[X86] Remove the single AdSize indicator and replace it with separate AdSize16/32...

[oota-llvm.git] / lib / Target / X86 / X86InstrSSE.td

diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index e66813fa1516f9758c1ad0f1bdaea934fb07f2ed..971cef12dabd20936d9299304aeb79c6c04597cb 100644 (file)
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -8966,20 +8966,26 @@ defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
                                 int_x86_avx2_maskstore_q_256>, VEX_W;
 
 def: Pat<(masked_store addr:$ptr, (v8i32 VR256:$mask), (v8f32 VR256:$src)),
-         (VPMASKMOVDYmr addr:$ptr, VR256:$mask, VR256:$src)>;
+         (VMASKMOVPSYmr addr:$ptr, VR256:$mask, VR256:$src)>;
 
 def: Pat<(masked_store addr:$ptr, (v8i32 VR256:$mask), (v8i32 VR256:$src)),
          (VPMASKMOVDYmr addr:$ptr, VR256:$mask, VR256:$src)>;
 
+def: Pat<(masked_store addr:$ptr, (v4i32 VR128:$mask), (v4f32 VR128:$src)),
+         (VMASKMOVPSmr addr:$ptr, VR128:$mask, VR128:$src)>;
+
+def: Pat<(masked_store addr:$ptr, (v4i32 VR128:$mask), (v4i32 VR128:$src)),
+         (VPMASKMOVDmr addr:$ptr, VR128:$mask, VR128:$src)>;
+
 def: Pat<(v8f32 (masked_load addr:$ptr, (v8i32 VR256:$mask), undef)),
-         (VPMASKMOVDYrm VR256:$mask, addr:$ptr)>;
+         (VMASKMOVPSYrm VR256:$mask, addr:$ptr)>;
 
 def: Pat<(v8f32 (masked_load addr:$ptr, (v8i32 VR256:$mask),
                              (bc_v8f32 (v8i32 immAllZerosV)))),
-         (VPMASKMOVDYrm VR256:$mask, addr:$ptr)>;
+         (VMASKMOVPSYrm VR256:$mask, addr:$ptr)>;
 
 def: Pat<(v8f32 (masked_load addr:$ptr, (v8i32 VR256:$mask), (v8f32 VR256:$src0))),
-         (VBLENDVPSYrr VR256:$src0, (VPMASKMOVDYrm VR256:$mask, addr:$ptr),
+         (VBLENDVPSYrr VR256:$src0, (VMASKMOVPSYrm VR256:$mask, addr:$ptr),
                        VR256:$mask)>;
 
 def: Pat<(v8i32 (masked_load addr:$ptr, (v8i32 VR256:$mask), undef)),
@@ -8992,21 +8998,42 @@ def: Pat<(v8i32 (masked_load addr:$ptr, (v8i32 VR256:$mask), (v8i32 VR256:$src0)
          (VBLENDVPSYrr VR256:$src0, (VPMASKMOVDYrm VR256:$mask, addr:$ptr),
                        VR256:$mask)>;
 
+def: Pat<(v4f32 (masked_load addr:$ptr, (v4i32 VR128:$mask), undef)),
+         (VMASKMOVPSrm VR128:$mask, addr:$ptr)>;
+
+def: Pat<(v4f32 (masked_load addr:$ptr, (v4i32 VR128:$mask),
+                             (bc_v4f32 (v4i32 immAllZerosV)))),
+         (VMASKMOVPSrm VR128:$mask, addr:$ptr)>;
+
+def: Pat<(v4f32 (masked_load addr:$ptr, (v4i32 VR128:$mask), (v4f32 VR128:$src0))),
+         (VBLENDVPSrr VR128:$src0, (VMASKMOVPSrm VR128:$mask, addr:$ptr),
+                       VR128:$mask)>;
+
+def: Pat<(v4i32 (masked_load addr:$ptr, (v4i32 VR128:$mask), undef)),
+         (VPMASKMOVDrm VR128:$mask, addr:$ptr)>;
+
+def: Pat<(v4i32 (masked_load addr:$ptr, (v4i32 VR128:$mask), (v4i32 immAllZerosV))),
+         (VPMASKMOVDrm VR128:$mask, addr:$ptr)>;
+
+def: Pat<(v4i32 (masked_load addr:$ptr, (v4i32 VR128:$mask), (v4i32 VR128:$src0))),
+         (VBLENDVPSrr VR128:$src0, (VPMASKMOVDrm VR128:$mask, addr:$ptr),
+                       VR128:$mask)>;
+
 def: Pat<(masked_store addr:$ptr, (v4i64 VR256:$mask), (v4f64 VR256:$src)),
-         (VPMASKMOVQYmr addr:$ptr, VR256:$mask, VR256:$src)>;
+         (VMASKMOVPDYmr addr:$ptr, VR256:$mask, VR256:$src)>;
 
