R600: Add SI load support for v[24]i32 and store for v2i32

[oota-llvm.git] / lib / Target / R600 / SIInstructions.td

diff --git a/lib/Target/R600/SIInstructions.td b/lib/Target/R600/SIInstructions.td
index 03eced0af5ffaf3845c868f61c5c353b5e9fcaf2..9c96c080c0811f759b6eebc6cf36e71b6c84b2e6 100644 (file)
--- a/lib/Target/R600/SIInstructions.td
+++ b/lib/Target/R600/SIInstructions.td
@@ -22,8 +22,8 @@ def InterpSlot : Operand<i32> {
   let PrintMethod = "printInterpSlot";
 }
 
-def isSI : Predicate<"Subtarget.device()"
-                            "->getGeneration() == AMDGPUDeviceInfo::HD7XXX">;
+def isSI : Predicate<"Subtarget.getGeneration() "
+                      "== AMDGPUSubtarget::SOUTHERN_ISLANDS">;
 
 let Predicates = [isSI] in {
 
@@ -399,7 +399,7 @@ defm BUFFER_LOAD_FORMAT_XYZW : MUBUF_Load_Helper <0x00000003, "BUFFER_LOAD_FORMA
 //def BUFFER_STORE_FORMAT_XY : MUBUF_ <0x00000005, "BUFFER_STORE_FORMAT_XY", []>;
 //def BUFFER_STORE_FORMAT_XYZ : MUBUF_ <0x00000006, "BUFFER_STORE_FORMAT_XYZ", []>;
 //def BUFFER_STORE_FORMAT_XYZW : MUBUF_ <0x00000007, "BUFFER_STORE_FORMAT_XYZW", []>;
-//def BUFFER_LOAD_UBYTE : MUBUF_ <0x00000008, "BUFFER_LOAD_UBYTE", []>;
+defm BUFFER_LOAD_UBYTE : MUBUF_Load_Helper <0x00000008, "BUFFER_LOAD_UBYTE", VReg_32>;
 //def BUFFER_LOAD_SBYTE : MUBUF_ <0x00000009, "BUFFER_LOAD_SBYTE", []>;
 //def BUFFER_LOAD_USHORT : MUBUF_ <0x0000000a, "BUFFER_LOAD_USHORT", []>;
 //def BUFFER_LOAD_SSHORT : MUBUF_ <0x0000000b, "BUFFER_LOAD_SSHORT", []>;
@@ -416,7 +416,10 @@ def BUFFER_STORE_DWORD : MUBUF_Store_Helper <
 def BUFFER_STORE_DWORDX2 : MUBUF_Store_Helper <
   0x0000001d, "BUFFER_STORE_DWORDX2", VReg_64, i64
 >;
-//def BUFFER_STORE_DWORDX4 : MUBUF_DWORDX4 <0x0000001e, "BUFFER_STORE_DWORDX4", []>;
+
+def BUFFER_STORE_DWORDX4 : MUBUF_Store_Helper <
+  0x0000001e, "BUFFER_STORE_DWORDX4", VReg_128, v4i32
+>;
 //def BUFFER_ATOMIC_SWAP : MUBUF_ <0x00000030, "BUFFER_ATOMIC_SWAP", []>;
 //def BUFFER_ATOMIC_CMPSWAP : MUBUF_ <0x00000031, "BUFFER_ATOMIC_CMPSWAP", []>;
 //def BUFFER_ATOMIC_ADD : MUBUF_ <0x00000032, "BUFFER_ATOMIC_ADD", []>;
@@ -1200,6 +1203,19 @@ def SI_INDIRECT_DST_V16 : SI_INDIRECT_DST<VReg_512>;
 
 } // Uses = [EXEC,VCC,M0], Defs = [EXEC,VCC,M0]
 
+// This psuedo instruction takes a pointer as input and outputs a resource
+// constant that can be used with the ADDR64 MUBUF instructions.
+
+let usesCustomInserter = 1 in {
+
+def SI_ADDR64_RSRC : InstSI <
+  (outs SReg_128:$srsrc),
+  (ins SReg_64:$ptr),
+  "", []
+>;
+
+} // end usesCustomInserter
+
 } // end IsCodeGenOnly, isPseudo
 
 def : Pat<
@@ -1591,6 +1607,60 @@ defm : SMRD_Pattern <S_LOAD_DWORDX2_IMM, S_LOAD_DWORDX2_SGPR, i64>;
 defm : SMRD_Pattern <S_LOAD_DWORDX4_IMM, S_LOAD_DWORDX4_SGPR, v16i8>;
 defm : SMRD_Pattern <S_LOAD_DWORDX8_IMM, S_LOAD_DWORDX8_SGPR, v32i8>;
 
+//===----------------------------------------------------------------------===//
+// MUBUF Patterns
+//===----------------------------------------------------------------------===//
+
+multiclass MUBUFLoad_Pattern <MUBUF Instr_ADDR64, ValueType vt,
+                              PatFrag global_ld, PatFrag constant_ld> {
+  def : Pat <
+    (vt (global_ld (add i64:$ptr, (i64 IMM12bit:$offset)))),
+    (Instr_ADDR64 (SI_ADDR64_RSRC (i64 0)), $ptr, (as_i16imm $offset))
+  >;
+
+  def : Pat <
+    (vt (global_ld i64:$ptr)),
+    (Instr_ADDR64 (SI_ADDR64_RSRC (i64 0)), $ptr, 0)
+  >;
+
+  def : Pat <
+     (vt (global_ld (add i64:$ptr, i64:$offset))),
+     (Instr_ADDR64 (SI_ADDR64_RSRC $ptr), $offset, 0)
+  >;
+
+  def : Pat <
+     (vt (constant_ld (add i64:$ptr, i64:$offset))),
+     (Instr_ADDR64 (SI_ADDR64_RSRC $ptr), $offset, 0)
+  >;
+}
+
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_DWORD_ADDR64, i32,
+                          global_load, constant_load>;
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_UBYTE_ADDR64, i32,
+                          zextloadi8_global, zextloadi8_constant>;
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_DWORDX2_ADDR64, v2i32,
+                          global_load, constant_load>;
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_DWORDX4_ADDR64, v4i32,
+                          global_load, constant_load>;
+
+multiclass MUBUFStore_Pattern <MUBUF Instr, ValueType vt> {
+
+  def : Pat <
+    (global_store vt:$value, i64:$ptr),
+    (Instr $value, (SI_ADDR64_RSRC (i64 0)), $ptr, 0)
+  >;
+
+  def : Pat <
+    (global_store vt:$value, (add i64:$ptr, i64:$offset)),
+    (Instr $value, (SI_ADDR64_RSRC $ptr), $offset, 0)
+   >;
+}
+
+defm : MUBUFStore_Pattern <BUFFER_STORE_DWORD, i32>;
+defm : MUBUFStore_Pattern <BUFFER_STORE_DWORDX2, i64>;
+defm : MUBUFStore_Pattern <BUFFER_STORE_DWORDX2, v2i32>;
+defm : MUBUFStore_Pattern <BUFFER_STORE_DWORDX4, v4i32>;
+
 /********** ====================== **********/
 /**********   Indirect adressing   **********/
 /********** ====================== **********/