(v4f32 (movhlps VR128:$src1, VR128:$src2)))]>;
}
-def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
- (MOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
-let AddedComplexity = 20 in {
- def : Pat<(v4f32 (movddup VR128:$src, (undef))),
- (MOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
- def : Pat<(v2i64 (movddup VR128:$src, (undef))),
- (MOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
+let Predicates = [HasAVX] in {
+ // MOVHPS patterns
+ def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
+ (VMOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
+ def : Pat<(X86Movlhps VR128:$src1,
+ (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
+ (VMOVHPSrm VR128:$src1, addr:$src2)>;
+ def : Pat<(X86Movlhps VR128:$src1,
+ (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
+ (VMOVHPSrm VR128:$src1, addr:$src2)>;
+
+ // MOVLHPS patterns
+ let AddedComplexity = 20 in {
+ def : Pat<(v4f32 (movddup VR128:$src, (undef))),
+ (VMOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
+ def : Pat<(v2i64 (movddup VR128:$src, (undef))),
+ (VMOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
+
+ // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
+ def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
+ (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
+ }
+ def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
+ (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
+ def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
+ (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
+ def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
+ (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
+
+ // MOVHLPS patterns
+ let AddedComplexity = 20 in {
+ // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
+ def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
+ (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
+
+ // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
+ def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
+ (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
+ def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
+ (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
+ }
+
+ def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
+ (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
+ def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
+ (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
+}
+
+let Predicates = [HasSSE1] in {
+ // MOVHPS patterns
+ def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
+ (MOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
+ def : Pat<(X86Movlhps VR128:$src1,
+ (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
+ (MOVHPSrm VR128:$src1, addr:$src2)>;
+ def : Pat<(X86Movlhps VR128:$src1,
+ (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
+ (MOVHPSrm VR128:$src1, addr:$src2)>;
+
+ // MOVLHPS patterns
+ let AddedComplexity = 20 in {
+ def : Pat<(v4f32 (movddup VR128:$src, (undef))),
+ (MOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
+ def : Pat<(v2i64 (movddup VR128:$src, (undef))),
+ (MOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
+
+ // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
+ def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
+ (MOVLHPSrr VR128:$src1, VR128:$src2)>;
+ }
+ def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
+ (MOVLHPSrr VR128:$src1, VR128:$src2)>;
+ def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
+ (MOVLHPSrr VR128:$src1, VR128:$src2)>;
+ def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
+ (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
+
+ // MOVHLPS patterns
+ let AddedComplexity = 20 in {
+ // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
+ def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
+ (MOVHLPSrr VR128:$src1, VR128:$src2)>;
+
+ // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
+ def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
+ (MOVHLPSrr VR128:$src1, VR128:$src1)>;
+ def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
+ (MOVHLPSrr VR128:$src1, VR128:$src1)>;
+ }
+
+ def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
+ (MOVHLPSrr VR128:$src1, VR128:$src2)>;
+ def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
+ (MOVHLPSrr VR128:$src1, VR128:$src2)>;
}
//===----------------------------------------------------------------------===//
memopv2f64, SSEPackedDouble>, TB, OpSize;
}
+let Predicates = [HasSSE1] in {
+ def : Pat<(v4f32 (X86Shufps VR128:$src1,
+ (memopv4f32 addr:$src2), (i8 imm:$imm))),
+ (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
+ def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+ (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
+ def : Pat<(v4i32 (X86Shufps VR128:$src1,
+ (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
+ (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
+ def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+ (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
+ // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
+ // fall back to this for SSE1)
+ def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
+ (SHUFPSrri VR128:$src2, VR128:$src1,
+ (SHUFFLE_get_shuf_imm VR128:$src3))>;
+ // Special unary SHUFPSrri case.
+ def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
+ (SHUFPSrri VR128:$src1, VR128:$src1,
+ (SHUFFLE_get_shuf_imm VR128:$src3))>;
+}
+
+let Predicates = [HasSSE2] in {
+ // Special binary v4i32 shuffle cases with SHUFPS.
+ def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
+ (SHUFPSrri VR128:$src1, VR128:$src2,
+ (SHUFFLE_get_shuf_imm VR128:$src3))>;
+ def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
+ (bc_v4i32 (memopv2i64 addr:$src2)))),
+ (SHUFPSrmi VR128:$src1, addr:$src2,
+ (SHUFFLE_get_shuf_imm VR128:$src3))>;
+ // Special unary SHUFPDrri cases.
