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.
42 lines
1.8 KiB
Plaintext
42 lines
1.8 KiB
Plaintext
// #target: macos-x64
|
|
module test;
|
|
fn int main()
|
|
{
|
|
int[<4>] x = { 1, 2, 3, 4 };
|
|
int[<4>] z = $$matrix_mul(x, x, 2, 2, 2);
|
|
int[<2>] a = { 1, 2 };
|
|
int[<3>] b = { 1, 2, 3 };
|
|
int[<6>] c = $$matrix_mul(a, b, 2, 1, 3);
|
|
c = $$matrix_transpose(c, 2, 3);
|
|
return 0;
|
|
}
|
|
|
|
/* #expect: test.ll
|
|
|
|
define i32 @main() #0 {
|
|
entry:
|
|
%x = alloca <4 x i32>, align 16
|
|
%z = alloca <4 x i32>, align 16
|
|
%a = alloca <2 x i32>, align 8
|
|
%b = alloca <3 x i32>, align 16
|
|
%c = alloca <6 x i32>, align 32
|
|
store <4 x i32> <i32 1, i32 2, i32 3, i32 4>, ptr %x, align 16
|
|
%0 = load <4 x i32>, ptr %x, align 16
|
|
%1 = load <4 x i32>, ptr %x, align 16
|
|
%2 = call <4 x i32> @llvm.matrix.multiply.v4i32.v4i32.v4i32(<4 x i32> %0, <4 x i32> %1, i32 2, i32 2, i32 2)
|
|
store <4 x i32> %2, ptr %z, align 16
|
|
store <2 x i32> <i32 1, i32 2>, ptr %a, align 8
|
|
store <4 x i32> <i32 1, i32 2, i32 3, i32 undef>, ptr %b, align 16
|
|
%3 = load <2 x i32>, ptr %a, align 8
|
|
%4 = load <4 x i32>, ptr %b, align 16
|
|
%extractvec = shufflevector <4 x i32> %4, <4 x i32> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%5 = call <6 x i32> @llvm.matrix.multiply.v6i32.v2i32.v3i32(<2 x i32> %3, <3 x i32> %extractvec, i32 2, i32 1, i32 3)
|
|
%expandvec = shufflevector <6 x i32> %5, <6 x i32> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 poison, i32 poison>
|
|
store <8 x i32> %expandvec, ptr %c, align 32
|
|
%6 = load <8 x i32>, ptr %c, align 32
|
|
%extractvec1 = shufflevector <8 x i32> %6, <8 x i32> poison, <6 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5>
|
|
%7 = call <6 x i32> @llvm.matrix.transpose.v6i32(<6 x i32> %extractvec1, i32 2, i32 3)
|
|
%expandvec2 = shufflevector <6 x i32> %7, <6 x i32> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 poison, i32 poison>
|
|
store <8 x i32> %expandvec2, ptr %c, align 32
|
|
ret i32 0
|
|
} |