mirror of
https://github.com/c64scene-ar/llvm-6502.git
synced 2024-12-15 20:29:48 +00:00
704d1347c5
This is a minor extension of SROA to handle a special case that is important for some ARM NEON operations. Some of the NEON intrinsics return multiple values, which are handled as struct types containing multiple elements of the same vector type. The corresponding return types declared in the arm_neon.h header have equivalent arrays. We need SROA to recognize that it can split up those arrays and structs into separate vectors, even though they are not always accessed with the same type. SROA already handles loads and stores of an entire alloca by using insertvalue/extractvalue to access the individual pieces, and that code works the same regardless of whether the type is a struct or an array. So, all that needs to be done is to check for compatible arrays and homogeneous structs. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@123381 91177308-0d34-0410-b5e6-96231b3b80d8
109 lines
3.3 KiB
LLVM
109 lines
3.3 KiB
LLVM
; RUN: opt < %s -scalarrepl -S | FileCheck %s
|
|
; PR3290
|
|
target datalayout = "E-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64"
|
|
|
|
;; Store of integer to whole alloca struct.
|
|
define i32 @test1(i64 %V) nounwind {
|
|
; CHECK: test1
|
|
; CHECK-NOT: alloca
|
|
%X = alloca {{i32, i32}}
|
|
%Y = bitcast {{i32,i32}}* %X to i64*
|
|
store i64 %V, i64* %Y
|
|
|
|
%A = getelementptr {{i32,i32}}* %X, i32 0, i32 0, i32 0
|
|
%B = getelementptr {{i32,i32}}* %X, i32 0, i32 0, i32 1
|
|
%a = load i32* %A
|
|
%b = load i32* %B
|
|
%c = add i32 %a, %b
|
|
ret i32 %c
|
|
}
|
|
|
|
;; Store of integer to whole struct/array alloca.
|
|
define float @test2(i128 %V) nounwind {
|
|
; CHECK: test2
|
|
; CHECK-NOT: alloca
|
|
%X = alloca {[4 x float]}
|
|
%Y = bitcast {[4 x float]}* %X to i128*
|
|
store i128 %V, i128* %Y
|
|
|
|
%A = getelementptr {[4 x float]}* %X, i32 0, i32 0, i32 0
|
|
%B = getelementptr {[4 x float]}* %X, i32 0, i32 0, i32 3
|
|
%a = load float* %A
|
|
%b = load float* %B
|
|
%c = fadd float %a, %b
|
|
ret float %c
|
|
}
|
|
|
|
;; Load of whole alloca struct as integer
|
|
define i64 @test3(i32 %a, i32 %b) nounwind {
|
|
; CHECK: test3
|
|
; CHECK-NOT: alloca
|
|
%X = alloca {{i32, i32}}
|
|
|
|
%A = getelementptr {{i32,i32}}* %X, i32 0, i32 0, i32 0
|
|
%B = getelementptr {{i32,i32}}* %X, i32 0, i32 0, i32 1
|
|
store i32 %a, i32* %A
|
|
store i32 %b, i32* %B
|
|
|
|
%Y = bitcast {{i32,i32}}* %X to i64*
|
|
%Z = load i64* %Y
|
|
ret i64 %Z
|
|
}
|
|
|
|
;; load of integer from whole struct/array alloca.
|
|
define i128 @test4(float %a, float %b) nounwind {
|
|
; CHECK: test4
|
|
; CHECK-NOT: alloca
|
|
%X = alloca {[4 x float]}
|
|
%A = getelementptr {[4 x float]}* %X, i32 0, i32 0, i32 0
|
|
%B = getelementptr {[4 x float]}* %X, i32 0, i32 0, i32 3
|
|
store float %a, float* %A
|
|
store float %b, float* %B
|
|
|
|
%Y = bitcast {[4 x float]}* %X to i128*
|
|
%V = load i128* %Y
|
|
ret i128 %V
|
|
}
|
|
|
|
;; If the elements of a struct or array alloca contain padding, SROA can still
|
|
;; split up the alloca as long as there is no padding between the elements.
|
|
%padded = type { i16, i8 }
|
|
%arr = type [4 x %padded]
|
|
define void @test5(%arr* %p, %arr* %q) {
|
|
entry:
|
|
; CHECK: test5
|
|
; CHECK-NOT: i128
|
|
%var = alloca %arr, align 4
|
|
%vari8 = bitcast %arr* %var to i8*
|
|
%pi8 = bitcast %arr* %p to i8*
|
|
call void @llvm.memcpy.p0i8.p0i8.i32(i8* %vari8, i8* %pi8, i32 16, i32 4, i1 false)
|
|
%qi8 = bitcast %arr* %q to i8*
|
|
call void @llvm.memcpy.p0i8.p0i8.i32(i8* %qi8, i8* %vari8, i32 16, i32 4, i1 false)
|
|
ret void
|
|
}
|
|
|
|
;; Check that an array alloca can be split up when it is also accessed with
|
|
;; a load or store as a homogeneous structure with the same element type and
|
|
;; number of elements as the array.
|
|
%homogeneous = type { <8 x i16>, <8 x i16>, <8 x i16> }
|
|
%wrapped_array = type { [3 x <8 x i16>] }
|
|
define void @test6(i8* %p, %wrapped_array* %arr) {
|
|
entry:
|
|
; CHECK: test6
|
|
; CHECK: store <8 x i16>
|
|
; CHECK: store <8 x i16>
|
|
; CHECK: store <8 x i16>
|
|
%var = alloca %wrapped_array, align 16
|
|
%res = call %homogeneous @test6callee(i8* %p)
|
|
%varcast = bitcast %wrapped_array* %var to %homogeneous*
|
|
store %homogeneous %res, %homogeneous* %varcast
|
|
%tmp1 = bitcast %wrapped_array* %arr to i8*
|
|
%tmp2 = bitcast %wrapped_array* %var to i8*
|
|
call void @llvm.memcpy.p0i8.p0i8.i32(i8* %tmp1, i8* %tmp2, i32 48, i32 16, i1 false)
|
|
ret void
|
|
}
|
|
|
|
declare %homogeneous @test6callee(i8* nocapture) nounwind
|
|
|
|
declare void @llvm.memcpy.p0i8.p0i8.i32(i8* nocapture, i8* nocapture, i32, i32, i1) nounwind
|