+ def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
+ (SHUFPDrri VR128:$src1, VR128:$src1,
+ (SHUFFLE_get_shuf_imm VR128:$src3))>;
+ def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
+ (SHUFPDrri VR128:$src1, VR128:$src1,
+ (SHUFFLE_get_shuf_imm VR128:$src3))>;
+ // Special binary v2i64 shuffle cases using SHUFPDrri.
+ def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
+ (SHUFPDrri VR128:$src1, VR128:$src2,
+ (SHUFFLE_get_shuf_imm VR128:$src3))>;
+ // Generic SHUFPD patterns
+ def : Pat<(v2f64 (X86Shufps VR128:$src1,
+ (memopv2f64 addr:$src2), (i8 imm:$imm))),
+ (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
+ def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+ (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
+ def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+ (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
+}
+
+let Predicates = [HasAVX] in {
+ def : Pat<(v4f32 (X86Shufps VR128:$src1,
+ (memopv4f32 addr:$src2), (i8 imm:$imm))),
+ (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
+ def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+ (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
+ def : Pat<(v4i32 (X86Shufps VR128:$src1,
+ (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
+ (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
+ def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+ (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
+ // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
+ // fall back to this for SSE1)
+ def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
+ (VSHUFPSrri VR128:$src2, VR128:$src1,
+ (SHUFFLE_get_shuf_imm VR128:$src3))>;
+ // Special unary SHUFPSrri case.
+ def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
+ (VSHUFPSrri VR128:$src1, VR128:$src1,
+ (SHUFFLE_get_shuf_imm VR128:$src3))>;
+ // Special binary v4i32 shuffle cases with SHUFPS.
+ def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
+ (VSHUFPSrri VR128:$src1, VR128:$src2,
+ (SHUFFLE_get_shuf_imm VR128:$src3))>;
+ def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
+ (bc_v4i32 (memopv2i64 addr:$src2)))),
+ (VSHUFPSrmi VR128:$src1, addr:$src2,
+ (SHUFFLE_get_shuf_imm VR128:$src3))>;
+ // Special unary SHUFPDrri cases.
+ def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
+ (VSHUFPDrri VR128:$src1, VR128:$src1,
+ (SHUFFLE_get_shuf_imm VR128:$src3))>;
+ def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
+ (VSHUFPDrri VR128:$src1, VR128:$src1,
+ (SHUFFLE_get_shuf_imm VR128:$src3))>;
+ // Special binary v2i64 shuffle cases using SHUFPDrri.
+ def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
+ (VSHUFPDrri VR128:$src1, VR128:$src2,
+ (SHUFFLE_get_shuf_imm VR128:$src3))>;
+ // Generic VSHUFPD patterns
+ def : Pat<(v2f64 (X86Shufps VR128:$src1,
+ (memopv2f64 addr:$src2), (i8 imm:$imm))),
+ (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
+ def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+ (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
+ def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+ (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
+}
+
//===----------------------------------------------------------------------===//
// SSE 1 & 2 - Unpack Instructions
//===----------------------------------------------------------------------===//
!strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>, REX_W;
}
-// Mask creation
-defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
- "movmskps", SSEPackedSingle>, VEX;
-defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
- "movmskpd", SSEPackedDouble>, OpSize,
- VEX;
-defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
- "movmskps", SSEPackedSingle>, VEX;
-defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
- "movmskpd", SSEPackedDouble>, OpSize,
- VEX;
defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
SSEPackedSingle>, TB;
defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
SSEPackedDouble>, TB, OpSize;
-// X86fgetsign
-def MOVMSKPDrr32_alt : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins FR64:$src),
- "movmskpd\t{$src, $dst|$dst, $src}",
- [(set GR32:$dst, (X86fgetsign FR64:$src))], SSEPackedDouble>, TB, OpSize;
-def MOVMSKPDrr64_alt : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins FR64:$src),
- "movmskpd\t{$src, $dst|$dst, $src}",
- [(set GR64:$dst, (X86fgetsign FR64:$src))], SSEPackedDouble>, TB, OpSize;
-def MOVMSKPSrr32_alt : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins FR32:$src),
- "movmskps\t{$src, $dst|$dst, $src}",
- [(set GR32:$dst, (X86fgetsign FR32:$src))], SSEPackedSingle>, TB;
-def MOVMSKPSrr64_alt : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins FR32:$src),
- "movmskps\t{$src, $dst|$dst, $src}",
- [(set GR64:$dst, (X86fgetsign FR32:$src))], SSEPackedSingle>, TB;
-
-// Assembler Only
-def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
- "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
-def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
- "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
- VEX;
-def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
- "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
-def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
- "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
- VEX;
+def : Pat<(i32 (X86fgetsign FR32:$src)),
+ (MOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
+ sub_ss))>, Requires<[HasSSE1]>;
+def : Pat<(i64 (X86fgetsign FR32:$src)),
+ (MOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
+ sub_ss))>, Requires<[HasSSE1]>;
+def : Pat<(i32 (X86fgetsign FR64:$src)),
