, #7, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z12, [sp, #8, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z11, [sp, #9, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z10, [sp, #10, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z9, [sp, #11, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z8, [sp, #12, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: ld1b { z24.b - z27.b }, pn8/z, [x0]
+; CHECK-NEXT: ld1b { z4.b - z7.b }, pn8/z, [x0, x1]
+; CHECK-NEXT: ld1b { z8.b - z11.b }, pn8/z, [x0, x9]
+; CHECK-NEXT: add x9, x9, x1
+; CHECK-NEXT: ld1b { z28.b - z31.b }, pn8/z, [x0, x9]
+; CHECK-NEXT: mov z1.d, z24.d
+; CHECK-NEXT: mov z2.d, z4.d
+; CHECK-NEXT: mov z3.d, z8.d
+; CHECK-NEXT: mov z12.d, z25.d
+; CHECK-NEXT: mov z13.d, z5.d
+; CHECK-NEXT: mov z14.d, z9.d
+; CHECK-NEXT: mov z4.d, z28.d
+; CHECK-NEXT: mov z15.d, z29.d
+; CHECK-NEXT: mov z16.d, z26.d
+; CHECK-NEXT: mov z17.d, z6.d
+; CHECK-NEXT: mov z18.d, z10.d
+; CHECK-NEXT: mov z19.d, z30.d
+; CHECK-NEXT: mov z28.d, z27.d
+; CHECK-NEXT: mov z29.d, z7.d
+; CHECK-NEXT: mov z30.d, z11.d
+; CHECK-NEXT: usdot za.s[w8, 0, vgx4], { z1.b - z4.b }, z0.b
+; CHECK-NEXT: usdot za.s[w8, 0, vgx4], { z12.b - z15.b }, z0.b
+; CHECK-NEXT: usdot za.s[w8, 0, vgx4], { z16.b - z19.b }, z0.b
+; CHECK-NEXT: usdot za.s[w8, 0, vgx4], { z28.b - z31.b }, z0.b
+; CHECK-NEXT: ldr z19, [sp, #1, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z18, [sp, #2, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z17, [sp, #3, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z16, [sp, #4, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z15, [sp, #5, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z14, [sp, #6, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z13, [sp, #7, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z12, [sp, #8, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z11, [sp, #9, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z10, [sp, #10, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z9, [sp, #11, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z8, [sp, #12, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr p8, [sp, #7, mul vl] // 2-byte Folded Reload
+; CHECK-NEXT: addvl sp, sp, #13
+; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload
+; CHECK-NEXT: ret
+entry:
+ %0 = tail call target("aarch64.svcount") @llvm.aarch64.sve.ptrue.c8()
+ %1 = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld1.pn.x4.nxv16i8(target("aarch64.svcount") %0, ptr %ptr)
+ %2 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %1, 0
+ %3 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %1, 1
+ %4 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %1, 2
+ %5 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %1, 3
+ %arrayidx2 = getelementptr inbounds i8, ptr %ptr, i64 %stride
+ %6 = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld1.pn.x4.nxv16i8(target("aarch64.svcount") %0, ptr %arrayidx2)
+ %7 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %6, 0
+ %8 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %6, 1
+ %9 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %6, 2
+ %10 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %6, 3
+ %mul3 = shl i64 %stride, 1
+ %arrayidx4 = getelementptr inbounds i8, ptr %ptr, i64 %mul3
+ %11 = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld1.pn.x4.nxv16i8(target("aarch64.svcount") %0, ptr %arrayidx4)
+ %12 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %11, 0
+ %13 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %11, 1
+ %14 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %11, 2
+ %15 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %11, 3
+ %mul5 = mul i64 %stride, 3
+ %arrayidx6 = getelementptr inbounds i8, ptr %ptr, i64 %mul5
+ %16 = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld1.pn.x4.nxv16i8(target("aarch64.svcount") %0, ptr %arrayidx6)
+ %17 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %16, 0
+ %18 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %16, 1
+ %19 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %16, 2
+ %20 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %16, 3
+ call void @llvm.aarch64.sme.usdot.single.za32.vg1x4.nxv16i8(i32 0, <vscale x 16 x i8> %2, <vscale x 16 x i8> %7, <vscale x 16 x i8> %12, <vscale x 16 x i8> %17, <vscale x 16 x i8> %zn)
+ call void @llvm.aarch64.sme.usdot.single.za32.vg1x4.nxv16i8(i32 0, <vscale x 16 x i8> %3, <vscale x 16 x i8> %8, <vscale x 16 x i8> %13, <vscale x 16 x i8> %18, <vscale x 16 x i8> %zn)
+ call void @llvm.aarch64.sme.usdot.single.za32.vg1x4.nxv16i8(i32 0, <vscale x 16 x i8> %4, <vscale x 16 x i8> %9, <vscale x 16 x i8> %14, <vscale x 16 x i8> %19, <vscale x 16 x i8> %zn)
+ call void @llvm.aarch64.sme.usdot.single.za32.vg1x4.nxv16i8(i32 0, <vscale x 16 x i8> %5, <vscale x 16 x i8> %10, <vscale x 16 x i8> %15, <vscale x 16 x i8> %20, <vscale x 16 x i8> %zn)
+ ret void
+}
; == Multi, single (signed) ==
@@ -426,6 +681,39 @@ define void @sdot_single_za32_u16_vg1x2(i32 %slice, <vscale x 16 x i8> %unused,
ret void
}
+define void @sdot_single_za32_u16_vg1x2_tuple(ptr %ptr, i64 %stride, <vscale x 8 x i16> %zn) #0 {
+; CHECK-LABEL: sdot_single_za32_u16_vg1x2_tuple:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill
+; CHECK-NEXT: addvl sp, sp, #-1
+; CHECK-NEXT: str p8, [sp, #7, mul vl] // 2-byte Folded Spill
+; CHECK-NEXT: ptrue pn8.b
+; CHECK-NEXT: add x9, x0, x1
+; CHECK-NEXT: ld1h { z2.h, z3.h }, pn8/z, [x0]
+; CHECK-NEXT: ld1h { z6.h, z7.h }, pn8/z, [x9]
+; CHECK-NEXT: mov w8, wzr
+; CHECK-NEXT: mov z5.d, z2.d
+; CHECK-NEXT: mov z4.d, z7.d
+; CHECK-NEXT: sdot za.s[w8, 0, vgx2], { z5.h, z6.h }, z0.h
+; CHECK-NEXT: sdot za.s[w8, 0, vgx2], { z3.h, z4.h }, z0.h
+; CHECK-NEXT: ldr p8, [sp, #7, mul vl] // 2-byte Folded Reload
+; CHECK-NEXT: addvl sp, sp, #1
+; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload
+; CHECK-NEXT: ret
+entry:
+ %0 = tail call target("aarch64.svcount") @llvm.aarch64.sve.ptrue.c8()
+ %1 = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld1.pn.x2.nxv8i16(target("aarch64.svcount") %0, ptr %ptr)
+ %2 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } %1, 0
+ %3 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } %1, 1
+ %arrayidx2 = getelementptr inbounds i8, ptr %ptr, i64 %stride
+ %4 = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld1.pn.x2.nxv8i16(target("aarch64.svcount") %0, ptr %arrayidx2)
+ %5 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } %4, 0
+ %6 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } %4, 1
+ call void @llvm.aarch64.sme.sdot.single.za32.vg1x2.nxv8i16(i32 0, <vscale x 8 x i16> %2, <vscale x 8 x i16> %5, <vscale x 8 x i16> %zn)
+ call void @llvm.aarch64.sme.sdot.single.za32.vg1x2.nxv8i16(i32 0, <vscale x 8 x i16> %3, <vscale x 8 x i16> %6, <vscale x 8 x i16> %zn)
+ ret void
+}
+
define void @sdot_single_za32_u16_vg1x4(i32 %slice, <vscale x 16 x i8> %unused, <vscale x 8 x i16> %zn0, <vscale x 8 x i16> %zn1, <vscale x 8 x i16> %zn2, <vscale x 8 x i16> %zn3, <vscale x 8 x i16> %zn4) #0 {
; CHECK-LABEL: sdot_single_za32_u16_vg1x4:
; CHECK: // %bb.0:
@@ -439,6 +727,102 @@ define void @sdot_single_za32_u16_vg1x4(i32 %slice, <vscale x 16 x i8> %unused,
ret void
}
+define void @sdot_single_za32_u16_vg1x4_tuple(ptr %ptr, i64 %stride, <vscale x 8 x i16> %zn) #0 {
+; CHECK-LABEL: sdot_single_za32_u16_vg1x4_tuple:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill
+; CHECK-NEXT: addvl sp, sp, #-13
+; CHECK-NEXT: add x10, x1, x1, lsl #1
+; CHECK-NEXT: str p8, [sp, #7, mul vl] // 2-byte Folded Spill
+; CHECK-NEXT: ptrue pn8.b
+; CHECK-NEXT: str z19, [sp, #1, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: add x9, x0, x1
+; CHECK-NEXT: mov w8, wzr
