90ba33099c
This patch canonicalizes getelementptr instructions with constant indices to use the `i8` source element type. This makes it easier for optimizations to recognize that two GEPs are identical, because they don't need to see past many different ways to express the same offset. This is a first step towards https://discourse.llvm.org/t/rfc-replacing-getelementptr-with-ptradd/68699. This is limited to constant GEPs only for now, as they have a clear canonical form, while we're not yet sure how exactly to deal with variable indices. The test llvm/test/Transforms/PhaseOrdering/switch_with_geps.ll gives two representative examples of the kind of optimization improvement we expect from this change. In the first test SimplifyCFG can now realize that all switch branches are actually the same. In the second test it can convert it into simple arithmetic. These are representative of common optimization failures we see in Rust. Fixes https://github.com/llvm/llvm-project/issues/69841.
58 lines
4.1 KiB
LLVM
58 lines
4.1 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
|
|
; RUN: opt -passes='lower-matrix-intrinsics<minimal>,instcombine,verify<domtree>' -fuse-matrix-tile-size=2 -matrix-allow-contract -force-fuse-matrix %s -S | FileCheck %s
|
|
|
|
; Test for the minimal version of the matrix lowering pass, which does not
|
|
; require DT or AA. Make sure no tiling is happening, even though it was
|
|
; requested.
|
|
|
|
; REQUIRES: aarch64-registered-target
|
|
|
|
target datalayout = "e-m:o-i64:64-f80:128-n8:8:32:64-S128"
|
|
target triple = "aarch64-apple-ios"
|
|
|
|
define void @multiply(ptr %A, ptr %B, ptr %C) {
|
|
; CHECK-LABEL: @multiply(
|
|
; CHECK-NEXT: entry:
|
|
; CHECK-NEXT: [[COL_LOAD:%.*]] = load <2 x double>, ptr [[A:%.*]], align 8
|
|
; CHECK-NEXT: [[VEC_GEP:%.*]] = getelementptr i8, ptr [[A]], i64 16
|
|
; CHECK-NEXT: [[COL_LOAD1:%.*]] = load <2 x double>, ptr [[VEC_GEP]], align 8
|
|
; CHECK-NEXT: [[VEC_GEP2:%.*]] = getelementptr i8, ptr [[A]], i64 32
|
|
; CHECK-NEXT: [[COL_LOAD3:%.*]] = load <2 x double>, ptr [[VEC_GEP2]], align 8
|
|
; CHECK-NEXT: [[VEC_GEP4:%.*]] = getelementptr i8, ptr [[A]], i64 48
|
|
; CHECK-NEXT: [[COL_LOAD5:%.*]] = load <2 x double>, ptr [[VEC_GEP4]], align 8
|
|
; CHECK-NEXT: [[COL_LOAD6:%.*]] = load <4 x double>, ptr [[B:%.*]], align 8
|
|
; CHECK-NEXT: [[VEC_GEP7:%.*]] = getelementptr i8, ptr [[B]], i64 32
|
|
; CHECK-NEXT: [[COL_LOAD8:%.*]] = load <4 x double>, ptr [[VEC_GEP7]], align 8
|
|
; CHECK-NEXT: [[SPLAT_SPLAT:%.*]] = shufflevector <4 x double> [[COL_LOAD6]], <4 x double> poison, <2 x i32> zeroinitializer
|
|