+ (MOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
+ sub_sd))>, Requires<[HasSSE2]>;
+def : Pat<(i64 (X86fgetsign FR64:$src)),
+ (MOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
+ sub_sd))>, Requires<[HasSSE2]>;
+
+let Predicates = [HasAVX] in {
+ defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
+ "movmskps", SSEPackedSingle>, TB, VEX;
+ defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
+ "movmskpd", SSEPackedDouble>, TB, OpSize,
+ VEX;
+ defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
+ "movmskps", SSEPackedSingle>, TB, VEX;
+ defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
+ "movmskpd", SSEPackedDouble>, TB, OpSize,
+ VEX;
+
+ def : Pat<(i32 (X86fgetsign FR32:$src)),
+ (VMOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
+ sub_ss))>;
+ def : Pat<(i64 (X86fgetsign FR32:$src)),
+ (VMOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
+ sub_ss))>;
+ def : Pat<(i32 (X86fgetsign FR64:$src)),
+ (VMOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
+ sub_sd))>;
+ def : Pat<(i64 (X86fgetsign FR64:$src)),
+ (VMOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
+ sub_sd))>;
+
+ // Assembler Only
+ def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
+ "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
+ def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
+ "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
+ VEX;
+ def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
+ "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
+ def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
+ "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
+ VEX;
+}
//===----------------------------------------------------------------------===//
// SSE 1 & 2 - Misc aliasing of packed SSE 1 & 2 instructions
multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
SDNode OpNode> {
defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
- FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, VEX_4V;
+ FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, TB, VEX_4V;
defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
- FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, OpSize, VEX_4V;
+ FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, TB, OpSize, VEX_4V;
let Constraints = "$src1 = $dst" in {
defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
///
multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
SDNode OpNode> {
- let Pattern = []<dag> in {
- defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
- !strconcat(OpcodeStr, "ps"), f128mem,
- [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
- [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
- (memopv2i64 addr:$src2)))], 0>, VEX_4V;
-
- defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
- !strconcat(OpcodeStr, "pd"), f128mem,
- [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
- (bc_v2i64 (v2f64 VR128:$src2))))],
- [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
- (memopv2i64 addr:$src2)))], 0>,
- OpSize, VEX_4V;
- }
+ // In AVX no need to add a pattern for 128-bit logical rr ps, because they
+ // are all promoted to v2i64, and the patterns are covered by the int
+ // version. This is needed in SSE only, because v2i64 isn't supported on
+ // SSE1, but only on SSE2.
+ defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
+ !strconcat(OpcodeStr, "ps"), f128mem, [],
+ [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
+ (memopv2i64 addr:$src2)))], 0>, TB, VEX_4V;
+
+ defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
+ !strconcat(OpcodeStr, "pd"), f128mem,
+ [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
+ (bc_v2i64 (v2f64 VR128:$src2))))],
+ [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
+ (memopv2i64 addr:$src2)))], 0>,
+ TB, OpSize, VEX_4V;
let Constraints = "$src1 = $dst" in {
defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
!strconcat(OpcodeStr, "ps"), f128mem,
!strconcat(OpcodeStr, "ps"), f256mem,
[(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
[(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
- (memopv4i64 addr:$src2)))], 0>, VEX_4V;
+ (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V;
defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
!strconcat(OpcodeStr, "pd"), f256mem,
(bc_v4i64 (v4f64 VR256:$src2))))],
[(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
(memopv4i64 addr:$src2)))], 0>,
- OpSize, VEX_4V;
+ TB, OpSize, VEX_4V;
}
// AVX 256-bit packed logical ops forms
}
/// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
-multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
- SDNode OpNode, Intrinsic F32Int> {
+multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
!strconcat(OpcodeStr,
"ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
- def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1, f32mem:$src2),
+ def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
!strconcat(OpcodeStr,
- "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- []>, XS, Requires<[HasAVX, OptForSize]>;
- def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
- !strconcat(OpcodeStr,
- "ss\t{$src, $dst, $dst|$dst, $dst, $src}"),
- [(set VR128:$dst, (F32Int VR128:$src))]>;
- def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
+ "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
+ def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
+ (ins ssmem:$src1, VR128:$src2),
!strconcat(OpcodeStr,
- "ss\t{$src, $dst, $dst|$dst, $dst, $src}"),
- [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
+ "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
}
/// sse1_fp_unop_p - SSE1 unops in packed form.
