Add AVX ld/st XCSR register.

Add VEX encoding bits for MRMXm x86 form



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@107204 91177308-0d34-0410-b5e6-96231b3b80d8

4 files changed, 50 insertions, 15 deletions

diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index c8f2d3e06b..e34e45e4d3 100644
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -2019,6 +2019,8 @@ defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
 defm RCP   : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
              sse1_fp_unop_p<0x53, "rcp", X86frcp, int_x86_sse_rcp_ps>;
 
+// There is no f64 version of the reciprocal approximation instructions.
+
 //===----------------------------------------------------------------------===//
 // SSE 1 & 2 - Non-temporal stores
 //===----------------------------------------------------------------------===//
@@ -2111,7 +2113,7 @@ def MOVNTImr_Int  :   I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
                   TB, Requires<[HasSSE2]>;
 
 //===----------------------------------------------------------------------===//
-// SSE 1 & 2 - Misc Instructions
+// SSE 1 & 2 - Misc Instructions (No AVX form)
 //===----------------------------------------------------------------------===//
 
 // Prefetch intrinsic.
@@ -2128,12 +2130,6 @@ def PREFETCHNTA  : PSI<0x18, MRM0m, (outs), (ins i8mem:$src),
 def SFENCE : I<0xAE, MRM_F8, (outs), (ins), "sfence", [(int_x86_sse_sfence)]>,
              TB, Requires<[HasSSE1]>;
 
-// MXCSR register
-def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
-                  "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
-def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
-                  "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
-
 // Alias instructions that map zero vector to pxor / xorp* for sse.
 // We set canFoldAsLoad because this can be converted to a constant-pool
 // load of an all-zeros value if folding it would be beneficial.
@@ -2156,13 +2152,26 @@ def : Pat<(v16i8 immAllZerosV), (V_SET0PI)>;
 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
           (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
 
+//===----------------------------------------------------------------------===//
+// SSE 1 & 2 - Load/Store XCSR register
+//===----------------------------------------------------------------------===//
+
+let isAsmParserOnly = 1 in {
+  def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
+                    "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
+  def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
+                    "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
+}
+
+def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
+                  "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
+def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
+                  "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
+
 //===---------------------------------------------------------------------===//
 // SSE2 Instructions
 //===---------------------------------------------------------------------===//
 
-
-// There is no f64 version of the reciprocal approximation instructions.
-
 //===---------------------------------------------------------------------===//
 // SSE integer instructions
 let ExeDomain = SSEPackedInt in {
diff --git a/lib/Target/X86/X86MCCodeEmitter.cpp b/lib/Target/X86/X86MCCodeEmitter.cpp
index 3b9b20a665..5dd668400b 100644
--- a/lib/Target/X86/X86MCCodeEmitter.cpp
+++ b/lib/Target/X86/X86MCCodeEmitter.cpp
@@ -455,15 +455,17 @@ void X86MCCodeEmitter::EmitVEXOpcodePrefix(uint64_t TSFlags, unsigned &CurByte,
   unsigned NumOps = MI.getNumOperands();
   unsigned CurOp = 0;
 
-  if ((TSFlags & X86II::FormMask) == X86II::MRMDestMem)
-    NumOps = CurOp = X86AddrNumOperands;
-
   switch (TSFlags & X86II::FormMask) {
   case X86II::MRMInitReg: assert(0 && "FIXME: Remove this!");
-  case X86II::MRMSrcMem:
+  case X86II::MRM0m: case X86II::MRM1m:
+  case X86II::MRM2m: case X86II::MRM3m:
+  case X86II::MRM4m: case X86II::MRM5m:
+  case X86II::MRM6m: case X86II::MRM7m:
   case X86II::MRMDestMem:
+    NumOps = CurOp = X86AddrNumOperands;
+  case X86II::MRMSrcMem:
   case X86II::MRMSrcReg:
-    if (MI.getOperand(CurOp).isReg() &&
+    if (MI.getNumOperands() > CurOp && MI.getOperand(CurOp).isReg() &&
         X86InstrInfo::isX86_64ExtendedReg(MI.getOperand(CurOp).getReg()))
       VEX_R = 0x0;
 
diff --git a/test/MC/AsmParser/X86/x86_32-encoding.s b/test/MC/AsmParser/X86/x86_32-encoding.s
index 4ad9d799f3..fa45c4e188 100644
--- a/test/MC/AsmParser/X86/x86_32-encoding.s
+++ b/test/MC/AsmParser/X86/x86_32-encoding.s
@@ -10950,3 +10950,19 @@
 // CHECK: encoding: [0xc5,0xf8,0x2b,0x08]
           vmovntps  %xmm1, (%eax)
 
+// CHECK: vldmxcsr  (%eax)
+// CHECK: encoding: [0xc5,0xf8,0xae,0x10]
+          vldmxcsr  (%eax)
+
+// CHECK: vstmxcsr  (%eax)
+// CHECK: encoding: [0xc5,0xf8,0xae,0x18]
+          vstmxcsr  (%eax)
+
+// CHECK: vldmxcsr  3735928559
+// CHECK: encoding: [0xc5,0xf8,0xae,0x15,0xef,0xbe,0xad,0xde]
+          vldmxcsr  0xdeadbeef
+
+// CHECK: vstmxcsr  3735928559
+// CHECK: encoding: [0xc5,0xf8,0xae,0x1d,0xef,0xbe,0xad,0xde]
+          vstmxcsr  0xdeadbeef
+
diff --git a/test/MC/AsmParser/X86/x86_64-encoding.s b/test/MC/AsmParser/X86/x86_64-encoding.s
index 071f9d538d..b030caf077 100644
--- a/test/MC/AsmParser/X86/x86_64-encoding.s
+++ b/test/MC/AsmParser/X86/x86_64-encoding.s
@@ -998,3 +998,11 @@ pshufb	CPI1_0(%rip), %xmm1
 // CHECK: encoding: [0xc5,0x78,0x2b,0x18]
           vmovntps  %xmm11, (%rax)
 
+// CHECK: vldmxcsr  -4(%rip)
+// CHECK: encoding: [0xc5,0xf8,0xae,0x15,0xfc,0xff,0xff,0xff]
+          vldmxcsr  -4(%rip)
+
+// CHECK: vstmxcsr  -4(%rsp)
+// CHECK: encoding: [0xc5,0xf8,0xae,0x5c,0x24,0xfc]
+          vstmxcsr  -4(%rsp)
+