Skip to content

Commit

Permalink
[llvm][CodeGen] IR intrinsics for SVE2 contiguous conflict detection …
Browse files Browse the repository at this point in the history 
…instructions.

Summary:
The IR intrinsics are mapped to the following SVE2 instructions:

* WHILERW <Pd>.<T>, <Xn>, <Xm>
* WHILEWR <Pd>.<T>, <Xn>, <Xm>

The intrinsics introduced in this patch are the IR counterpart of the
SVE ACLE functions `svwhilerw` and `svwhilewr` (all data type
variants).

Patch by Maciej Gąbka <maciej.gabka@arm.com>.

Reviewers: kmclaughlin, rengolin

Reviewed By: kmclaughlin

Subscribers: tschuett, kristof.beyls, hiraditya, danielkiss, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D75862
  • Loading branch information
Francesco Petrogalli committed Mar 11, 2020
1 parent 9801e54 commit 4dde9e9
Show file tree
Hide file tree
Showing 4 changed files with 166 additions and 3 deletions.
18 changes: 18 additions & 0 deletions llvm/include/llvm/IR/IntrinsicsAArch64.td
View file
Expand Up @@ -1146,6 +1146,11 @@ let TargetPrefix = "aarch64" in { // All intrinsics start with "llvm.aarch64.".
llvm_i32_ty],
[IntrNoMem, ImmArg<2>]>;

class SVE2_CONFLICT_DETECT_Intrinsic
: Intrinsic<[llvm_anyvector_ty],
[LLVMAnyPointerType<llvm_any_ty>,
LLVMMatchType<1>]>;

class SVE2_3VectorArg_Indexed_Intrinsic
: Intrinsic<[llvm_anyvector_ty],
[LLVMMatchType<0>,
Expand Down Expand Up @@ -2167,3 +2172,16 @@ def int_aarch64_sve_bext_x : AdvSIMD_2VectorArg_Intrinsic;
def int_aarch64_sve_bgrp_x : AdvSIMD_2VectorArg_Intrinsic;

}

//
// SVE2 - Contiguous conflict detection
//

def int_aarch64_sve_whilerw_b : SVE2_CONFLICT_DETECT_Intrinsic;
def int_aarch64_sve_whilerw_h : SVE2_CONFLICT_DETECT_Intrinsic;
def int_aarch64_sve_whilerw_s : SVE2_CONFLICT_DETECT_Intrinsic;
def int_aarch64_sve_whilerw_d : SVE2_CONFLICT_DETECT_Intrinsic;
def int_aarch64_sve_whilewr_b : SVE2_CONFLICT_DETECT_Intrinsic;
def int_aarch64_sve_whilewr_h : SVE2_CONFLICT_DETECT_Intrinsic;
def int_aarch64_sve_whilewr_s : SVE2_CONFLICT_DETECT_Intrinsic;
def int_aarch64_sve_whilewr_d : SVE2_CONFLICT_DETECT_Intrinsic;
4 changes: 2 additions & 2 deletions llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td
View file
Expand Up @@ -1975,8 +1975,8 @@ let Predicates = [HasSVE2] in {
defm WHILEHI_PXX : sve_int_while8_rr<0b101, "whilehi", int_aarch64_sve_whilehi>;

// SVE2 pointer conflict compare
defm WHILEWR_PXX : sve2_int_while_rr<0b0, "whilewr">;
defm WHILERW_PXX : sve2_int_while_rr<0b1, "whilerw">;
defm WHILEWR_PXX : sve2_int_while_rr<0b0, "whilewr", "int_aarch64_sve_whilewr">;
defm WHILERW_PXX : sve2_int_while_rr<0b1, "whilerw", "int_aarch64_sve_whilerw">;
}

let Predicates = [HasSVE2AES] in {
Expand Down
8 changes: 7 additions & 1 deletion llvm/lib/Target/AArch64/SVEInstrFormats.td
View file
Expand Up @@ -4385,11 +4385,17 @@ class sve2_int_while_rr<bits<2> sz8_64, bits<1> rw, string asm,
let Defs = [NZCV];
}

multiclass sve2_int_while_rr<bits<1> rw, string asm> {
multiclass sve2_int_while_rr<bits<1> rw, string asm, string op> {
def _B : sve2_int_while_rr<0b00, rw, asm, PPR8>;
def _H : sve2_int_while_rr<0b01, rw, asm, PPR16>;
def _S : sve2_int_while_rr<0b10, rw, asm, PPR32>;
def _D : sve2_int_while_rr<0b11, rw, asm, PPR64>;