+; CHECK-NEXT: str z18, [sp, #2, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z17, [sp, #3, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z16, [sp, #4, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z15, [sp, #5, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z14, [sp, #6, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z13, [sp, #7, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z12, [sp, #8, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z11, [sp, #9, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z10, [sp, #10, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z9, [sp, #11, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z8, [sp, #12, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: ld1h { z4.h - z7.h }, pn8/z, [x0]
+; CHECK-NEXT: ld1h { z28.h - z31.h }, pn8/z, [x9]
+; CHECK-NEXT: add x9, x0, x10
+; CHECK-NEXT: ld1h { z24.h - z27.h }, pn8/z, [x0, x1, lsl #1]
+; CHECK-NEXT: ld1h { z8.h - z11.h }, pn8/z, [x9]
+; CHECK-NEXT: mov z1.d, z4.d
+; CHECK-NEXT: mov z2.d, z28.d
+; CHECK-NEXT: mov z12.d, z5.d
+; CHECK-NEXT: mov z3.d, z24.d
+; CHECK-NEXT: mov z4.d, z8.d
+; CHECK-NEXT: mov z13.d, z29.d
+; CHECK-NEXT: mov z14.d, z25.d
+; CHECK-NEXT: mov z15.d, z9.d
+; CHECK-NEXT: mov z16.d, z6.d
+; CHECK-NEXT: mov z17.d, z30.d
+; CHECK-NEXT: mov z18.d, z26.d
+; CHECK-NEXT: mov z19.d, z10.d
+; CHECK-NEXT: mov z8.d, z7.d
+; CHECK-NEXT: sdot za.s[w8, 0, vgx4], { z1.h - z4.h }, z0.h
+; CHECK-NEXT: mov z9.d, z31.d
+; CHECK-NEXT: mov z10.d, z27.d
+; CHECK-NEXT: sdot za.s[w8, 0, vgx4], { z12.h - z15.h }, z0.h
+; CHECK-NEXT: sdot za.s[w8, 0, vgx4], { z16.h - z19.h }, z0.h
+; CHECK-NEXT: sdot za.s[w8, 0, vgx4], { z8.h - z11.h }, z0.h
+; CHECK-NEXT: ldr z19, [sp, #1, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z18, [sp, #2, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z17, [sp, #3, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z16, [sp, #4, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z15, [sp, #5, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z14, [sp, #6, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z13, [sp, #7, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z12, [sp, #8, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z11, [sp, #9, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z10, [sp, #10, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z9, [sp, #11, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z8, [sp, #12, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr p8, [sp, #7, mul vl] // 2-byte Folded Reload
+; CHECK-NEXT: addvl sp, sp, #13
+; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload
+; CHECK-NEXT: ret
+entry:
+ %0 = tail call target("aarch64.svcount") @llvm.aarch64.sve.ptrue.c8()
+ %1 = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld1.pn.x4.nxv8i16(target("aarch64.svcount") %0, ptr %ptr)
+ %2 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %1, 0
+ %3 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %1, 1
+ %4 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %1, 2
+ %5 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %1, 3
+ %arrayidx2 = getelementptr inbounds i8, ptr %ptr, i64 %stride
+ %6 = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld1.pn.x4.nxv8i16(target("aarch64.svcount") %0, ptr %arrayidx2)
+ %7 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %6, 0
+ %8 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %6, 1
+ %9 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %6, 2
+ %10 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %6, 3
+ %mul3 = shl i64 %stride, 1
+ %arrayidx4 = getelementptr inbounds i8, ptr %ptr, i64 %mul3
+ %11 = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld1.pn.x4.nxv8i16(target("aarch64.svcount") %0, ptr %arrayidx4)
+ %12 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %11, 0
+ %13 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %11, 1
+ %14 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %11, 2
+ %15 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %11, 3
+ %mul5 = mul i64 %stride, 3
+ %arrayidx6 = getelementptr inbounds i8, ptr %ptr, i64 %mul5
+ %16 = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld1.pn.x4.nxv8i16(target("aarch64.svcount") %0, ptr %arrayidx6)
+ %17 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %16, 0
+ %18 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %16, 1
+ %19 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %16, 2
+ %20 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %16, 3
+ call void @llvm.aarch64.sme.sdot.single.za32.vg1x4.nxv8i16(i32 0, <vscale x 8 x i16> %2, <vscale x 8 x i16> %7, <vscale x 8 x i16> %12, <vscale x 8 x i16> %17, <vscale x 8 x i16> %zn)
+ call void @llvm.aarch64.sme.sdot.single.za32.vg1x4.nxv8i16(i32 0, <vscale x 8 x i16> %3, <vscale x 8 x i16> %8, <vscale x 8 x i16> %13, <vscale x 8 x i16> %18, <vscale x 8 x i16> %zn)
+ call void @llvm.aarch64.sme.sdot.single.za32.vg1x4.nxv8i16(i32 0, <vscale x 8 x i16> %4, <vscale x 8 x i16> %9, <vscale x 8 x i16> %14, <vscale x 8 x i16> %19, <vscale x 8 x i16> %zn)
+ call void @llvm.aarch64.sme.sdot.single.za32.vg1x4.nxv8i16(i32 0, <vscale x 8 x i16> %5, <vscale x 8 x i16> %10, <vscale x 8 x i16> %15, <vscale x 8 x i16> %20, <vscale x 8 x i16> %zn)
+ ret void
+}
+
define void @sdot_single_za32_u8_vg1x2(i32 %slice, <vscale x 16 x i8> %unused, <vscale x 16 x i8> %zn0, <vscale x 16 x i8> %zn1, <vscale x 16 x i8> %zn2) #0 {
; CHECK-LABEL: sdot_single_za32_u8_vg1x2:
; CHECK: // %bb.0:
@@ -504,6 +888,38 @@ define void @sudot_single_za32_u8_vg1x2(i32 %slice, <vscale x 16 x i8> %unused,
ret void
}
+define void @sudot_single_za32_u16_vg1x2_tuple(ptr %ptr, i64 %stride, <vscale x 16 x i8> %zn) #0 {
+; CHECK-LABEL: sudot_single_za32_u16_vg1x2_tuple:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill
+; CHECK-NEXT: addvl sp, sp, #-1
+; CHECK-NEXT: str p8, [sp, #7, mul vl] // 2-byte Folded Spill
+; CHECK-NEXT: ptrue pn8.b
+; CHECK-NEXT: mov w8, wzr
+; CHECK-NEXT: ld1b { z2.b, z3.b }, pn8/z, [x0]
+; CHECK-NEXT: ld1b { z6.b, z7.b }, pn8/z, [x0, x1]
+; CHECK-NEXT: mov z5.d, z2.d
+; CHECK-NEXT: mov z4.d, z7.d
+; CHECK-NEXT: sudot za.s[w8, 0, vgx2], { z5.b, z6.b }, z0.b
+; CHECK-NEXT: sudot za.s[w8, 0, vgx2], { z3.b, z4.b }, z0.b
+; CHECK-NEXT: ldr p8, [sp, #7, mul vl] // 2-byte Folded Reload
+; CHECK-NEXT: addvl sp, sp, #1
+; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload
+; CHECK-NEXT: ret
+entry:
+ %0 = tail call target("aarch64.svcount") @llvm.aarch64.sve.ptrue.c8()
+ %1 = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld1.pn.x2.nxv16i8(target("aarch64.svcount") %0, ptr %ptr)
+ %2 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } %1, 0
+ %3 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } %1, 1
+ %arrayidx2 = getelementptr inbounds i8, ptr %ptr, i64 %stride
+ %4 = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld1.pn.x2.nxv16i8(target("aarch64.svcount") %0, ptr %arrayidx2)
+ %5 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } %4, 0
+ %6 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } %4, 1
+ call void @llvm.aarch64.sme.sudot.single.za32.vg1x2.nxv16i8(i32 0, <vscale x 16 x i8> %2, <vscale x 16 x i8> %5, <vscale x 16 x i8> %zn)
+ call void @llvm.aarch64.sme.sudot.single.za32.vg1x2.nxv16i8(i32 0, <vscale x 16 x i8> %3, <vscale x 16 x i8> %6, <vscale x 16 x i8> %zn)
+ ret void
+}
+
define void @sudot_single_za32_u8_vg1x4(i32 %slice, <vscale x 16 x i8> %unused, <vscale x 16 x i8> %zn0, <vscale x 16 x i8> %zn1, <vscale x 16 x i8> %zn2, <vscale x 16 x i8> %zn3, <vscale x 16 x i8> %zn4) #0 {
; CHECK-LABEL: sudot_single_za32_u8_vg1x4:
; CHECK: // %bb.0:
@@ -517,6 +933,101 @@ define void @sudot_single_za32_u8_vg1x4(i32 %slice, <vscale x 16 x i8> %unused,
ret void
}
+define void @sudot_single_za32_u16_vg1x4_tuple(ptr %ptr, i64 %stride, <vscale x 16 x i8> %zn) #0 {