}
/// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
-multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
- SDNode OpNode, Intrinsic F64Int> {
+multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
!strconcat(OpcodeStr,
"sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
- def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst),
- (ins FR64:$src1, f64mem:$src2),
+ def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
+ !strconcat(OpcodeStr,
+ "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
+ def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
+ (ins VR128:$src1, sdmem:$src2),
!strconcat(OpcodeStr,
"sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
- def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
- !strconcat(OpcodeStr, "sd\t{$src, $dst, $dst|$dst, $dst, $src}"),
- [(set VR128:$dst, (F64Int VR128:$src))]>;
- def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
- !strconcat(OpcodeStr, "sd\t{$src, $dst, $dst|$dst, $dst, $src}"),
- [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
}
/// sse2_fp_unop_p - SSE2 unops in vector forms.
let Predicates = [HasAVX] in {
// Square root.
- defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt", fsqrt, int_x86_sse_sqrt_ss>,
- sse2_fp_unop_s_avx<0x51, "vsqrt", fsqrt, int_x86_sse2_sqrt_sd>,
- VEX_4V;
+ defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
+ sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V;
defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
// Reciprocal approximations. Note that these typically require refinement
// in order to obtain suitable precision.
- defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt", X86frsqrt,
- int_x86_sse_rsqrt_ss>, VEX_4V;
+ defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V;
defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
- defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp", X86frcp, int_x86_sse_rcp_ss>,
- VEX_4V;
+ defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V;
defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
def : Pat<(f32 (fsqrt FR32:$src)),
(VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
+def : Pat<(f32 (fsqrt (load addr:$src))),
+ (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
+ Requires<[HasAVX, OptForSize]>;
def : Pat<(f64 (fsqrt FR64:$src)),
(VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
def : Pat<(f64 (fsqrt (load addr:$src))),
(VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
Requires<[HasAVX, OptForSize]>;
-def : Pat<(f32 (fsqrt (load addr:$src))),
- (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
+
+def : Pat<(f32 (X86frsqrt FR32:$src)),
+ (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
+def : Pat<(f32 (X86frsqrt (load addr:$src))),
+ (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
Requires<[HasAVX, OptForSize]>;
+def : Pat<(f32 (X86frcp FR32:$src)),
+ (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
+def : Pat<(f32 (X86frcp (load addr:$src))),
+ (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
+ Requires<[HasAVX, OptForSize]>;
+
+let Predicates = [HasAVX] in {
+def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
+ (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
+ (VSQRTSSr (f32 (IMPLICIT_DEF)),
+ (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
+ sub_ss)>;
+def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
+ (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
+
+def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
+ (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)),
+ (VSQRTSDr (f64 (IMPLICIT_DEF)),
+ (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd)),
+ sub_sd)>;
+def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
+ (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
+
+def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
+ (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
+ (VRSQRTSSr (f32 (IMPLICIT_DEF)),
+ (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
+ sub_ss)>;
+def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
+ (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
+
+def : Pat<(int_x86_sse_rcp_ss VR128:$src),
+ (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
+ (VRCPSSr (f32 (IMPLICIT_DEF)),
+ (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
+ sub_ss)>;
+def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
+ (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
+}
+
// Square root.
defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
def VPANDNrr : PDI<0xDF, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
"vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
- VR128:$src2)))]>, VEX_4V;
+ [(set VR128:$dst,
+ (v2i64 (X86andnp VR128:$src1, VR128:$src2)))]>,VEX_4V;
def VPANDNrm : PDI<0xDF, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
"vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
- (memopv2i64 addr:$src2))))]>,
- VEX_4V;
+ [(set VR128:$dst, (X86andnp VR128:$src1,
+ (memopv2i64 addr:$src2)))]>, VEX_4V;
}
}
0>, VEX_4V;
defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d, 0,
0>, VEX_4V;
+
+ def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
+ (VPCMPEQBrr VR128:$src1, VR128:$src2)>;
+ def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, (memop addr:$src2))),
+ (VPCMPEQBrm VR128:$src1, addr:$src2)>;
+ def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
+ (VPCMPEQWrr VR128:$src1, VR128:$src2)>;
+ def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, (memop addr:$src2))),
+ (VPCMPEQWrm VR128:$src1, addr:$src2)>;
+ def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
+ (VPCMPEQDrr VR128:$src1, VR128:$src2)>;
+ def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, (memop addr:$src2))),
+ (VPCMPEQDrm VR128:$src1, addr:$src2)>;
+
+ def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
+ (VPCMPGTBrr VR128:$src1, VR128:$src2)>;
+ def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, (memop addr:$src2))),
+ (VPCMPGTBrm VR128:$src1, addr:$src2)>;
+ def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
+ (VPCMPGTWrr VR128:$src1, VR128:$src2)>;
+ def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, (memop addr:$src2))),
+ (VPCMPGTWrm VR128:$src1, addr:$src2)>;
+ def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
+ (VPCMPGTDrr VR128:$src1, VR128:$src2)>;
+ def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, (memop addr:$src2))),
+ (VPCMPGTDrm VR128:$src1, addr:$src2)>;
}
let Constraints = "$src1 = $dst" in {
// SSE2 with ImmT == Imm8 and XD prefix.
defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
VEX;
+
+ let AddedComplexity = 5 in
+ def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
+ (VPSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
+ // Unary v4f32 shuffle with VPSHUF* in order to fold a load.
+ def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
+ (VPSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
+
+ def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
+ (i8 imm:$imm))),
+ (VPSHUFDmi addr:$src1, imm:$imm)>, Requires<[HasAVX]>;
+ def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
+ (i8 imm:$imm))),
+ (VPSHUFDmi addr:$src1, imm:$imm)>;
+ def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
+ (VPSHUFDri VR128:$src1, imm:$imm)>, Requires<[HasAVX]>;
+ def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
+ (VPSHUFDri VR128:$src1, imm:$imm)>, Requires<[HasAVX]>;
+ def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
+ (VPSHUFHWri VR128:$src, imm:$imm)>;
+ def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
+ (i8 imm:$imm))),
+ (VPSHUFHWmi addr:$src, imm:$imm)>;
+ def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
+ (VPSHUFLWri VR128:$src, imm:$imm)>;
+ def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
+ (i8 imm:$imm))),
+ (VPSHUFLWmi addr:$src, imm:$imm)>;
}
let Predicates = [HasSSE2] in {
// SSE2 with ImmT == Imm8 and XD prefix.
defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
+
+ let AddedComplexity = 5 in
+ def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
+ (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
+ // Unary v4f32 shuffle with PSHUF* in order to fold a load.
+ def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
+ (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
+
+ def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
+ (i8 imm:$imm))),
+ (PSHUFDmi addr:$src1, imm:$imm)>;
+ def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
+ (i8 imm:$imm))),
+ (PSHUFDmi addr:$src1, imm:$imm)>;
+ def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
+ (PSHUFDri VR128:$src1, imm:$imm)>;
+ def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
+ (PSHUFDri VR128:$src1, imm:$imm)>;
+ def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
+ (PSHUFHWri VR128:$src, imm:$imm)>;
+ def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
+ (i8 imm:$imm))),
+ (PSHUFHWmi addr:$src, imm:$imm)>;
+ def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
+ (PSHUFLWri VR128:$src, imm:$imm)>;
+ def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
+ (i8 imm:$imm))),
+ (PSHUFLWmi addr:$src, imm:$imm)>;
}
//===---------------------------------------------------------------------===//
(VCVTDQ2PDYrm addr:$src)>;
//===---------------------------------------------------------------------===//
-// SSE3 - Move Instructions
+// SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
//===---------------------------------------------------------------------===//
-
-//===---------------------------------------------------------------------===//
-// Replicate Single FP - MOVSHDUP and MOVSLDUP
-//
multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
ValueType vt, RegisterClass RC, PatFrag mem_frag,
X86MemOperand x86memop> {
}
//===---------------------------------------------------------------------===//
-// Replicate Double FP - MOVDDUP
-//
+// SSE3 - Replicate Double FP - MOVDDUP
+//===---------------------------------------------------------------------===//
+
multiclass sse3_replicate_dfp<string OpcodeStr> {
def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
(undef))))]>;
}