def : SVE_2_Op_Pat<nxv16i1, !cast<SDPatternOperator>(op # _b), i64, i64, !cast<Instruction>(NAME # _B)>;
def : SVE_2_Op_Pat<nxv8i1, !cast<SDPatternOperator>(op # _h), i64, i64, !cast<Instruction>(NAME # _H)>;
def : SVE_2_Op_Pat<nxv4i1, !cast<SDPatternOperator>(op # _s), i64, i64, !cast<Instruction>(NAME # _S)>;
def : SVE_2_Op_Pat<nxv2i1, !cast<SDPatternOperator>(op # _d), i64, i64, !cast<Instruction>(NAME # _D)>;

}

//===----------------------------------------------------------------------===//
Expand Down
139 changes: 139 additions & 0 deletions llvm/test/CodeGen/AArch64/sve2-intrinsics-contiguous-conflict-detection.ll
View file
@@ -0,0 +1,139 @@
; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve2 -asm-verbose=0 < %s | FileCheck %s

;
; WHILERW
;

define <vscale x 16 x i1> @whilerw_i8(i8* %a, i8* %b) {
; CHECK-LABEL: whilerw_i8:
; CHECK: whilerw p0.b, x0, x1
; CHECK-NEXT: ret
%out = call <vscale x 16 x i1> @llvm.aarch64.sve.whilerw.b.nx16i1(i8* %a, i8* %b)
ret <vscale x 16 x i1> %out
}

define <vscale x 8 x i1> @whilerw_i16(i16* %a, i16* %b) {
; CHECK-LABEL: whilerw_i16:
; CHECK: whilerw p0.h, x0, x1
; CHECK-NEXT: ret
%out = call <vscale x 8 x i1> @llvm.aarch64.sve.whilerw.h.nx8i1(i16* %a, i16* %b)
ret <vscale x 8 x i1> %out
}

define <vscale x 4 x i1> @whilerw_i32(i32* %a, i32* %b) {
; CHECK-LABEL: whilerw_i32:
; CHECK: whilerw p0.s, x0, x1
; CHECK-NEXT: ret
%out = call <vscale x 4 x i1> @llvm.aarch64.sve.whilerw.s.nx4i1(i32* %a, i32* %b)
ret <vscale x 4 x i1> %out
}

define <vscale x 2 x i1> @whilerw_i64(i64* %a, i64* %b) {
; CHECK-LABEL: whilerw_i64:
; CHECK: whilerw p0.d, x0, x1
; CHECK-NEXT: ret
%out = call <vscale x 2 x i1> @llvm.aarch64.sve.whilerw.d.nx2i1(i64* %a, i64* %b)
ret <vscale x 2 x i1> %out
}

define <vscale x 8 x i1> @whilerw_half(half* %a, half* %b) {
; CHECK-LABEL: whilerw_half:
; CHECK: whilerw p0.h, x0, x1
; CHECK-NEXT: ret
%out = call <vscale x 8 x i1> @llvm.aarch64.sve.whilerw.h.nx8i1.f16.f16(half* %a, half* %b)
ret <vscale x 8 x i1> %out
}

define <vscale x 4 x i1> @whilerw_float(float* %a, float* %b) {
; CHECK-LABEL: whilerw_float:
; CHECK: whilerw p0.s, x0, x1
; CHECK-NEXT: ret
%out = call <vscale x 4 x i1> @llvm.aarch64.sve.whilerw.s.nx4i1.f32.f32(float* %a, float* %b)
ret <vscale x 4 x i1> %out
}

define <vscale x 2 x i1> @whilerw_double(double* %a, double* %b) {
; CHECK-LABEL: whilerw_double:
; CHECK: whilerw p0.d, x0, x1
; CHECK-NEXT: ret
%out = call <vscale x 2 x i1> @llvm.aarch64.sve.whilerw.d.nx2i1.f64.f64(double* %a, double* %b)
ret <vscale x 2 x i1> %out
}

