From 19110099c04a831dd1b47ee8084e48f516712d95 Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault@amd.com>
Date: Sat, 22 Aug 2015 00:54:31 +0000
Subject: [PATCH] AMDGPU: Allow specifying different opcode on VI for SMRD/SMEM

Although the basic s_load_* instructions happen to use the same
opcode, some of the special case SMRD instructions have
different opcodes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@245775 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/AMDGPU/SIInstrInfo.td    | 16 +++++++++++-----
 lib/Target/AMDGPU/SIInstructions.td | 20 ++++++++++----------
 2 files changed, 21 insertions(+), 15 deletions(-)

diff --git a/lib/Target/AMDGPU/SIInstrInfo.td b/lib/Target/AMDGPU/SIInstrInfo.td
index 402ae05664b..5772573e9c7 100644
--- a/lib/Target/AMDGPU/SIInstrInfo.td
+++ b/lib/Target/AMDGPU/SIInstrInfo.td
@@ -72,6 +72,12 @@ class sopk <bits<5> si, bits<5> vi = si> {
   field bits<5> VI = vi;
 }
 
+// Specify an SMRD opcode for SI and SMEM opcode for VI
+class smrd<bits<5> si, bits<5> vi = si> {
+  field bits<5> SI = si;
+  field bits<8> VI = { 0, 0, 0, vi };
+}
+
 // Execpt for the NONE field, this must be kept in sync with the SISubtarget enum
 // in AMDGPUInstrInfo.cpp
 def SISubtarget {
@@ -900,21 +906,21 @@ class SMRD_Real_vi <bits<8> op, string opName, bit imm, dag outs, dag ins,
   let AssemblerPredicates = [isVI];
 }
 
-multiclass SMRD_m <bits<5> op, string opName, bit imm, dag outs, dag ins,
+multiclass SMRD_m <smrd op, string opName, bit imm, dag outs, dag ins,
                    string asm, list<dag> pattern> {
 
   def "" : SMRD_Pseudo <opName, outs, ins, pattern>;
 
-  def _si : SMRD_Real_si <op, opName, imm, outs, ins, asm>;
+  def _si : SMRD_Real_si <op.SI, opName, imm, outs, ins, asm>;
 
   // glc is only applicable to scalar stores, which are not yet
   // implemented.
   let glc = 0 in {
-    def _vi : SMRD_Real_vi <{0, 0, 0, op}, opName, imm, outs, ins, asm>;
+    def _vi : SMRD_Real_vi <op.VI, opName, imm, outs, ins, asm>;
   }
 }
 
-multiclass SMRD_Helper <bits<5> op, string opName, RegisterClass baseClass,
+multiclass SMRD_Helper <smrd op, string opName, RegisterClass baseClass,
                         RegisterClass dstClass> {
   defm _IMM : SMRD_m <
     op, opName#"_IMM", 1, (outs dstClass:$dst),
@@ -924,7 +930,7 @@ multiclass SMRD_Helper <bits<5> op, string opName, RegisterClass baseClass,
 
   def _IMM_ci : SMRD <
     (outs dstClass:$dst), (ins baseClass:$sbase, smrd_literal_offset:$offset),
-    opName#" $dst, $sbase, $offset", []>, SMRD_IMMe_ci <op> {
+    opName#" $dst, $sbase, $offset", []>, SMRD_IMMe_ci <op.SI> {
     let AssemblerPredicates = [isCIOnly];
   }
 
diff --git a/lib/Target/AMDGPU/SIInstructions.td b/lib/Target/AMDGPU/SIInstructions.td
index 833a19f2d43..5c50580822f 100644
--- a/lib/Target/AMDGPU/SIInstructions.td
+++ b/lib/Target/AMDGPU/SIInstructions.td
@@ -62,30 +62,30 @@ let mayLoad = 1 in {
 // We are using the SGPR_32 and not the SReg_32 register class for 32-bit
 // SMRD instructions, because the SGPR_32 register class does not include M0
 // and writing to M0 from an SMRD instruction will hang the GPU.
-defm S_LOAD_DWORD : SMRD_Helper <0x00, "s_load_dword", SReg_64, SGPR_32>;
-defm S_LOAD_DWORDX2 : SMRD_Helper <0x01, "s_load_dwordx2", SReg_64, SReg_64>;
-defm S_LOAD_DWORDX4 : SMRD_Helper <0x02, "s_load_dwordx4", SReg_64, SReg_128>;
-defm S_LOAD_DWORDX8 : SMRD_Helper <0x03, "s_load_dwordx8", SReg_64, SReg_256>;
-defm S_LOAD_DWORDX16 : SMRD_Helper <0x04, "s_load_dwordx16", SReg_64, SReg_512>;
+defm S_LOAD_DWORD : SMRD_Helper <smrd<0x00>, "s_load_dword", SReg_64, SGPR_32>;
+defm S_LOAD_DWORDX2 : SMRD_Helper <smrd<0x01>, "s_load_dwordx2", SReg_64, SReg_64>;
+defm S_LOAD_DWORDX4 : SMRD_Helper <smrd<0x02>, "s_load_dwordx4", SReg_64, SReg_128>;
+defm S_LOAD_DWORDX8 : SMRD_Helper <smrd<0x03>, "s_load_dwordx8", SReg_64, SReg_256>;
+defm S_LOAD_DWORDX16 : SMRD_Helper <smrd<0x04>, "s_load_dwordx16", SReg_64, SReg_512>;
 
 defm S_BUFFER_LOAD_DWORD : SMRD_Helper <
-  0x08, "s_buffer_load_dword", SReg_128, SGPR_32
+  smrd<0x08>, "s_buffer_load_dword", SReg_128, SGPR_32
 >;
 
 defm S_BUFFER_LOAD_DWORDX2 : SMRD_Helper <
-  0x09, "s_buffer_load_dwordx2", SReg_128, SReg_64
+  smrd<0x09>, "s_buffer_load_dwordx2", SReg_128, SReg_64
 >;
 
 defm S_BUFFER_LOAD_DWORDX4 : SMRD_Helper <
-  0x0a, "s_buffer_load_dwordx4", SReg_128, SReg_128
+  smrd<0x0a>, "s_buffer_load_dwordx4", SReg_128, SReg_128
 >;
 
 defm S_BUFFER_LOAD_DWORDX8 : SMRD_Helper <
-  0x0b, "s_buffer_load_dwordx8", SReg_128, SReg_256
+  smrd<0x0b>, "s_buffer_load_dwordx8", SReg_128, SReg_256
 >;
 
 defm S_BUFFER_LOAD_DWORDX16 : SMRD_Helper <
-  0x0c, "s_buffer_load_dwordx16", SReg_128, SReg_512
+  smrd<0x0c>, "s_buffer_load_dwordx16", SReg_128, SReg_512
 >;
 
 } // mayLoad = 1
-- 
2.34.1