+; CHECK-LABEL: sudot_single_za32_u16_vg1x4_tuple:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill
+; CHECK-NEXT: addvl sp, sp, #-13
+; CHECK-NEXT: lsl x9, x1, #1
+; CHECK-NEXT: str p8, [sp, #7, mul vl] // 2-byte Folded Spill
+; CHECK-NEXT: ptrue pn8.b
+; CHECK-NEXT: str z19, [sp, #1, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: mov w8, wzr
+; CHECK-NEXT: str z18, [sp, #2, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z17, [sp, #3, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z16, [sp, #4, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z15, [sp, #5, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z14, [sp, #6, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z13, [sp, #7, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z12, [sp, #8, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z11, [sp, #9, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z10, [sp, #10, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z9, [sp, #11, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z8, [sp, #12, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: ld1b { z24.b - z27.b }, pn8/z, [x0]
+; CHECK-NEXT: ld1b { z4.b - z7.b }, pn8/z, [x0, x1]
+; CHECK-NEXT: ld1b { z8.b - z11.b }, pn8/z, [x0, x9]
+; CHECK-NEXT: add x9, x9, x1
+; CHECK-NEXT: ld1b { z28.b - z31.b }, pn8/z, [x0, x9]
+; CHECK-NEXT: mov z1.d, z24.d
+; CHECK-NEXT: mov z2.d, z4.d
+; CHECK-NEXT: mov z3.d, z8.d
+; CHECK-NEXT: mov z12.d, z25.d
+; CHECK-NEXT: mov z13.d, z5.d
+; CHECK-NEXT: mov z14.d, z9.d
+; CHECK-NEXT: mov z4.d, z28.d
+; CHECK-NEXT: mov z15.d, z29.d
+; CHECK-NEXT: mov z16.d, z26.d
+; CHECK-NEXT: mov z17.d, z6.d
+; CHECK-NEXT: mov z18.d, z10.d
+; CHECK-NEXT: mov z19.d, z30.d
+; CHECK-NEXT: mov z28.d, z27.d
+; CHECK-NEXT: mov z29.d, z7.d
+; CHECK-NEXT: mov z30.d, z11.d
+; CHECK-NEXT: sudot za.s[w8, 0, vgx4], { z1.b - z4.b }, z0.b
+; CHECK-NEXT: sudot za.s[w8, 0, vgx4], { z12.b - z15.b }, z0.b
+; CHECK-NEXT: sudot za.s[w8, 0, vgx4], { z16.b - z19.b }, z0.b
+; CHECK-NEXT: sudot za.s[w8, 0, vgx4], { z28.b - z31.b }, z0.b
+; CHECK-NEXT: ldr z19, [sp, #1, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z18, [sp, #2, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z17, [sp, #3, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z16, [sp, #4, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z15, [sp, #5, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z14, [sp, #6, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z13, [sp, #7, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z12, [sp, #8, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z11, [sp, #9, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z10, [sp, #10, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z9, [sp, #11, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z8, [sp, #12, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr p8, [sp, #7, mul vl] // 2-byte Folded Reload
+; CHECK-NEXT: addvl sp, sp, #13
+; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload
+; CHECK-NEXT: ret
+entry:
+ %0 = tail call target("aarch64.svcount") @llvm.aarch64.sve.ptrue.c8()
+ %1 = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld1.pn.x4.nxv16i8(target("aarch64.svcount") %0, ptr %ptr)
+ %2 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %1, 0
+ %3 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %1, 1
+ %4 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %1, 2
+ %5 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %1, 3
+ %arrayidx2 = getelementptr inbounds i8, ptr %ptr, i64 %stride
+ %6 = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld1.pn.x4.nxv16i8(target("aarch64.svcount") %0, ptr %arrayidx2)
+ %7 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %6, 0
+ %8 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %6, 1
+ %9 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %6, 2
+ %10 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %6, 3
+ %mul3 = shl i64 %stride, 1
+ %arrayidx4 = getelementptr inbounds i8, ptr %ptr, i64 %mul3
+ %11 = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld1.pn.x4.nxv16i8(target("aarch64.svcount") %0, ptr %arrayidx4)
+ %12 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %11, 0
+ %13 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %11, 1
+ %14 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %11, 2
+ %15 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %11, 3
+ %mul5 = mul i64 %stride, 3
+ %arrayidx6 = getelementptr inbounds i8, ptr %ptr, i64 %mul5
+ %16 = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld1.pn.x4.nxv16i8(target("aarch64.svcount") %0, ptr %arrayidx6)
+ %17 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %16, 0
+ %18 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %16, 1
+ %19 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %16, 2
+ %20 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %16, 3
+ call void @llvm.aarch64.sme.sudot.single.za32.vg1x4.nxv16i8(i32 0, <vscale x 16 x i8> %2, <vscale x 16 x i8> %7, <vscale x 16 x i8> %12, <vscale x 16 x i8> %17, <vscale x 16 x i8> %zn)
+ call void @llvm.aarch64.sme.sudot.single.za32.vg1x4.nxv16i8(i32 0, <vscale x 16 x i8> %3, <vscale x 16 x i8> %8, <vscale x 16 x i8> %13, <vscale x 16 x i8> %18, <vscale x 16 x i8> %zn)
+ call void @llvm.aarch64.sme.sudot.single.za32.vg1x4.nxv16i8(i32 0, <vscale x 16 x i8> %4, <vscale x 16 x i8> %9, <vscale x 16 x i8> %14, <vscale x 16 x i8> %19, <vscale x 16 x i8> %zn)
+ call void @llvm.aarch64.sme.sudot.single.za32.vg1x4.nxv16i8(i32 0, <vscale x 16 x i8> %5, <vscale x 16 x i8> %10, <vscale x 16 x i8> %15, <vscale x 16 x i8> %20, <vscale x 16 x i8> %zn)
+ ret void
+}
+
; == Multi, indexed (unsigned) ==
define void @udot_lane_za32_u16_vg1x2(i32 %slice, <vscale x 16 x i8> %unused, <vscale x 8 x i16> %zn0, <vscale x 8 x i16> %zn1, <vscale x 8 x i16> %zn2) #0 {
>From 463c0fa28d528e41fed19ce14db5035f844a568d Mon Sep 17 00:00:00 2001
From: Kerry McLaughlin <kerry.mclaughlin at arm.com>
Date: Thu, 16 Jan 2025 10:05:50 +0000
Subject: [PATCH 2/2] [AArch64][SME] Create separate FORM_TRANSPOSE pseudos for
ZPR & ZPRMul classes
The FORM_TRANSPOSED_REG_TUPLE pseudo nodes use either the ZPR2Mul2 or ZPR4Mul4
register classes for output. To extend these to other multi-vector intrinsics
which instead create a ZPR2/ZPR4 REG_SEQUENCE, a new pseudo has been added and
the existing one renamed.
---
.../AArch64/AArch64ExpandPseudoInsts.cpp | 2 +
.../Target/AArch64/AArch64ISelLowering.cpp | 10 +-
llvm/lib/Target/AArch64/AArch64InstrInfo.h | 12 +
.../Target/AArch64/AArch64RegisterInfo.cpp | 14 +-
llvm/lib/Target/AArch64/SMEInstrFormats.td | 26 +-
.../AArch64/sme2-intrinsics-int-dots.ll | 412 ++++++++----------
6 files changed, 223 insertions(+), 253 deletions(-)
diff --git a/llvm/lib/Target/AArch64/AArch64ExpandPseudoInsts.cpp b/llvm/lib/Target/AArch64/AArch64ExpandPseudoInsts.cpp
index b44c48afe705ba..239070925aa3aa 100644
--- a/llvm/lib/Target/AArch64/AArch64ExpandPseudoInsts.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ExpandPseudoInsts.cpp
@@ -1755,8 +1755,10 @@ bool AArch64ExpandPseudo::expandMI(MachineBasicBlock &MBB,
MBB, MBBI, AArch64::ZPR4RegClass, AArch64::ZPR4StridedRegClass,
AArch64::LDNT1D_4Z, AArch64::LDNT1D_4Z_STRIDED);
case AArch64::FORM_TRANSPOSED_REG_TUPLE_X2_PSEUDO:
+ case AArch64::FORM_TRANSPOSED_REG_TUPLE_MULX2_PSEUDO:
return expandFormTuplePseudo(MBB, MBBI, NextMBBI, 2);
case AArch64::FORM_TRANSPOSED_REG_TUPLE_X4_PSEUDO:
+ case AArch64::FORM_TRANSPOSED_REG_TUPLE_MULX4_PSEUDO:
return expandFormTuplePseudo(MBB, MBBI, NextMBBI, 4);
}
return false;
diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
index d4a114c275fb76..289ada6b4f86f3 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -8759,7 +8759,7 @@ static bool checkZExtBool(SDValue Arg, const SelectionDAG &DAG) {
// %6:zpr2stridedorcontiguous = LD1B_2Z_PSEUDO ..
