From 4a706472692db58ca724b64e1bff17a67b369eb6 Mon Sep 17 00:00:00 2001
From: Tom Stellard <thomas.stellard@amd.com>
Date: Wed, 2 Jul 2014 20:53:56 +0000
Subject: [PATCH] R600/SI: Use a ComplexPattern for ADDR64 addressing of MUBUF
 loads

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@212217 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/R600/SIInstrInfo.td    |  8 +++-
 lib/Target/R600/SIInstructions.td | 64 ++++++++++++++-----------------
 2 files changed, 35 insertions(+), 37 deletions(-)

diff --git a/lib/Target/R600/SIInstrInfo.td b/lib/Target/R600/SIInstrInfo.td
index 76330cf443e..774c9d13b17 100644
--- a/lib/Target/R600/SIInstrInfo.td
+++ b/lib/Target/R600/SIInstrInfo.td
@@ -576,7 +576,9 @@ class MTBUF_Store_Helper <bits<3> op, string asm, RegisterClass regClass> : MTBU
   let mayLoad = 0;
 }
 
-multiclass MUBUF_Load_Helper <bits<7> op, string asm, RegisterClass regClass> {
+multiclass MUBUF_Load_Helper <bits<7> op, string asm, RegisterClass regClass,
+                              ValueType load_vt = i32,
+                              SDPatternOperator ld = null_frag> {
 
   let lds = 0, mayLoad = 1 in {
 
@@ -618,7 +620,9 @@ multiclass MUBUF_Load_Helper <bits<7> op, string asm, RegisterClass regClass> {
     let offen = 0, idxen = 0, addr64 = 1, glc = 0, slc = 0, tfe = 0, soffset = 128 /* ZERO */ in {
       def _ADDR64 : MUBUF <op, (outs regClass:$vdata),
                            (ins SReg_128:$srsrc, VReg_64:$vaddr, u16imm:$offset),
-                           asm#" $vdata, $srsrc + $vaddr + $offset", []>;
+                           asm#" $vdata, $srsrc + $vaddr + $offset",
+                           [(set load_vt:$vdata, (ld (MUBUFAddr64 v4i32:$srsrc,
+                                                  i64:$vaddr, u16imm:$offset)))]>;
     }
   }
 }
diff --git a/lib/Target/R600/SIInstructions.td b/lib/Target/R600/SIInstructions.td
index 7e71bbd9eb6..291d63455c9 100644
--- a/lib/Target/R600/SIInstructions.td
+++ b/lib/Target/R600/SIInstructions.td
@@ -846,13 +846,27 @@ 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", []>;
-defm BUFFER_LOAD_UBYTE : MUBUF_Load_Helper <0x00000008, "BUFFER_LOAD_UBYTE", VReg_32>;
-defm BUFFER_LOAD_SBYTE : MUBUF_Load_Helper <0x00000009, "BUFFER_LOAD_SBYTE", VReg_32>;
-defm BUFFER_LOAD_USHORT : MUBUF_Load_Helper <0x0000000a, "BUFFER_LOAD_USHORT", VReg_32>;
-defm BUFFER_LOAD_SSHORT : MUBUF_Load_Helper <0x0000000b, "BUFFER_LOAD_SSHORT", VReg_32>;
-defm BUFFER_LOAD_DWORD : MUBUF_Load_Helper <0x0000000c, "BUFFER_LOAD_DWORD", VReg_32>;
-defm BUFFER_LOAD_DWORDX2 : MUBUF_Load_Helper <0x0000000d, "BUFFER_LOAD_DWORDX2", VReg_64>;
-defm BUFFER_LOAD_DWORDX4 : MUBUF_Load_Helper <0x0000000e, "BUFFER_LOAD_DWORDX4", VReg_128>;
+defm BUFFER_LOAD_UBYTE : MUBUF_Load_Helper <
+  0x00000008, "BUFFER_LOAD_UBYTE", VReg_32, i32, az_extloadi8_global
+>;
+defm BUFFER_LOAD_SBYTE : MUBUF_Load_Helper <
+  0x00000009, "BUFFER_LOAD_SBYTE", VReg_32, i32, sextloadi8_global
+>;
+defm BUFFER_LOAD_USHORT : MUBUF_Load_Helper <
+  0x0000000a, "BUFFER_LOAD_USHORT", VReg_32, i32, az_extloadi16_global
+>;
+defm BUFFER_LOAD_SSHORT : MUBUF_Load_Helper <
+  0x0000000b, "BUFFER_LOAD_SSHORT", VReg_32, i32, sextloadi16_global
+>;
+defm BUFFER_LOAD_DWORD : MUBUF_Load_Helper <
+  0x0000000c, "BUFFER_LOAD_DWORD", VReg_32, i32, global_load
+>;
+defm BUFFER_LOAD_DWORDX2 : MUBUF_Load_Helper <
+  0x0000000d, "BUFFER_LOAD_DWORDX2", VReg_64, v2i32, global_load
+>;
+defm BUFFER_LOAD_DWORDX4 : MUBUF_Load_Helper <
+  0x0000000e, "BUFFER_LOAD_DWORDX4", VReg_128, v4i32, global_load
+>;
 