;
; WHILEWR
;

define <vscale x 16 x i1> @whilewr_i8(i8* %a, i8* %b) {
; CHECK-LABEL: whilewr_i8:
; CHECK: whilewr p0.b, x0, x1
; CHECK-NEXT: ret
%out = call <vscale x 16 x i1> @llvm.aarch64.sve.whilewr.b.nx16i1(i8* %a, i8* %b)
ret <vscale x 16 x i1> %out
}

define <vscale x 8 x i1> @whilewr_i16(i16* %a, i16* %b) {
; CHECK-LABEL: whilewr_i16:
; CHECK: whilewr p0.h, x0, x1
; CHECK-NEXT: ret
%out = call <vscale x 8 x i1> @llvm.aarch64.sve.whilewr.h.nx8i1(i16* %a, i16* %b)
ret <vscale x 8 x i1> %out
}

define <vscale x 4 x i1> @whilewr_i32(i32* %a, i32* %b) {
; CHECK-LABEL: whilewr_i32:
; CHECK: whilewr p0.s, x0, x1
; CHECK-NEXT: ret
%out = call <vscale x 4 x i1> @llvm.aarch64.sve.whilewr.s.nx4i1(i32* %a, i32* %b)
ret <vscale x 4 x i1> %out
}

define <vscale x 2 x i1> @whilewr_i64(i64* %a, i64* %b) {
; CHECK-LABEL: whilewr_i64:
; CHECK: whilewr p0.d, x0, x1
; CHECK-NEXT: ret
%out = call <vscale x 2 x i1> @llvm.aarch64.sve.whilewr.d.nx2i1(i64* %a, i64* %b)
ret <vscale x 2 x i1> %out
}

define <vscale x 8 x i1> @whilewr_half(half* %a, half* %b) {
; CHECK-LABEL: whilewr_half:
; CHECK: whilewr p0.h, x0, x1
; CHECK-NEXT: ret
%out = call <vscale x 8 x i1> @llvm.aarch64.sve.whilewr.h.nx8i1.f16.f16(half* %a, half* %b)
ret <vscale x 8 x i1> %out
}

define <vscale x 4 x i1> @whilewr_float(float* %a, float* %b) {
; CHECK-LABEL: whilewr_float:
; CHECK: whilewr p0.s, x0, x1
; CHECK-NEXT: ret
%out = call <vscale x 4 x i1> @llvm.aarch64.sve.whilewr.s.nx4i1.f32.f32(float* %a, float* %b)
ret <vscale x 4 x i1> %out
}

define <vscale x 2 x i1> @whilewr_double(double* %a, double* %b) {
; CHECK-LABEL: whilewr_double:
; CHECK: whilewr p0.d, x0, x1
; CHECK-NEXT: ret
%out = call <vscale x 2 x i1> @llvm.aarch64.sve.whilewr.d.nx2i1.f64.f64(double* %a, double* %b)
ret <vscale x 2 x i1> %out
}

declare <vscale x 16 x i1> @llvm.aarch64.sve.whilerw.b.nx16i1(i8* %a, i8* %b)
declare <vscale x 8 x i1> @llvm.aarch64.sve.whilerw.h.nx8i1(i16* %a, i16* %b)
declare <vscale x 4 x i1> @llvm.aarch64.sve.whilerw.s.nx4i1(i32* %a, i32* %b)
declare <vscale x 2 x i1> @llvm.aarch64.sve.whilerw.d.nx2i1(i64* %a, i64* %b)

declare <vscale x 8 x i1> @llvm.aarch64.sve.whilerw.h.nx8i1.f16.f16(half* %a, half* %b)
declare <vscale x 4 x i1> @llvm.aarch64.sve.whilerw.s.nx4i1.f32.f32(float* %a, float* %b)
declare <vscale x 2 x i1> @llvm.aarch64.sve.whilerw.d.nx2i1.f64.f64(double* %a, double* %b)

declare <vscale x 16 x i1> @llvm.aarch64.sve.whilewr.b.nx16i1(i8* %a, i8* %b)
declare <vscale x 8 x i1> @llvm.aarch64.sve.whilewr.h.nx8i1(i16* %a, i16* %b)
declare <vscale x 4 x i1> @llvm.aarch64.sve.whilewr.s.nx4i1(i32* %a, i32* %b)
declare <vscale x 2 x i1> @llvm.aarch64.sve.whilewr.d.nx2i1(i64* %a, i64* %b)

declare <vscale x 8 x i1> @llvm.aarch64.sve.whilewr.h.nx8i1.f16.f16(half* %a, half* %b)
declare <vscale x 4 x i1> @llvm.aarch64.sve.whilewr.s.nx4i1.f32.f32(float* %a, float* %b)
declare <vscale x 2 x i1> @llvm.aarch64.sve.whilewr.d.nx2i1.f64.f64(double* %a, double* %b)

0 comments on commit 4dde9e9

Please sign in to comment.