+// FIXME: Merge with above classe when there're patterns for the ymm version
multiclass sse3_replicate_dfp_y<string OpcodeStr> {
-def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
- !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- []>;
-def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
- !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- []>;
+let Predicates = [HasAVX] in {
+ def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
+ !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
+ []>;
+ def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
+ !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
+ []>;
+ }
+}
+
+defm MOVDDUP : sse3_replicate_dfp<"movddup">;
+defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
+defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
+
+let Predicates = [HasSSE3] in {
+ def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
+ (undef)),
+ (MOVDDUPrm addr:$src)>;
+ let AddedComplexity = 5 in {
+ def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
+ def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
+ (MOVDDUPrm addr:$src)>;
+ def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
+ def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
+ (MOVDDUPrm addr:$src)>;
+ }
+ def : Pat<(X86Movddup (memopv2f64 addr:$src)),
+ (MOVDDUPrm addr:$src)>;
+ def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
+ (MOVDDUPrm addr:$src)>;
+ def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
+ (MOVDDUPrm addr:$src)>;
+ def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
+ (MOVDDUPrm addr:$src)>;
+ def : Pat<(X86Movddup (bc_v2f64
+ (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
+ (MOVDDUPrm addr:$src)>;
}
let Predicates = [HasAVX] in {
- // FIXME: Merge above classes when we have patterns for the ymm version
- defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
- defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
+ def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
+ (undef)),
+ (VMOVDDUPrm addr:$src)>;
+ let AddedComplexity = 5 in {
+ def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
+ def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
+ (VMOVDDUPrm addr:$src)>;
+ def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
+ def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
+ (VMOVDDUPrm addr:$src)>;
+ }
+ def : Pat<(X86Movddup (memopv2f64 addr:$src)),
+ (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
+ def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
+ (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
+ def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
+ (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
+ def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
+ (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
+ def : Pat<(X86Movddup (bc_v2f64
+ (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
+ (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
}
-defm MOVDDUP : sse3_replicate_dfp<"movddup">;
-// Move Unaligned Integer
+//===---------------------------------------------------------------------===//
+// SSE3 - Move Unaligned Integer
+//===---------------------------------------------------------------------===//
+
let Predicates = [HasAVX] in {
def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
"vlddqu\t{$src, $dst|$dst, $src}",
"lddqu\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
-def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
- (undef)),
- (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
-
-// Several Move patterns
-let AddedComplexity = 5 in {
-def : Pat<(movddup (memopv2f64 addr:$src), (undef)),
- (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
-def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
- (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
-def : Pat<(movddup (memopv2i64 addr:$src), (undef)),
- (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
-def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
- (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
-}
-
//===---------------------------------------------------------------------===//
// SSE3 - Arithmetic
//===---------------------------------------------------------------------===//
(PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
}
-// Special unary SHUFPSrri case.
-def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
- (SHUFPSrri VR128:$src1, VR128:$src1,
- (SHUFFLE_get_shuf_imm VR128:$src3))>;
-let AddedComplexity = 5 in
-def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
- (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
- Requires<[HasSSE2]>;
-// Special unary SHUFPDrri case.
-def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
- (SHUFPDrri VR128:$src1, VR128:$src1,
- (SHUFFLE_get_shuf_imm VR128:$src3))>,
- Requires<[HasSSE2]>;
-// Special unary SHUFPDrri case.
-def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
- (SHUFPDrri VR128:$src1, VR128:$src1,
- (SHUFFLE_get_shuf_imm VR128:$src3))>,
- Requires<[HasSSE2]>;
-// Unary v4f32 shuffle with PSHUF* in order to fold a load.
-def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
- (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
- Requires<[HasSSE2]>;
-
-// Special binary v4i32 shuffle cases with SHUFPS.
-def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
- (SHUFPSrri VR128:$src1, VR128:$src2,
- (SHUFFLE_get_shuf_imm VR128:$src3))>,
- Requires<[HasSSE2]>;
-def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (bc_v4i32 (memopv2i64 addr:$src2)))),
- (SHUFPSrmi VR128:$src1, addr:$src2,
- (SHUFFLE_get_shuf_imm VR128:$src3))>,
- Requires<[HasSSE2]>;
-// Special binary v2i64 shuffle cases using SHUFPDrri.
