// in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
//===----------------------------------------------------------------------===//
-class sse12_move_rr<RegisterClass RC, ValueType vt, string asm> :
+class sse12_move_rr<RegisterClass RC, SDNode OpNode, ValueType vt, string asm> :
SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), asm,
- [(set (vt VR128:$dst), (movl VR128:$src1, (scalar_to_vector RC:$src2)))]>;
+ [(set VR128:$dst, (vt (OpNode VR128:$src1,
+ (scalar_to_vector RC:$src2))))]>;
// Loading from memory automatically zeroing upper bits.
class sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
[(set RC:$dst, (mem_pat addr:$src))]>;
// AVX
-def VMOVSSrr : sse12_move_rr<FR32, v4f32,
+def VMOVSSrr : sse12_move_rr<FR32, X86Movss, v4f32,
"movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V,
VEX_LIG;
-def VMOVSDrr : sse12_move_rr<FR64, v2f64,
+def VMOVSDrr : sse12_move_rr<FR64, X86Movsd, v2f64,
"movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V,
VEX_LIG;
// SSE1 & 2
let Constraints = "$src1 = $dst" in {
- def MOVSSrr : sse12_move_rr<FR32, v4f32,
+ def MOVSSrr : sse12_move_rr<FR32, X86Movss, v4f32,
"movss\t{$src2, $dst|$dst, $src2}">, XS;
- def MOVSDrr : sse12_move_rr<FR64, v2f64,
+ def MOVSDrr : sse12_move_rr<FR64, X86Movsd, v2f64,
"movsd\t{$src2, $dst|$dst, $src2}">, XD;
// For the disassembler
// Patterns
let Predicates = [HasAVX] in {
let AddedComplexity = 15 in {
- // Extract the low 32-bit value from one vector and insert it into another.
- def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
- (VMOVSSrr (v4f32 VR128:$src1),
- (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
- def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
- (VMOVSSrr (v4i32 VR128:$src1),
- (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
-
- // Extract the low 64-bit value from one vector and insert it into another.
- def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
- (VMOVSDrr (v2f64 VR128:$src1),
- (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
- def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
- (VMOVSDrr (v2i64 VR128:$src1),
- (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
-
- // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
- def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
- (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
- def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
- (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
-
// Move scalar to XMM zero-extended, zeroing a VR128 then do a
// MOVS{S,D} to the lower bits.
def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
(EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
// Shuffle with VMOVSS
- def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
- (VMOVSSrr VR128:$src1, FR32:$src2)>;
def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
(VMOVSSrr (v4i32 VR128:$src1),
(EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
(EXTRACT_SUBREG (v8f32 VR256:$src2), sub_ss)), sub_xmm)>;
// Shuffle with VMOVSD
- def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
- (VMOVSDrr VR128:$src1, FR64:$src2)>;
def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
(VMOVSDrr (v2i64 VR128:$src1),
(EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
let Predicates = [HasSSE1] in {
let AddedComplexity = 15 in {
- // Extract the low 32-bit value from one vector and insert it into another.
- def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
- (MOVSSrr (v4f32 VR128:$src1),
- (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
- def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
- (MOVSSrr (v4i32 VR128:$src1),
- (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
-
// Move scalar to XMM zero-extended, zeroing a VR128 then do a
// MOVSS to the lower bits.
def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
(EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
// Shuffle with MOVSS
- def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
- (MOVSSrr VR128:$src1, FR32:$src2)>;
def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
(MOVSSrr (v4i32 VR128:$src1),
(EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
let Predicates = [HasSSE2] in {
let AddedComplexity = 15 in {
- // Extract the low 64-bit value from one vector and insert it into another.
- def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
- (MOVSDrr (v2f64 VR128:$src1),
- (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
- def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
- (MOVSDrr (v2i64 VR128:$src1),
- (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
-
- // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
- def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
- (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
- def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
- (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
-
// Move scalar to XMM zero-extended, zeroing a VR128 then do a
// MOVSD to the lower bits.
def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
(EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
// Shuffle with MOVSD
- def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
- (MOVSDrr VR128:$src1, FR64:$src2)>;
def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
(MOVSDrr (v2i64 VR128:$src1),
(EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;