// %7:zpr = COPY %6.zsub1:zpr2stridedorcontiguous
// %8:zpr = COPY %6.zsub0:zpr2stridedorcontiguous
-// %9:zpr2mul2 = FORM_TRANSPOSED_REG_TUPLE_X2_PSEUDO %5:zpr, %8:zpr
+// %9:zpr2mul2 = FORM_TRANSPOSED_REG_TUPLE_MULX2_PSEUDO %5:zpr, %8:zpr
//
bool shouldUseFormStridedPseudo(MachineInstr &MI) {
MachineRegisterInfo &MRI = MI.getMF()->getRegInfo();
@@ -8767,9 +8767,11 @@ bool shouldUseFormStridedPseudo(MachineInstr &MI) {
const TargetRegisterClass *RegClass = nullptr;
switch (MI.getOpcode()) {
case AArch64::FORM_TRANSPOSED_REG_TUPLE_X2_PSEUDO:
+ case AArch64::FORM_TRANSPOSED_REG_TUPLE_MULX2_PSEUDO:
RegClass = &AArch64::ZPR2StridedOrContiguousRegClass;
break;
case AArch64::FORM_TRANSPOSED_REG_TUPLE_X4_PSEUDO:
+ case AArch64::FORM_TRANSPOSED_REG_TUPLE_MULX4_PSEUDO:
RegClass = &AArch64::ZPR4StridedOrContiguousRegClass;
break;
default:
@@ -8824,14 +8826,14 @@ void AArch64TargetLowering::AdjustInstrPostInstrSelection(MachineInstr &MI,
}
}
- if (MI.getOpcode() == AArch64::FORM_TRANSPOSED_REG_TUPLE_X2_PSEUDO ||
- MI.getOpcode() == AArch64::FORM_TRANSPOSED_REG_TUPLE_X4_PSEUDO) {
+ const AArch64InstrInfo *TII =
+ MI.getMF()->getSubtarget<AArch64Subtarget>().getInstrInfo();
+ if (TII->isFormTransposedOpcode(MI.getOpcode())) {
// If input values to the FORM_TRANSPOSED_REG_TUPLE pseudo aren't copies
// from a StridedOrContiguous class, fall back on REG_SEQUENCE node.
if (shouldUseFormStridedPseudo(MI))
return;
- const TargetInstrInfo *TII = Subtarget->getInstrInfo();
MachineInstrBuilder MIB = BuildMI(*MI.getParent(), MI, MI.getDebugLoc(),
TII->get(TargetOpcode::REG_SEQUENCE),
MI.getOperand(0).getReg());
diff --git a/llvm/lib/Target/AArch64/AArch64InstrInfo.h b/llvm/lib/Target/AArch64/AArch64InstrInfo.h
index e37f70f7d985de..f45d868546e7f4 100644
--- a/llvm/lib/Target/AArch64/AArch64InstrInfo.h
+++ b/llvm/lib/Target/AArch64/AArch64InstrInfo.h
@@ -548,6 +548,18 @@ class AArch64InstrInfo final : public AArch64GenInstrInfo {
Register TargetReg,
bool FrameSetup) const;
+ bool isFormTransposedOpcode(unsigned Opc) const {
+ switch (Opc) {
+ case AArch64::FORM_TRANSPOSED_REG_TUPLE_X2_PSEUDO:
+ case AArch64::FORM_TRANSPOSED_REG_TUPLE_X4_PSEUDO:
+ case AArch64::FORM_TRANSPOSED_REG_TUPLE_MULX2_PSEUDO:
+ case AArch64::FORM_TRANSPOSED_REG_TUPLE_MULX4_PSEUDO:
+ return true;
+ default:
+ return false;
+ }
+ }
+
#define GET_INSTRINFO_HELPER_DECLS
#include "AArch64GenInstrInfo.inc"
diff --git a/llvm/lib/Target/AArch64/AArch64RegisterInfo.cpp b/llvm/lib/Target/AArch64/AArch64RegisterInfo.cpp
index 5973b63b5a8024..773d9946c192f1 100644
--- a/llvm/lib/Target/AArch64/AArch64RegisterInfo.cpp
+++ b/llvm/lib/Target/AArch64/AArch64RegisterInfo.cpp
@@ -1109,13 +1109,14 @@ bool AArch64RegisterInfo::getRegAllocationHints(
// so we add the strided registers as a hint.
unsigned RegID = MRI.getRegClass(VirtReg)->getID();
// Look through uses of the register for FORM_TRANSPOSED_REG_TUPLE.
+ const AArch64InstrInfo *TII =
+ MF.getSubtarget<AArch64Subtarget>().getInstrInfo();
if ((RegID == AArch64::ZPR2StridedOrContiguousRegClassID ||
RegID == AArch64::ZPR4StridedOrContiguousRegClassID) &&
- any_of(MRI.use_nodbg_instructions(VirtReg), [](const MachineInstr &Use) {
- return Use.getOpcode() ==
- AArch64::FORM_TRANSPOSED_REG_TUPLE_X2_PSEUDO ||
- Use.getOpcode() == AArch64::FORM_TRANSPOSED_REG_TUPLE_X4_PSEUDO;
- })) {
+ any_of(MRI.use_nodbg_instructions(VirtReg),
+ [&TII](const MachineInstr &Use) {
+ return TII->isFormTransposedOpcode(Use.getOpcode());
+ })) {
const TargetRegisterClass *StridedRC =
RegID == AArch64::ZPR2StridedOrContiguousRegClassID
? &AArch64::ZPR2StridedRegClass
@@ -1130,8 +1131,7 @@ bool AArch64RegisterInfo::getRegAllocationHints(
}
for (MachineInstr &MI : MRI.def_instructions(VirtReg)) {
- if (MI.getOpcode() != AArch64::FORM_TRANSPOSED_REG_TUPLE_X2_PSEUDO &&
- MI.getOpcode() != AArch64::FORM_TRANSPOSED_REG_TUPLE_X4_PSEUDO)
+ if (!TII->isFormTransposedOpcode(MI.getOpcode()))
return TargetRegisterInfo::getRegAllocationHints(VirtReg, Order, Hints,
MF, VRM);
diff --git a/llvm/lib/Target/AArch64/SMEInstrFormats.td b/llvm/lib/Target/AArch64/SMEInstrFormats.td
index 81004e70dc179b..918637a9c6d3cb 100644
--- a/llvm/lib/Target/AArch64/SMEInstrFormats.td
+++ b/llvm/lib/Target/AArch64/SMEInstrFormats.td
@@ -45,20 +45,26 @@ def am_sme_indexed_b4 : ComplexPattern<iPTR, 2, "SelectAddrModeIndexedSVE<0, 15>
// If the operands do not match this pattern, the pseudos are expanded
// to a REG_SEQUENCE using the post-isel hook.
