[avx] Add patterns for combining vextractf128 + vmovaps/vmovups/vmobdqu to

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

diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index 02205bb2c799654dff424ec2e26cce1620a27e2f..df426279ea5440b9470640b254466adc1b603b15 100644 (file)
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -7309,6 +7309,23 @@ def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
           []>, VEX;
 }
 
+// Extract and store.
+let Predicates = [HasAVX] in {
+  def : Pat<(alignedstore (int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2), addr:$dst),
+          (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
+  def : Pat<(alignedstore (int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2), addr:$dst),
+          (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
+  def : Pat<(alignedstore (int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2), addr:$dst),
+          (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
+
+  def : Pat<(int_x86_sse_storeu_ps addr:$dst, (int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2)),
+          (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
+  def : Pat<(int_x86_sse2_storeu_pd addr:$dst, (int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2)),
+          (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
+  def : Pat<(int_x86_sse2_storeu_dq addr:$dst, (bc_v16i8 (int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2))),
+          (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
+}
+
 // AVX1 patterns
 let Predicates = [HasAVX] in {
 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),