We generate a shufflevector instruction, so we don't need the builtin
intrinsic.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@35269 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/lib/Target/X86/X86InstrMMX.td b/lib/Target/X86/X86InstrMMX.td
index f6c74f7..cddd765 100644
--- a/lib/Target/X86/X86InstrMMX.td
+++ b/lib/Target/X86/X86InstrMMX.td
@@ -152,6 +152,16 @@
defm MMX_PMADDWD : MMXI_binop_rm_int<0xF5, "pmaddwd", int_x86_mmx_pmadd_wd, 1>;
+// MMX_SHUFFLE_get_shuf_imm xform function: convert vector_shuffle mask to
+// MMX_PSHUF*, MMX_SHUFP* etc. imm.
+def MMX_SHUFFLE_get_shuf_imm : SDNodeXForm<build_vector, [{
+ return getI8Imm(X86::getShuffleSHUFImmediate(N));
+}]>;
+
+def MMX_splat_mask : PatLeaf<(build_vector), [{
+ return X86::isSplatMask(N);
+}], MMX_SHUFFLE_get_shuf_imm>;
+
def MMX_UNPCKH_shuffle_mask : PatLeaf<(build_vector), [{
return X86::isUNPCKHMask(N);
}]>;
@@ -315,16 +325,13 @@
// Splat v2i32
let AddedComplexity = 10 in {
def : Pat<(vector_shuffle (v2i32 VR64:$src), (undef),
+ MMX_splat_mask:$sm),
+ (MMX_PUNPCKHDQrr VR64:$src, VR64:$src)>;
+ def : Pat<(vector_shuffle (v2i32 VR64:$src), (undef),
MMX_UNPCKH_shuffle_mask:$sm),
(MMX_PUNPCKHDQrr VR64:$src, VR64:$src)>;
}
-// FIXME: Temporary workaround because 2-wide shuffle is broken.
-def : Pat<(int_x86_mmx_punpckh_dq VR64:$src1, VR64:$src2),
- (v2i32 (MMX_PUNPCKHDQrr VR64:$src1, VR64:$src2))>;
-def : Pat<(int_x86_mmx_punpckh_dq VR64:$src1, (load addr:$src2)),
- (v2i32 (MMX_PUNPCKHDQrm VR64:$src1, addr:$src2))>;
-
def MMX_X86s2vec : SDNode<"X86ISD::S2VEC", SDTypeProfile<1, 1, []>, []>;
// Scalar to v4i16 / v8i8. The source may be a GR32, but only the lower 8 or