mirror of
https://github.com/c3lang/c3c.git
synced 2026-02-27 12:01:16 +00:00
423152202f
* Optimize vector load / store. Fixes to alignment. Support typedef with `@simd` and `@align` #2543. Update vector ABI #2542 * Fix alignment issue with indirect arguments.
54 lines
1.5 KiB
Plaintext
54 lines
1.5 KiB
Plaintext
// #target: macos-x64
|
|
|
|
module test;
|
|
|
|
typedef Int16 = int[<16>] @simd;
|
|
fn void test(int[16] x, Int16 y)
|
|
{
|
|
int a = x[4];
|
|
int b = x[^2];
|
|
int c = y[4];
|
|
int d = y[^2];
|
|
int j = 3;
|
|
int e = y[^j];
|
|
int f = x[^j];
|
|
}
|
|
|
|
/* #expect: test.ll
|
|
|
|
define void @test.test(ptr byval([16 x i32]) align 8 %0, ptr byval(<16 x i32>) align 64 %1) #0 {
|
|
entry:
|
|
%a = alloca i32, align 4
|
|
%b = alloca i32, align 4
|
|
%c = alloca i32, align 4
|
|
%d = alloca i32, align 4
|
|
%j = alloca i32, align 4
|
|
%e = alloca i32, align 4
|
|
%f = alloca i32, align 4
|
|
%ptradd = getelementptr inbounds i8, ptr %0, i64 16
|
|
%2 = load i32, ptr %ptradd, align 4
|
|
store i32 %2, ptr %a, align 4
|
|
%ptradd1 = getelementptr inbounds i8, ptr %0, i64 56
|
|
%3 = load i32, ptr %ptradd1, align 4
|
|
store i32 %3, ptr %b, align 4
|
|
%4 = load <16 x i32>, ptr %1, align 64
|
|
%5 = extractelement <16 x i32> %4, i64 4
|
|
store i32 %5, ptr %c, align 4
|
|
%6 = load <16 x i32>, ptr %1, align 64
|
|
%7 = extractelement <16 x i32> %6, i64 14
|
|
store i32 %7, ptr %d, align 4
|
|
store i32 3, ptr %j, align 4
|
|
%8 = load <16 x i32>, ptr %1, align 64
|
|
%9 = load i32, ptr %j, align 4
|
|
%sext = sext i32 %9 to i64
|
|
%10 = sub nuw i64 16, %sext
|
|
%11 = extractelement <16 x i32> %8, i64 %10
|
|
store i32 %11, ptr %e, align 4
|
|
%12 = load i32, ptr %j, align 4
|
|
%sext2 = sext i32 %12 to i64
|
|
%13 = sub nuw i64 16, %sext2
|
|
%ptroffset = getelementptr inbounds [4 x i8], ptr %0, i64 %13
|
|
%14 = load i32, ptr %ptroffset, align 4
|
|
store i32 %14, ptr %f, align 4
|
|
ret void
|
|
} |