 def: Pat<(masked_store addr:$ptr, (v4i64 VR256:$mask), (v4i64 VR256:$src)),
          (VPMASKMOVQYmr addr:$ptr, VR256:$mask, VR256:$src)>;
 
 def: Pat<(v4f64 (masked_load addr:$ptr, (v4i64 VR256:$mask), undef)),
-         (VPMASKMOVQYrm VR256:$mask, addr:$ptr)>;
+         (VMASKMOVPDYrm VR256:$mask, addr:$ptr)>;
 
 def: Pat<(v4f64 (masked_load addr:$ptr, (v4i64 VR256:$mask),
                              (v4f64 immAllZerosV))),
-         (VPMASKMOVQYrm VR256:$mask, addr:$ptr)>;
+         (VMASKMOVPDYrm VR256:$mask, addr:$ptr)>;
 
 def: Pat<(v4f64 (masked_load addr:$ptr, (v4i64 VR256:$mask), (v4f64 VR256:$src0))),
-         (VBLENDVPDYrr VR256:$src0, (VPMASKMOVQYrm VR256:$mask, addr:$ptr),
+         (VBLENDVPDYrr VR256:$src0, (VMASKMOVPDYrm VR256:$mask, addr:$ptr),
                        VR256:$mask)>;
 
 def: Pat<(v4i64 (masked_load addr:$ptr, (v4i64 VR256:$mask), undef)),
@@ -9020,6 +9047,33 @@ def: Pat<(v4i64 (masked_load addr:$ptr, (v4i64 VR256:$mask), (v4i64 VR256:$src0)
          (VBLENDVPDYrr VR256:$src0, (VPMASKMOVQYrm VR256:$mask, addr:$ptr),
                        VR256:$mask)>;
 
+def: Pat<(masked_store addr:$ptr, (v2i64 VR128:$mask), (v2f64 VR128:$src)),
+         (VMASKMOVPDmr addr:$ptr, VR128:$mask, VR128:$src)>;
+
+def: Pat<(masked_store addr:$ptr, (v2i64 VR128:$mask), (v2i64 VR128:$src)),
+         (VPMASKMOVQmr addr:$ptr, VR128:$mask, VR128:$src)>;
+
+def: Pat<(v2f64 (masked_load addr:$ptr, (v2i64 VR128:$mask), undef)),
+         (VMASKMOVPDrm VR128:$mask, addr:$ptr)>;
+
+def: Pat<(v2f64 (masked_load addr:$ptr, (v2i64 VR128:$mask),
+                             (v2f64 immAllZerosV))),
+         (VMASKMOVPDrm VR128:$mask, addr:$ptr)>;
+
+def: Pat<(v2f64 (masked_load addr:$ptr, (v2i64 VR128:$mask), (v2f64 VR128:$src0))),
+         (VBLENDVPDrr VR128:$src0, (VMASKMOVPDrm VR128:$mask, addr:$ptr),
+                       VR128:$mask)>;
+
+def: Pat<(v2i64 (masked_load addr:$ptr, (v2i64 VR128:$mask), undef)),
+         (VPMASKMOVQrm VR128:$mask, addr:$ptr)>;
+
+def: Pat<(v2i64 (masked_load addr:$ptr, (v2i64 VR128:$mask),
+                             (bc_v2i64 (v4i32 immAllZerosV)))),
+         (VPMASKMOVQrm VR128:$mask, addr:$ptr)>;
+
+def: Pat<(v2i64 (masked_load addr:$ptr, (v2i64 VR128:$mask), (v2i64 VR128:$src0))),
+         (VBLENDVPDrr VR128:$src0, (VPMASKMOVQrm VR128:$mask, addr:$ptr),
+                       VR128:$mask)>;
 
 //===----------------------------------------------------------------------===//
 // Variable Bit Shifts