-def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
- (SHUFPDrri VR128:$src1, VR128:$src2,
- (SHUFFLE_get_shuf_imm VR128:$src3))>,
- Requires<[HasSSE2]>;
-
-let AddedComplexity = 20 in {
-// vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
-def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
- (MOVLHPSrr VR128:$src1, VR128:$src2)>;
-
-// vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
-def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
- (MOVHLPSrr VR128:$src1, VR128:$src2)>;
-
-// vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
-def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
- (MOVHLPSrr VR128:$src1, VR128:$src1)>;
-def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
- (MOVHLPSrr VR128:$src1, VR128:$src1)>;
-}
-
let AddedComplexity = 20 in {
// vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
Requires<[HasSSE2]>;
}
-// vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
-// fall back to this for SSE1)
-def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
- (SHUFPSrri VR128:$src2, VR128:$src1,
- (SHUFFLE_get_shuf_imm VR128:$src3))>;
-
// Set lowest element and zero upper elements.
def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
(MOVZPQILo2PQIrr VR128:$src)>, Requires<[HasSSE2]>;
0>, VEX_4V;
defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
0>, VEX_4V;
+
+ def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
+ (VPCMPEQQrr VR128:$src1, VR128:$src2)>;
+ def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
+ (VPCMPEQQrm VR128:$src1, addr:$src2)>;
}
let Constraints = "$src1 = $dst" in {
(bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
}
-let Predicates = [HasAVX] in
+let Predicates = [HasAVX] in {
defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
0>, VEX_4V;
+
+ def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
+ (VPCMPGTQrr VR128:$src1, VR128:$src2)>;
+ def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
+ (VPCMPGTQrm VR128:$src1, addr:$src2)>;
+}
+
let Constraints = "$src1 = $dst" in
defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
(VBROADCASTF128 addr:$src)>;
+def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
+ (VBROADCASTSSY addr:$src)>;
+def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
+ (VBROADCASTSD addr:$src)>;
+def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
+ (VBROADCASTSSY addr:$src)>;
+def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
+ (VBROADCASTSD addr:$src)>;
+
+def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
+ (VBROADCASTSS addr:$src)>;
+def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
+ (VBROADCASTSS addr:$src)>;
+
//===----------------------------------------------------------------------===//
// VINSERTF128 - Insert packed floating-point values
//
def : Pat<(v2f64 (extract_subvector (v4f64 VR256:$src), (i32 0))),
(v2f64 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm))>;
+def : Pat<(v8i16 (extract_subvector (v16i16 VR256:$src), (i32 0))),
+ (v8i16 (EXTRACT_SUBREG (v16i16 VR256:$src), sub_xmm))>;
+def : Pat<(v16i8 (extract_subvector (v32i8 VR256:$src), (i32 0))),
+ (v16i8 (EXTRACT_SUBREG (v32i8 VR256:$src), sub_xmm))>;
+
//===----------------------------------------------------------------------===//
// VMASKMOV - Conditional SIMD Packed Loads and Stores
VR256:$src1, (memopv8i32 addr:$src2), imm:$src3),
(VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
+def : Pat<(v8f32 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
+ (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
+def : Pat<(v8i32 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
+ (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
+def : Pat<(v4i64 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
+ (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
+def : Pat<(v4f64 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
+ (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
+def : Pat<(v32i8 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
+ (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
+def : Pat<(v16i16 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
+ (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
+
//===----------------------------------------------------------------------===//
// VZERO - Zero YMM registers
//
-// Zero All YMM registers
-def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
- [(int_x86_avx_vzeroall)]>, VEX, VEX_L, Requires<[HasAVX]>;
+let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
+ YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
+ // Zero All YMM registers
+ def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
+ [(int_x86_avx_vzeroall)]>, VEX, VEX_L, Requires<[HasAVX]>;
+
+}
// Zero Upper bits of YMM registers
def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
// The AVX version of some but not all of them are described here, and more
// should come in a near future.
-// Shuffle with PSHUFD instruction folding loads. The first two patterns match
-// SSE2 loads, which are always promoted to v2i64. The last one should match
-// the SSE1 case, where the only legal load is v4f32, but there is no PSHUFD
-// in SSE2, how does it ever worked? Anyway, the pattern will remain here until
-// we investigate further.
-def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
- (i8 imm:$imm))),
- (VPSHUFDmi addr:$src1, imm:$imm)>, Requires<[HasAVX]>;
-def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
- (i8 imm:$imm))),
- (PSHUFDmi addr:$src1, imm:$imm)>;
-def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
- (i8 imm:$imm))),
- (PSHUFDmi addr:$src1, imm:$imm)>; // FIXME: has this ever worked?
