llvm/test/CodeGen/SystemZ/vec-move-07.ll
Ulrich Weigand 878c6281d3 [SystemZ] Add CodeGen support for v4f32
The architecture doesn't really have any native v4f32 operations except
v4f32->v2f64 and v2f64->v4f32 conversions, with only half of the v4f32
elements being used.  Even so, using vector registers for <4 x float>
and scalarising individual operations is much better than generating
completely scalar code, since there's much less register pressure.
It's also more efficient to do v4f32 comparisons by extending to 2
v2f64s, comparing those, then packing the result.

This particularly helps with llvmpipe.

Based on a patch by Richard Sandiford.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236523 91177308-0d34-0410-b5e6-96231b3b80d8
2015-05-05 19:27:45 +00:00

58 lines
1.3 KiB
LLVM

; Test scalar_to_vector expansion.
;
; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z13 | FileCheck %s
; Test v16i8.
define <16 x i8> @f1(i8 %val) {
; CHECK-LABEL: f1:
; CHECK: vlvgb %v24, %r2, 0
; CHECK: br %r14
%ret = insertelement <16 x i8> undef, i8 %val, i32 0
ret <16 x i8> %ret
}
; Test v8i16.
define <8 x i16> @f2(i16 %val) {
; CHECK-LABEL: f2:
; CHECK: vlvgh %v24, %r2, 0
; CHECK: br %r14
%ret = insertelement <8 x i16> undef, i16 %val, i32 0
ret <8 x i16> %ret
}
; Test v4i32.
define <4 x i32> @f3(i32 %val) {
; CHECK-LABEL: f3:
; CHECK: vlvgf %v24, %r2, 0
; CHECK: br %r14
%ret = insertelement <4 x i32> undef, i32 %val, i32 0
ret <4 x i32> %ret
}
; Test v2i64. Here we load %val into both halves.
define <2 x i64> @f4(i64 %val) {
; CHECK-LABEL: f4:
; CHECK: vlvgp %v24, %r2, %r2
; CHECK: br %r14
%ret = insertelement <2 x i64> undef, i64 %val, i32 0
ret <2 x i64> %ret
}
; Test v4f32, which is just a move.
define <4 x float> @f5(float %val) {
; CHECK-LABEL: f5:
; CHECK: vlr %v24, %v0
; CHECK: br %r14
%ret = insertelement <4 x float> undef, float %val, i32 0
ret <4 x float> %ret
}
; Likewise v2f64.
define <2 x double> @f6(double %val) {
; CHECK-LABEL: f6:
; CHECK: vlr %v24, %v0
; CHECK: br %r14
%ret = insertelement <2 x double> undef, double %val, i32 0
ret <2 x double> %ret
}