-def FORM_TRANSPOSED_REG_TUPLE_X2_PSEUDO :
- Pseudo<(outs ZPR2Mul2:$tup),
- (ins ZPR:$zn0, ZPR:$zn1), []>, Sched<[]>{
+class sme_form_transpose_x2_pseudo<RegisterClass multi_vector_class>
+ : Pseudo<(outs multi_vector_class:$tup), (ins ZPR:$zn0, ZPR:$zn1), []>,
+ Sched<[]> {
let hasSideEffects = 0;
let hasPostISelHook = 1;
}
-def FORM_TRANSPOSED_REG_TUPLE_X4_PSEUDO :
- Pseudo<(outs ZPR4Mul4:$tup),
- (ins ZPR:$zn0, ZPR:$zn1, ZPR:$zn2, ZPR:$zn3), []>, Sched<[]>{
+def FORM_TRANSPOSED_REG_TUPLE_X2_PSEUDO : sme_form_transpose_x2_pseudo<ZPR2>;
+def FORM_TRANSPOSED_REG_TUPLE_MULX2_PSEUDO : sme_form_transpose_x2_pseudo<ZPR2Mul2>;
+
+class sme_form_transpose_x4_pseudo<RegisterClass multi_vector_class>
+ : Pseudo<(outs multi_vector_class:$tup), (ins ZPR:$zn0, ZPR:$zn1, ZPR:$zn2, ZPR:$zn3), []>,
+ Sched<[]> {
let hasSideEffects = 0;
let hasPostISelHook = 1;
}
+def FORM_TRANSPOSED_REG_TUPLE_X4_PSEUDO : sme_form_transpose_x4_pseudo<ZPR4>;
+def FORM_TRANSPOSED_REG_TUPLE_MULX4_PSEUDO : sme_form_transpose_x4_pseudo<ZPR4Mul4>;
+
def SDTZALoadStore : SDTypeProfile<0, 3, [SDTCisInt<0>, SDTCisPtrTy<1>, SDTCisInt<2>]>;
def AArch64SMELdr : SDNode<"AArch64ISD::SME_ZA_LDR", SDTZALoadStore,
[SDNPHasChain, SDNPSideEffect, SDNPMayLoad]>;
@@ -164,14 +170,14 @@ class SME2_ZA_TwoOp_Multi_Single_Pat<string name, SDPatternOperator intrinsic, O
class SME2_ZA_TwoOp_VG2_Multi_Single_Pat<string name, SDPatternOperator intrinsic, Operand index_ty, ZPRRegOp zpr_ty,
ValueType vt, ComplexPattern tileslice>
: Pat<(intrinsic (i32 (tileslice MatrixIndexGPR32Op8_11:$base, index_ty:$offset)), vt:$Zn1, vt:$Zn2, vt:$Zm),
- (!cast<Instruction>(name # _PSEUDO) $base, $offset, (REG_SEQUENCE ZPR2, vt:$Zn1, zsub0, vt:$Zn2, zsub1),
+ (!cast<Instruction>(name # _PSEUDO) $base, $offset, (FORM_TRANSPOSED_REG_TUPLE_X2_PSEUDO vt:$Zn1, vt:$Zn2),
zpr_ty:$Zm)>;
class SME2_ZA_TwoOp_VG4_Multi_Single_Pat<string name, SDPatternOperator intrinsic, Operand index_ty, ZPRRegOp zpr_ty,
ValueType vt, ComplexPattern tileslice>
: Pat<(intrinsic (i32 (tileslice MatrixIndexGPR32Op8_11:$base, index_ty:$offset)),
vt:$Zn1, vt:$Zn2, vt:$Zn3, vt:$Zn4, vt:$Zm),
(!cast<Instruction>(name # _PSEUDO) $base, $offset,
- (REG_SEQUENCE ZPR4, vt:$Zn1, zsub0, vt:$Zn2, zsub1, vt:$Zn3, zsub2, vt:$Zn4, zsub3),
+ (FORM_TRANSPOSED_REG_TUPLE_X4_PSEUDO vt:$Zn1, vt:$Zn2, vt:$Zn3, vt:$Zn4),
zpr_ty:$Zm)>;
class SME2_ZA_TwoOp_VG2_Multi_Multi_Pat<string name, SDPatternOperator intrinsic, Operand index_ty, ValueType vt, ComplexPattern tileslice>
@@ -197,14 +203,14 @@ class SME2_ZA_TwoOp_VG2_Multi_Index_Pat<string name, SDPatternOperator intrinsic
Operand imm_ty, ComplexPattern tileslice>
: Pat<(intrinsic (i32 (tileslice MatrixIndexGPR32Op8_11:$base, index_ty:$offset)), vt:$Zn1, vt:$Zn2, vt:$Zm, (i32 imm_ty:$i)),
(!cast<Instruction>(name # _PSEUDO) $base, $offset,
- (FORM_TRANSPOSED_REG_TUPLE_X2_PSEUDO vt:$Zn1,vt:$Zn2), zpr_ty:$Zm, imm_ty:$i)>;
+ (FORM_TRANSPOSED_REG_TUPLE_MULX2_PSEUDO vt:$Zn1,vt:$Zn2), zpr_ty:$Zm, imm_ty:$i)>;
class SME2_ZA_TwoOp_VG4_Multi_Index_Pat<string name, SDPatternOperator intrinsic, Operand index_ty, ZPRRegOp zpr_ty, ValueType vt,
Operand imm_ty, ComplexPattern tileslice>
: Pat<(intrinsic (i32 (tileslice MatrixIndexGPR32Op8_11:$base, index_ty:$offset)),
vt:$Zn1, vt:$Zn2, vt:$Zn3, vt:$Zn4, vt:$Zm, (i32 imm_ty:$i)),
(!cast<Instruction>(name # _PSEUDO) $base, $offset,
- (FORM_TRANSPOSED_REG_TUPLE_X4_PSEUDO vt:$Zn1, vt:$Zn2, vt:$Zn3, vt:$Zn4),
+ (FORM_TRANSPOSED_REG_TUPLE_MULX4_PSEUDO vt:$Zn1, vt:$Zn2, vt:$Zn3, vt:$Zn4),
zpr_ty:$Zm, imm_ty:$i)>;
class SME2_Sat_Shift_VG2_Pat<string name, SDPatternOperator intrinsic, ValueType out_vt, ValueType in_vt, Operand imm_ty>
diff --git a/llvm/test/CodeGen/AArch64/sme2-intrinsics-int-dots.ll b/llvm/test/CodeGen/AArch64/sme2-intrinsics-int-dots.ll
index df422df5c13c81..967d168593a400 100644
--- a/llvm/test/CodeGen/AArch64/sme2-intrinsics-int-dots.ll
+++ b/llvm/test/CodeGen/AArch64/sme2-intrinsics-int-dots.ll
@@ -323,19 +323,21 @@ define void @udot_single_za32_u16_vg1x2_tuple(ptr %ptr, i64 %stride, <vscale x 8
; CHECK-LABEL: udot_single_za32_u16_vg1x2_tuple:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill
-; CHECK-NEXT: addvl sp, sp, #-1
+; CHECK-NEXT: addvl sp, sp, #-3
; CHECK-NEXT: str p8, [sp, #7, mul vl] // 2-byte Folded Spill
; CHECK-NEXT: ptrue pn8.b
; CHECK-NEXT: add x9, x0, x1
-; CHECK-NEXT: ld1h { z2.h, z3.h }, pn8/z, [x0]
-; CHECK-NEXT: ld1h { z6.h, z7.h }, pn8/z, [x9]
+; CHECK-NEXT: str z10, [sp, #1, mul vl] // 16-byte Folded Spill
; CHECK-NEXT: mov w8, wzr
-; CHECK-NEXT: mov z5.d, z2.d
-; CHECK-NEXT: mov z4.d, z7.d
-; CHECK-NEXT: udot za.s[w8, 0, vgx2], { z5.h, z6.h }, z0.h
-; CHECK-NEXT: udot za.s[w8, 0, vgx2], { z3.h, z4.h }, z0.h
+; CHECK-NEXT: str z9, [sp, #2, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: ld1h { z1.h, z9.h }, pn8/z, [x0]
+; CHECK-NEXT: ld1h { z2.h, z10.h }, pn8/z, [x9]
+; CHECK-NEXT: udot za.s[w8, 0, vgx2], { z1.h, z2.h }, z0.h
+; CHECK-NEXT: udot za.s[w8, 0, vgx2], { z9.h, z10.h }, z0.h
+; CHECK-NEXT: ldr z10, [sp, #1, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z9, [sp, #2, mul vl] // 16-byte Folded Reload
; CHECK-NEXT: ldr p8, [sp, #7, mul vl] // 2-byte Folded Reload
-; CHECK-NEXT: addvl sp, sp, #1
+; CHECK-NEXT: addvl sp, sp, #3
; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload
; CHECK-NEXT: ret
entry:
@@ -369,62 +371,47 @@ define void @udot_single_za32_u16_vg1x4_tuple(ptr %ptr, i64 %stride, <vscale x 8
; CHECK-LABEL: udot_single_za32_u16_vg1x4_tuple:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill
-; CHECK-NEXT: addvl sp, sp, #-13
-; CHECK-NEXT: add x10, x1, x1, lsl #1
+; CHECK-NEXT: addvl sp, sp, #-11
+; CHECK-NEXT: add x9, x1, x1, lsl #1
; CHECK-NEXT: str p8, [sp, #7, mul vl] // 2-byte Folded Spill
; CHECK-NEXT: ptrue pn8.b
-; CHECK-NEXT: str z19, [sp, #1, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: add x9, x0, x1
+; CHECK-NEXT: str z20, [sp, #1, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: add x10, x0, x1
; CHECK-NEXT: mov w8, wzr
-; CHECK-NEXT: str z18, [sp, #2, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z17, [sp, #3, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z16, [sp, #4, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z15, [sp, #5, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z14, [sp, #6, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z13, [sp, #7, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z12, [sp, #8, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z11, [sp, #9, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z10, [sp, #10, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z9, [sp, #11, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z8, [sp, #12, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: ld1h { z4.h - z7.h }, pn8/z, [x0]
-; CHECK-NEXT: ld1h { z28.h - z31.h }, pn8/z, [x9]
-; CHECK-NEXT: add x9, x0, x10
-; CHECK-NEXT: ld1h { z24.h - z27.h }, pn8/z, [x0, x1, lsl #1]
-; CHECK-NEXT: ld1h { z8.h - z11.h }, pn8/z, [x9]
-; CHECK-NEXT: mov z1.d, z4.d
-; CHECK-NEXT: mov z2.d, z28.d
-; CHECK-NEXT: mov z12.d, z5.d
-; CHECK-NEXT: mov z3.d, z24.d
-; CHECK-NEXT: mov z4.d, z8.d
-; CHECK-NEXT: mov z13.d, z29.d
-; CHECK-NEXT: mov z14.d, z25.d
-; CHECK-NEXT: mov z15.d, z9.d
-; CHECK-NEXT: mov z16.d, z6.d
-; CHECK-NEXT: mov z17.d, z30.d
-; CHECK-NEXT: mov z18.d, z26.d
-; CHECK-NEXT: mov z19.d, z10.d
-; CHECK-NEXT: mov z8.d, z7.d
+; CHECK-NEXT: str z16, [sp, #2, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: add x9, x0, x9