-
-// Shuffle with PSHUFD instruction.
-def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
- (VPSHUFDri VR128:$src1, imm:$imm)>, Requires<[HasAVX]>;
-def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
- (PSHUFDri VR128:$src1, imm:$imm)>;
-
-def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
- (VPSHUFDri VR128:$src1, imm:$imm)>, Requires<[HasAVX]>;
-def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
- (PSHUFDri VR128:$src1, imm:$imm)>;
-
-// Shuffle with SHUFPD instruction.
-def : Pat<(v2f64 (X86Shufps VR128:$src1,
- (memopv2f64 addr:$src2), (i8 imm:$imm))),
- (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
-def : Pat<(v2f64 (X86Shufps VR128:$src1,
- (memopv2f64 addr:$src2), (i8 imm:$imm))),
- (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
-
-def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
- (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
-def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
- (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
-
-def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
- (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
-def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
- (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
-
-// Shuffle with SHUFPS instruction.
-def : Pat<(v4f32 (X86Shufps VR128:$src1,
- (memopv4f32 addr:$src2), (i8 imm:$imm))),
- (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
-def : Pat<(v4f32 (X86Shufps VR128:$src1,
- (memopv4f32 addr:$src2), (i8 imm:$imm))),
- (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
-
-def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
- (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
-def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
- (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
-
-def : Pat<(v4i32 (X86Shufps VR128:$src1,
- (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
- (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
-def : Pat<(v4i32 (X86Shufps VR128:$src1,
- (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
- (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
-
-def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
- (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
-def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
- (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
-
-// Shuffle with MOVHLPS instruction
-def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
- (MOVHLPSrr VR128:$src1, VR128:$src2)>;
-def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
- (MOVHLPSrr VR128:$src1, VR128:$src2)>;
-
-// Shuffle with MOVDDUP instruction
-def : Pat<(X86Movddup (memopv2f64 addr:$src)),
- (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
-def : Pat<(X86Movddup (memopv2f64 addr:$src)),
- (MOVDDUPrm addr:$src)>;
-
-def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
- (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
-def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
- (MOVDDUPrm addr:$src)>;
-
-def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
- (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
-def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
- (MOVDDUPrm addr:$src)>;
-
-def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
- (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
-def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
- (MOVDDUPrm addr:$src)>;
-
-def : Pat<(X86Movddup (bc_v2f64
- (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
- (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
-def : Pat<(X86Movddup (bc_v2f64
- (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
- (MOVDDUPrm addr:$src)>;
// Shuffle with UNPCKLPS
def : Pat<(v4i64 (X86Unpckhpdy VR256:$src1, VR256:$src2)),
(VUNPCKHPDYrr VR256:$src1, VR256:$src2)>, Requires<[HasAVX]>;
-// Shuffle with MOVLHPS
-def : Pat<(X86Movlhps VR128:$src1,
- (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
- (MOVHPSrm VR128:$src1, addr:$src2)>;
-def : Pat<(X86Movlhps VR128:$src1,
- (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
- (MOVHPSrm VR128:$src1, addr:$src2)>;
-def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
- (MOVLHPSrr VR128:$src1, VR128:$src2)>;
-def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
- (MOVLHPSrr VR128:$src1, VR128:$src2)>;
-def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
- (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
-
// FIXME: Instead of X86Movddup, there should be a X86Unpcklpd here, the problem
// is during lowering, where it's not possible to recognize the load fold cause
// it has two uses through a bitcast. One use disappears at isel time and the
def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
(MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_sd))>;
-// Shuffle with PSHUFHW
-def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
- (PSHUFHWri VR128:$src, imm:$imm)>;
-def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)), (i8 imm:$imm))),
- (PSHUFHWmi addr:$src, imm:$imm)>;
-
-// Shuffle with PSHUFLW
-def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
- (PSHUFLWri VR128:$src, imm:$imm)>;
-def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)), (i8 imm:$imm))),
- (PSHUFLWmi addr:$src, imm:$imm)>;
-
// Shuffle with MOVLPS
def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
(MOVLPSrm VR128:$src1, addr:$src2)>;
def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
(MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_sd))>;
-def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
- (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_sd))>;
+def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
+ (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_sd))>;
// Shuffle with MOVLPD
def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
projects / oota-llvm.git / blobdiff