 def BUFFER_STORE_BYTE : MUBUF_Store_Helper <
   0x00000018, "BUFFER_STORE_BYTE", VReg_32, i32, truncstorei8_global
@@ -2482,27 +2496,7 @@ defm : DSAtomicCmpXChg<DS_CMPST_RTN_B64, i64, atomic_cmp_swap_64_local>;
 //===----------------------------------------------------------------------===//
 
 multiclass MUBUFLoad_Pattern <MUBUF Instr_ADDR64, ValueType vt,
-                              PatFrag global_ld, PatFrag constant_ld> {
-  def : Pat <
-    (vt (global_ld (mubuf_vaddr_offset i64:$ptr, i64:$offset, IMM12bit:$imm_offset))),
-    (Instr_ADDR64 (SI_ADDR64_RSRC $ptr), $offset, (as_i16imm $imm_offset))
-  >;
-
-  def : Pat <
-    (vt (global_ld (add i64:$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), $offset, 0)
-  >;
-
+                              PatFrag constant_ld> {
   def : Pat <
      (vt (constant_ld (add i64:$ptr, i64:$offset))),
      (Instr_ADDR64 (SI_ADDR64_RSRC $ptr), $offset, 0)
@@ -2510,19 +2504,19 @@ multiclass MUBUFLoad_Pattern <MUBUF Instr_ADDR64, ValueType vt,
 }
 
 defm : MUBUFLoad_Pattern <BUFFER_LOAD_SBYTE_ADDR64, i32,
-                          sextloadi8_global, sextloadi8_constant>;
+                          sextloadi8_constant>;
 defm : MUBUFLoad_Pattern <BUFFER_LOAD_UBYTE_ADDR64, i32,
-                          az_extloadi8_global, az_extloadi8_constant>;
+                          az_extloadi8_constant>;
 defm : MUBUFLoad_Pattern <BUFFER_LOAD_SSHORT_ADDR64, i32,
-                          sextloadi16_global, sextloadi16_constant>;
+                          sextloadi16_constant>;
 defm : MUBUFLoad_Pattern <BUFFER_LOAD_USHORT_ADDR64, i32,
-                          az_extloadi16_global, az_extloadi16_constant>;
+                          az_extloadi16_constant>;
 defm : MUBUFLoad_Pattern <BUFFER_LOAD_DWORD_ADDR64, i32,
-                          global_load, constant_load>;
+                          constant_load>;
 defm : MUBUFLoad_Pattern <BUFFER_LOAD_DWORDX2_ADDR64, v2i32,
-                          global_load, constant_load>;
+                          constant_load>;
 defm : MUBUFLoad_Pattern <BUFFER_LOAD_DWORDX4_ADDR64, v4i32,
-                          global_load, constant_load>;
+                          constant_load>;
 
 // BUFFER_LOAD_DWORD*, addr64=0
 multiclass MUBUF_Load_Dword <ValueType vt, MUBUF offset, MUBUF offen, MUBUF idxen,
-- 
2.34.1