+; CHECK-NEXT: str z15, [sp, #3, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z14, [sp, #4, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z13, [sp, #5, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z12, [sp, #6, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z11, [sp, #7, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z10, [sp, #8, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z9, [sp, #9, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z8, [sp, #10, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: ld1h { z1.h, z5.h, z9.h, z13.h }, pn8/z, [x0]
+; CHECK-NEXT: ld1h { z2.h, z6.h, z10.h, z14.h }, pn8/z, [x10]
+; CHECK-NEXT: ld1h { z3.h, z7.h, z11.h, z15.h }, pn8/z, [x0, x1, lsl #1]
+; CHECK-NEXT: ld1h { z16.h, z20.h, z24.h, z28.h }, pn8/z, [x9]
+; CHECK-NEXT: mov z4.d, z16.d
+; CHECK-NEXT: mov z8.d, z20.d
+; CHECK-NEXT: mov z12.d, z24.d
+; CHECK-NEXT: mov z16.d, z28.d
; CHECK-NEXT: udot za.s[w8, 0, vgx4], { z1.h - z4.h }, z0.h
-; CHECK-NEXT: mov z9.d, z31.d
-; CHECK-NEXT: mov z10.d, z27.d
-; CHECK-NEXT: udot za.s[w8, 0, vgx4], { z12.h - z15.h }, z0.h
-; CHECK-NEXT: udot za.s[w8, 0, vgx4], { z16.h - z19.h }, z0.h
-; CHECK-NEXT: udot za.s[w8, 0, vgx4], { z8.h - z11.h }, z0.h
-; CHECK-NEXT: ldr z19, [sp, #1, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z18, [sp, #2, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z17, [sp, #3, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z16, [sp, #4, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z15, [sp, #5, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z14, [sp, #6, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z13, [sp, #7, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z12, [sp, #8, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z11, [sp, #9, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z10, [sp, #10, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z9, [sp, #11, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z8, [sp, #12, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: udot za.s[w8, 0, vgx4], { z5.h - z8.h }, z0.h
+; CHECK-NEXT: udot za.s[w8, 0, vgx4], { z9.h - z12.h }, z0.h
+; CHECK-NEXT: udot za.s[w8, 0, vgx4], { z13.h - z16.h }, z0.h
+; CHECK-NEXT: ldr z20, [sp, #1, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z16, [sp, #2, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z15, [sp, #3, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z14, [sp, #4, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z13, [sp, #5, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z12, [sp, #6, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z11, [sp, #7, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z10, [sp, #8, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z9, [sp, #9, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z8, [sp, #10, mul vl] // 16-byte Folded Reload
; CHECK-NEXT: ldr p8, [sp, #7, mul vl] // 2-byte Folded Reload
-; CHECK-NEXT: addvl sp, sp, #13
+; CHECK-NEXT: addvl sp, sp, #11
; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload
; CHECK-NEXT: ret
entry:
@@ -530,18 +517,20 @@ define void @usdot_single_za32_u16_vg1x2_tuple(ptr %ptr, i64 %stride, <vscale x
; CHECK-LABEL: usdot_single_za32_u16_vg1x2_tuple:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill
-; CHECK-NEXT: addvl sp, sp, #-1
+; CHECK-NEXT: addvl sp, sp, #-3
; CHECK-NEXT: str p8, [sp, #7, mul vl] // 2-byte Folded Spill
; CHECK-NEXT: ptrue pn8.b
; CHECK-NEXT: mov w8, wzr
-; CHECK-NEXT: ld1b { z2.b, z3.b }, pn8/z, [x0]
-; CHECK-NEXT: ld1b { z6.b, z7.b }, pn8/z, [x0, x1]
-; CHECK-NEXT: mov z5.d, z2.d
-; CHECK-NEXT: mov z4.d, z7.d
-; CHECK-NEXT: usdot za.s[w8, 0, vgx2], { z5.b, z6.b }, z0.b
-; CHECK-NEXT: usdot za.s[w8, 0, vgx2], { z3.b, z4.b }, z0.b
+; CHECK-NEXT: str z10, [sp, #1, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z9, [sp, #2, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: ld1b { z1.b, z9.b }, pn8/z, [x0]
+; CHECK-NEXT: ld1b { z2.b, z10.b }, pn8/z, [x0, x1]
+; CHECK-NEXT: usdot za.s[w8, 0, vgx2], { z1.b, z2.b }, z0.b
+; CHECK-NEXT: usdot za.s[w8, 0, vgx2], { z9.b, z10.b }, z0.b
+; CHECK-NEXT: ldr z10, [sp, #1, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z9, [sp, #2, mul vl] // 16-byte Folded Reload
; CHECK-NEXT: ldr p8, [sp, #7, mul vl] // 2-byte Folded Reload
-; CHECK-NEXT: addvl sp, sp, #1
+; CHECK-NEXT: addvl sp, sp, #3
; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload
; CHECK-NEXT: ret
entry:
@@ -575,61 +564,46 @@ define void @usdot_single_za32_u16_vg1x4_tuple(ptr %ptr, i64 %stride, <vscale x
; CHECK-LABEL: usdot_single_za32_u16_vg1x4_tuple:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill
-; CHECK-NEXT: addvl sp, sp, #-13
+; CHECK-NEXT: addvl sp, sp, #-11
; CHECK-NEXT: lsl x9, x1, #1
; CHECK-NEXT: str p8, [sp, #7, mul vl] // 2-byte Folded Spill
; CHECK-NEXT: ptrue pn8.b
-; CHECK-NEXT: str z19, [sp, #1, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z20, [sp, #1, mul vl] // 16-byte Folded Spill
; CHECK-NEXT: mov w8, wzr
-; CHECK-NEXT: str z18, [sp, #2, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z17, [sp, #3, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z16, [sp, #4, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z15, [sp, #5, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z14, [sp, #6, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z13, [sp, #7, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z12, [sp, #8, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z11, [sp, #9, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z10, [sp, #10, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z9, [sp, #11, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z8, [sp, #12, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: ld1b { z24.b - z27.b }, pn8/z, [x0]
-; CHECK-NEXT: ld1b { z4.b - z7.b }, pn8/z, [x0, x1]
-; CHECK-NEXT: ld1b { z8.b - z11.b }, pn8/z, [x0, x9]
-; CHECK-NEXT: add x9, x9, x1
-; CHECK-NEXT: ld1b { z28.b - z31.b }, pn8/z, [x0, x9]
-; CHECK-NEXT: mov z1.d, z24.d
-; CHECK-NEXT: mov z2.d, z4.d
-; CHECK-NEXT: mov z3.d, z8.d
-; CHECK-NEXT: mov z12.d, z25.d
-; CHECK-NEXT: mov z13.d, z5.d
-; CHECK-NEXT: mov z14.d, z9.d
-; CHECK-NEXT: mov z4.d, z28.d
-; CHECK-NEXT: mov z15.d, z29.d
-; CHECK-NEXT: mov z16.d, z26.d
-; CHECK-NEXT: mov z17.d, z6.d
-; CHECK-NEXT: mov z18.d, z10.d
-; CHECK-NEXT: mov z19.d, z30.d
-; CHECK-NEXT: mov z28.d, z27.d
-; CHECK-NEXT: mov z29.d, z7.d
-; CHECK-NEXT: mov z30.d, z11.d
+; CHECK-NEXT: str z16, [sp, #2, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: add x10, x9, x1
+; CHECK-NEXT: str z15, [sp, #3, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z14, [sp, #4, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z13, [sp, #5, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z12, [sp, #6, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z11, [sp, #7, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z10, [sp, #8, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z9, [sp, #9, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z8, [sp, #10, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: ld1b { z1.b, z5.b, z9.b, z13.b }, pn8/z, [x0]
+; CHECK-NEXT: ld1b { z2.b, z6.b, z10.b, z14.b }, pn8/z, [x0, x1]
+; CHECK-NEXT: ld1b { z3.b, z7.b, z11.b, z15.b }, pn8/z, [x0, x9]
+; CHECK-NEXT: ld1b { z16.b, z20.b, z24.b, z28.b }, pn8/z, [x0, x10]
+; CHECK-NEXT: mov z4.d, z16.d
+; CHECK-NEXT: mov z8.d, z20.d
+; CHECK-NEXT: mov z12.d, z24.d
+; CHECK-NEXT: mov z16.d, z28.d
; CHECK-NEXT: usdot za.s[w8, 0, vgx4], { z1.b - z4.b }, z0.b
-; CHECK-NEXT: usdot za.s[w8, 0, vgx4], { z12.b - z15.b }, z0.b
-; CHECK-NEXT: usdot za.s[w8, 0, vgx4], { z16.b - z19.b }, z0.b