; CHECK-NEXT: [[TMP0:%.*]] = fmul contract <2 x double> [[COL_LOAD]], [[SPLAT_SPLAT]]
|
|
; CHECK-NEXT: [[SPLAT_SPLAT11:%.*]] = shufflevector <4 x double> [[COL_LOAD6]], <4 x double> poison, <2 x i32> <i32 1, i32 1>
|
|
; CHECK-NEXT: [[TMP1:%.*]] = call contract <2 x double> @llvm.fmuladd.v2f64(<2 x double> [[COL_LOAD1]], <2 x double> [[SPLAT_SPLAT11]], <2 x double> [[TMP0]])
|
|
; CHECK-NEXT: [[SPLAT_SPLAT14:%.*]] = shufflevector <4 x double> [[COL_LOAD6]], <4 x double> poison, <2 x i32> <i32 2, i32 2>
|
|
; CHECK-NEXT: [[TMP2:%.*]] = call contract <2 x double> @llvm.fmuladd.v2f64(<2 x double> [[COL_LOAD3]], <2 x double> [[SPLAT_SPLAT14]], <2 x double> [[TMP1]])
|
|
; CHECK-NEXT: [[SPLAT_SPLAT17:%.*]] = shufflevector <4 x double> [[COL_LOAD6]], <4 x double> poison, <2 x i32> <i32 3, i32 3>
|
|
; CHECK-NEXT: [[TMP3:%.*]] = call contract <2 x double> @llvm.fmuladd.v2f64(<2 x double> [[COL_LOAD5]], <2 x double> [[SPLAT_SPLAT17]], <2 x double> [[TMP2]])
|
|
; CHECK-NEXT: [[SPLAT_SPLAT20:%.*]] = shufflevector <4 x double> [[COL_LOAD8]], <4 x double> poison, <2 x i32> zeroinitializer
|
|
; CHECK-NEXT: [[TMP4:%.*]] = fmul contract <2 x double> [[COL_LOAD]], [[SPLAT_SPLAT20]]
|
|
; CHECK-NEXT: [[SPLAT_SPLAT23:%.*]] = shufflevector <4 x double> [[COL_LOAD8]], <4 x double> poison, <2 x i32> <i32 1, i32 1>
|
|
; CHECK-NEXT: [[TMP5:%.*]] = call contract <2 x double> @llvm.fmuladd.v2f64(<2 x double> [[COL_LOAD1]], <2 x double> [[SPLAT_SPLAT23]], <2 x double> [[TMP4]])
|
|
; CHECK-NEXT: [[SPLAT_SPLAT26:%.*]] = shufflevector <4 x double> [[COL_LOAD8]], <4 x double> poison, <2 x i32> <i32 2, i32 2>
|
|
; CHECK-NEXT: [[TMP6:%.*]] = call contract <2 x double> @llvm.fmuladd.v2f64(<2 x double> [[COL_LOAD3]], <2 x double> [[SPLAT_SPLAT26]], <2 x double> [[TMP5]])
|
|
; CHECK-NEXT: [[SPLAT_SPLAT29:%.*]] = shufflevector <4 x double> [[COL_LOAD8]], <4 x double> poison, <2 x i32> <i32 3, i32 3>
|
|
; CHECK-NEXT: [[TMP7:%.*]] = call contract <2 x double> @llvm.fmuladd.v2f64(<2 x double> [[COL_LOAD5]], <2 x double> [[SPLAT_SPLAT29]], <2 x double> [[TMP6]])
|
|
; CHECK-NEXT: store <2 x double> [[TMP3]], ptr [[C:%.*]], align 8
|
|
; CHECK-NEXT: [[VEC_GEP30:%.*]] = getelementptr i8, ptr [[C]], i64 16
|
|
; CHECK-NEXT: store <2 x double> [[TMP7]], ptr [[VEC_GEP30]], align 8
|
|
; CHECK-NEXT: ret void
|
|
;
|
|
entry:
|
|
%a = load <8 x double>, ptr %A, align 8
|
|
%b = load <8 x double>, ptr %B, align 8
|
|
|
|
%c = call <4 x double> @llvm.matrix.multiply(<8 x double> %a, <8 x double> %b, i32 2, i32 4, i32 2)
|
|
|
|
store <4 x double> %c, ptr %C, align 8
|
|
ret void
|
|
}
|
|
|
|
declare <4 x double> @llvm.matrix.multiply(<8 x double>, <8 x double>, i32, i32, i32)
|