[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.

llvm-svn: 236523

1 files changed, 38 insertions, 0 deletions

diff --git a/llvm/test/CodeGen/SystemZ/vec-perm-03.ll b/llvm/test/CodeGen/SystemZ/vec-perm-03.ll
index c30a87601a4..663815549c3 100644
--- a/llvm/test/CodeGen/SystemZ/vec-perm-03.ll
+++ b/llvm/test/CodeGen/SystemZ/vec-perm-03.ll
@@ -158,6 +158,44 @@ define <2 x i64> @f12(i64 *%base) {
   ret <2 x i64> %ret
 }
 
+; Test a v4f32 replicating load with no offset.
+define <4 x float> @f13(float *%ptr) {
+; CHECK-LABEL: f13:
+; CHECK: vlrepf %v24, 0(%r2)
+; CHECK: br %r14
+  %scalar = load float, float *%ptr
+  %val = insertelement <4 x float> undef, float %scalar, i32 0
+  %ret = shufflevector <4 x float> %val, <4 x float> undef,
+                       <4 x i32> zeroinitializer
+  ret <4 x float> %ret
+}
+
+; Test a v4f32 replicating load with the maximum in-range offset.
+define <4 x float> @f14(float *%base) {
+; CHECK-LABEL: f14:
+; CHECK: vlrepf %v24, 4092(%r2)
+; CHECK: br %r14
+  %ptr = getelementptr float, float *%base, i64 1023
+  %scalar = load float, float *%ptr
+  %val = insertelement <4 x float> undef, float %scalar, i32 0
+  %ret = shufflevector <4 x float> %val, <4 x float> undef,
+                       <4 x i32> zeroinitializer
+  ret <4 x float> %ret
+}
+
+; Test a v4f32 replicating load with the first out-of-range offset.
+define <4 x float> @f15(float *%base) {
+; CHECK-LABEL: f15:
+; CHECK: aghi %r2, 4096
+; CHECK: vlrepf %v24, 0(%r2)
+; CHECK: br %r14
+  %ptr = getelementptr float, float *%base, i64 1024
+  %scalar = load float, float *%ptr
+  %val = insertelement <4 x float> undef, float %scalar, i32 0
+  %ret = shufflevector <4 x float> %val, <4 x float> undef,
+                       <4 x i32> zeroinitializer
+  ret <4 x float> %ret
+}
 
 ; Test a v2f64 replicating load with no offset.
 define <2 x double> @f16(double *%ptr) {