-; CHECK-NEXT: usdot za.s[w8, 0, vgx4], { z28.b - z31.b }, z0.b
-; CHECK-NEXT: ldr z19, [sp, #1, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z18, [sp, #2, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z17, [sp, #3, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z16, [sp, #4, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z15, [sp, #5, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z14, [sp, #6, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z13, [sp, #7, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z12, [sp, #8, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z11, [sp, #9, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z10, [sp, #10, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z9, [sp, #11, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z8, [sp, #12, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: usdot za.s[w8, 0, vgx4], { z5.b - z8.b }, z0.b
+; CHECK-NEXT: usdot za.s[w8, 0, vgx4], { z9.b - z12.b }, z0.b
+; CHECK-NEXT: usdot za.s[w8, 0, vgx4], { z13.b - z16.b }, z0.b
+; CHECK-NEXT: ldr z20, [sp, #1, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z16, [sp, #2, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z15, [sp, #3, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z14, [sp, #4, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z13, [sp, #5, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z12, [sp, #6, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z11, [sp, #7, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z10, [sp, #8, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z9, [sp, #9, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z8, [sp, #10, mul vl] // 16-byte Folded Reload
; CHECK-NEXT: ldr p8, [sp, #7, mul vl] // 2-byte Folded Reload
-; CHECK-NEXT: addvl sp, sp, #13
+; CHECK-NEXT: addvl sp, sp, #11
; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload
; CHECK-NEXT: ret
entry:
@@ -685,19 +659,21 @@ define void @sdot_single_za32_u16_vg1x2_tuple(ptr %ptr, i64 %stride, <vscale x 8
; CHECK-LABEL: sdot_single_za32_u16_vg1x2_tuple:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill
-; CHECK-NEXT: addvl sp, sp, #-1
+; CHECK-NEXT: addvl sp, sp, #-3
; CHECK-NEXT: str p8, [sp, #7, mul vl] // 2-byte Folded Spill
; CHECK-NEXT: ptrue pn8.b
; CHECK-NEXT: add x9, x0, x1
-; CHECK-NEXT: ld1h { z2.h, z3.h }, pn8/z, [x0]
-; CHECK-NEXT: ld1h { z6.h, z7.h }, pn8/z, [x9]
+; CHECK-NEXT: str z10, [sp, #1, mul vl] // 16-byte Folded Spill
; CHECK-NEXT: mov w8, wzr
-; CHECK-NEXT: mov z5.d, z2.d
-; CHECK-NEXT: mov z4.d, z7.d
-; CHECK-NEXT: sdot za.s[w8, 0, vgx2], { z5.h, z6.h }, z0.h
-; CHECK-NEXT: sdot za.s[w8, 0, vgx2], { z3.h, z4.h }, z0.h
+; CHECK-NEXT: str z9, [sp, #2, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: ld1h { z1.h, z9.h }, pn8/z, [x0]
+; CHECK-NEXT: ld1h { z2.h, z10.h }, pn8/z, [x9]
+; CHECK-NEXT: sdot za.s[w8, 0, vgx2], { z1.h, z2.h }, z0.h
+; CHECK-NEXT: sdot za.s[w8, 0, vgx2], { z9.h, z10.h }, z0.h
+; CHECK-NEXT: ldr z10, [sp, #1, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z9, [sp, #2, mul vl] // 16-byte Folded Reload
; CHECK-NEXT: ldr p8, [sp, #7, mul vl] // 2-byte Folded Reload
-; CHECK-NEXT: addvl sp, sp, #1
+; CHECK-NEXT: addvl sp, sp, #3
; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload
; CHECK-NEXT: ret
entry:
@@ -731,62 +707,47 @@ define void @sdot_single_za32_u16_vg1x4_tuple(ptr %ptr, i64 %stride, <vscale x 8
; CHECK-LABEL: sdot_single_za32_u16_vg1x4_tuple:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill
-; CHECK-NEXT: addvl sp, sp, #-13
-; CHECK-NEXT: add x10, x1, x1, lsl #1
+; CHECK-NEXT: addvl sp, sp, #-11
+; CHECK-NEXT: add x9, x1, x1, lsl #1
; CHECK-NEXT: str p8, [sp, #7, mul vl] // 2-byte Folded Spill
; CHECK-NEXT: ptrue pn8.b
-; CHECK-NEXT: str z19, [sp, #1, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: add x9, x0, x1
+; CHECK-NEXT: str z20, [sp, #1, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: add x10, x0, x1
; CHECK-NEXT: mov w8, wzr
-; CHECK-NEXT: str z18, [sp, #2, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z17, [sp, #3, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z16, [sp, #4, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z15, [sp, #5, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z14, [sp, #6, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z13, [sp, #7, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z12, [sp, #8, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z11, [sp, #9, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z10, [sp, #10, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z9, [sp, #11, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z8, [sp, #12, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: ld1h { z4.h - z7.h }, pn8/z, [x0]
-; CHECK-NEXT: ld1h { z28.h - z31.h }, pn8/z, [x9]
-; CHECK-NEXT: add x9, x0, x10
-; CHECK-NEXT: ld1h { z24.h - z27.h }, pn8/z, [x0, x1, lsl #1]
-; CHECK-NEXT: ld1h { z8.h - z11.h }, pn8/z, [x9]
-; CHECK-NEXT: mov z1.d, z4.d
-; CHECK-NEXT: mov z2.d, z28.d
-; CHECK-NEXT: mov z12.d, z5.d
-; CHECK-NEXT: mov z3.d, z24.d
-; CHECK-NEXT: mov z4.d, z8.d
-; CHECK-NEXT: mov z13.d, z29.d
-; CHECK-NEXT: mov z14.d, z25.d
-; CHECK-NEXT: mov z15.d, z9.d
-; CHECK-NEXT: mov z16.d, z6.d
-; CHECK-NEXT: mov z17.d, z30.d
-; CHECK-NEXT: mov z18.d, z26.d
-; CHECK-NEXT: mov z19.d, z10.d
-; CHECK-NEXT: mov z8.d, z7.d
+; CHECK-NEXT: str z16, [sp, #2, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: add x9, x0, x9
+; CHECK-NEXT: str z15, [sp, #3, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z14, [sp, #4, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z13, [sp, #5, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z12, [sp, #6, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z11, [sp, #7, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z10, [sp, #8, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z9, [sp, #9, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z8, [sp, #10, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: ld1h { z1.h, z5.h, z9.h, z13.h }, pn8/z, [x0]
+; CHECK-NEXT: ld1h { z2.h, z6.h, z10.h, z14.h }, pn8/z, [x10]
+; CHECK-NEXT: ld1h { z3.h, z7.h, z11.h, z15.h }, pn8/z, [x0, x1, lsl #1]
+; CHECK-NEXT: ld1h { z16.h, z20.h, z24.h, z28.h }, pn8/z, [x9]
+; CHECK-NEXT: mov z4.d, z16.d
+; CHECK-NEXT: mov z8.d, z20.d
+; CHECK-NEXT: mov z12.d, z24.d
+; CHECK-NEXT: mov z16.d, z28.d
; CHECK-NEXT: sdot za.s[w8, 0, vgx4], { z1.h - z4.h }, z0.h
-; CHECK-NEXT: mov z9.d, z31.d
-; CHECK-NEXT: mov z10.d, z27.d
-; CHECK-NEXT: sdot za.s[w8, 0, vgx4], { z12.h - z15.h }, z0.h
-; CHECK-NEXT: sdot za.s[w8, 0, vgx4], { z16.h - z19.h }, z0.h
-; CHECK-NEXT: sdot za.s[w8, 0, vgx4], { z8.h - z11.h }, z0.h
-; CHECK-NEXT: ldr z19, [sp, #1, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z18, [sp, #2, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z17, [sp, #3, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z16, [sp, #4, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z15, [sp, #5, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z14, [sp, #6, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z13, [sp, #7, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z12, [sp, #8, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z11, [sp, #9, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z10, [sp, #10, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z9, [sp, #11, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z8, [sp, #12, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: sdot za.s[w8, 0, vgx4], { z5.h - z8.h }, z0.h
+; CHECK-NEXT: sdot za.s[w8, 0, vgx4], { z9.h - z12.h }, z0.h
+; CHECK-NEXT: sdot za.s[w8, 0, vgx4], { z13.h - z16.h }, z0.h
+; CHECK-NEXT: ldr z20, [sp, #1, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z16, [sp, #2, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z15, [sp, #3, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z14, [sp, #4, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z13, [sp, #5, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z12, [sp, #6, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z11, [sp, #7, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z10, [sp, #8, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z9, [sp, #9, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z8, [sp, #10, mul vl] // 16-byte Folded Reload
; CHECK-NEXT: ldr p8, [sp, #7, mul vl] // 2-byte Folded Reload
-; CHECK-NEXT: addvl sp, sp, #13
+; CHECK-NEXT: addvl sp, sp, #11
; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload
; CHECK-NEXT: ret
entry:
@@ -892,18 +853,20 @@ define void @sudot_single_za32_u16_vg1x2_tuple(ptr %ptr, i64 %stride, <vscale x
; CHECK-LABEL: sudot_single_za32_u16_vg1x2_tuple:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill
-; CHECK-NEXT: addvl sp, sp, #-1
+; CHECK-NEXT: addvl sp, sp, #-3
; CHECK-NEXT: str p8, [sp, #7, mul vl] // 2-byte Folded Spill
; CHECK-NEXT: ptrue pn8.b
; CHECK-NEXT: mov w8, wzr
-; CHECK-NEXT: ld1b { z2.b, z3.b }, pn8/z, [x0]
-; CHECK-NEXT: ld1b { z6.b, z7.b }, pn8/z, [x0, x1]
-; CHECK-NEXT: mov z5.d, z2.d
-; CHECK-NEXT: mov z4.d, z7.d
-; CHECK-NEXT: sudot za.s[w8, 0, vgx2], { z5.b, z6.b }, z0.b
-; CHECK-NEXT: sudot za.s[w8, 0, vgx2], { z3.b, z4.b }, z0.b
+; CHECK-NEXT: str z10, [sp, #1, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z9, [sp, #2, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: ld1b { z1.b, z9.b }, pn8/z, [x0]
+; CHECK-NEXT: ld1b { z2.b, z10.b }, pn8/z, [x0, x1]
+; CHECK-NEXT: sudot za.s[w8, 0, vgx2], { z1.b, z2.b }, z0.b
+; CHECK-NEXT: sudot za.s[w8, 0, vgx2], { z9.b, z10.b }, z0.b
+; CHECK-NEXT: ldr z10, [sp, #1, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z9, [sp, #2, mul vl] // 16-byte Folded Reload
; CHECK-NEXT: ldr p8, [sp, #7, mul vl] // 2-byte Folded Reload
-; CHECK-NEXT: addvl sp, sp, #1
+; CHECK-NEXT: addvl sp, sp, #3
; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload
; CHECK-NEXT: ret
entry:
@@ -937,61 +900,46 @@ define void @sudot_single_za32_u16_vg1x4_tuple(ptr %ptr, i64 %stride, <vscale x
; CHECK-LABEL: sudot_single_za32_u16_vg1x4_tuple:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill
-; CHECK-NEXT: addvl sp, sp, #-13
+; CHECK-NEXT: addvl sp, sp, #-11
; CHECK-NEXT: lsl x9, x1, #1
; CHECK-NEXT: str p8, [sp, #7, mul vl] // 2-byte Folded Spill
; CHECK-NEXT: ptrue pn8.b
-; CHECK-NEXT: str z19, [sp, #1, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z20, [sp, #1, mul vl] // 16-byte Folded Spill
; CHECK-NEXT: mov w8, wzr
-; CHECK-NEXT: str z18, [sp, #2, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z17, [sp, #3, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z16, [sp, #4, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z15, [sp, #5, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z14, [sp, #6, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z13, [sp, #7, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z12, [sp, #8, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z11, [sp, #9, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z10, [sp, #10, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z9, [sp, #11, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: str z8, [sp, #12, mul vl] // 16-byte Folded Spill
-; CHECK-NEXT: ld1b { z24.b - z27.b }, pn8/z, [x0]
-; CHECK-NEXT: ld1b { z4.b - z7.b }, pn8/z, [x0, x1]
-; CHECK-NEXT: ld1b { z8.b - z11.b }, pn8/z, [x0, x9]
-; CHECK-NEXT: add x9, x9, x1
-; CHECK-NEXT: ld1b { z28.b - z31.b }, pn8/z, [x0, x9]
-; CHECK-NEXT: mov z1.d, z24.d
-; CHECK-NEXT: mov z2.d, z4.d
-; CHECK-NEXT: mov z3.d, z8.d
-; CHECK-NEXT: mov z12.d, z25.d
-; CHECK-NEXT: mov z13.d, z5.d
-; CHECK-NEXT: mov z14.d, z9.d
-; CHECK-NEXT: mov z4.d, z28.d
-; CHECK-NEXT: mov z15.d, z29.d
-; CHECK-NEXT: mov z16.d, z26.d
-; CHECK-NEXT: mov z17.d, z6.d
-; CHECK-NEXT: mov z18.d, z10.d
-; CHECK-NEXT: mov z19.d, z30.d
-; CHECK-NEXT: mov z28.d, z27.d
-; CHECK-NEXT: mov z29.d, z7.d
-; CHECK-NEXT: mov z30.d, z11.d
+; CHECK-NEXT: str z16, [sp, #2, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: add x10, x9, x1
+; CHECK-NEXT: str z15, [sp, #3, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z14, [sp, #4, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z13, [sp, #5, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z12, [sp, #6, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z11, [sp, #7, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z10, [sp, #8, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z9, [sp, #9, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: str z8, [sp, #10, mul vl] // 16-byte Folded Spill
+; CHECK-NEXT: ld1b { z1.b, z5.b, z9.b, z13.b }, pn8/z, [x0]
+; CHECK-NEXT: ld1b { z2.b, z6.b, z10.b, z14.b }, pn8/z, [x0, x1]
+; CHECK-NEXT: ld1b { z3.b, z7.b, z11.b, z15.b }, pn8/z, [x0, x9]
+; CHECK-NEXT: ld1b { z16.b, z20.b, z24.b, z28.b }, pn8/z, [x0, x10]
+; CHECK-NEXT: mov z4.d, z16.d
+; CHECK-NEXT: mov z8.d, z20.d
+; CHECK-NEXT: mov z12.d, z24.d
+; CHECK-NEXT: mov z16.d, z28.d
; CHECK-NEXT: sudot za.s[w8, 0, vgx4], { z1.b - z4.b }, z0.b
-; CHECK-NEXT: sudot za.s[w8, 0, vgx4], { z12.b - z15.b }, z0.b
-; CHECK-NEXT: sudot za.s[w8, 0, vgx4], { z16.b - z19.b }, z0.b
-; CHECK-NEXT: sudot za.s[w8, 0, vgx4], { z28.b - z31.b }, z0.b
-; CHECK-NEXT: ldr z19, [sp, #1, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z18, [sp, #2, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z17, [sp, #3, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z16, [sp, #4, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z15, [sp, #5, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z14, [sp, #6, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z13, [sp, #7, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z12, [sp, #8, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z11, [sp, #9, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z10, [sp, #10, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z9, [sp, #11, mul vl] // 16-byte Folded Reload
-; CHECK-NEXT: ldr z8, [sp, #12, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: sudot za.s[w8, 0, vgx4], { z5.b - z8.b }, z0.b
+; CHECK-NEXT: sudot za.s[w8, 0, vgx4], { z9.b - z12.b }, z0.b
+; CHECK-NEXT: sudot za.s[w8, 0, vgx4], { z13.b - z16.b }, z0.b
+; CHECK-NEXT: ldr z20, [sp, #1, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z16, [sp, #2, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z15, [sp, #3, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z14, [sp, #4, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z13, [sp, #5, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z12, [sp, #6, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z11, [sp, #7, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z10, [sp, #8, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z9, [sp, #9, mul vl] // 16-byte Folded Reload
+; CHECK-NEXT: ldr z8, [sp, #10, mul vl] // 16-byte Folded Reload
; CHECK-NEXT: ldr p8, [sp, #7, mul vl] // 2-byte Folded Reload
-; CHECK-NEXT: addvl sp, sp, #13
+; CHECK-NEXT: addvl sp, sp, #11
; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload
; CHECK-NEXT: ret
entry:
More information about the llvm-commits
mailing list