R600/SI: Add support for global loads

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@183131 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/R600/AMDGPUInstructions.td b/lib/Target/R600/AMDGPUInstructions.td
index 54df7d0b1fd7db1bfe2ead92b02f17c3f329425d..29df37499a2aace497dc726c6b3deb4879dbf78c 100644 (file)
--- a/lib/Target/R600/AMDGPUInstructions.td
+++ b/lib/Target/R600/AMDGPUInstructions.td
@@ -90,6 +90,10 @@ def zextloadi8_global : PatFrag<(ops node:$ptr), (zextloadi8 node:$ptr), [{
     return isGlobalLoad(dyn_cast<LoadSDNode>(N));
 }]>;
 
+def zextloadi8_constant : PatFrag<(ops node:$ptr), (zextloadi8 node:$ptr), [{
+    return isGlobalLoad(dyn_cast<LoadSDNode>(N));
+}]>;
+
 class Constants {
 int TWO_PI = 0x40c90fdb;
 int PI = 0x40490fdb;

diff --git a/lib/Target/R600/SIInstrInfo.td b/lib/Target/R600/SIInstrInfo.td
index 19d9de1ffc89ce2b22d1689b053906a0295353a5..cb159baef13484eb4ca88cb370d4b325bfafde9d 100644 (file)
--- a/lib/Target/R600/SIInstrInfo.td
+++ b/lib/Target/R600/SIInstrInfo.td
@@ -35,9 +35,12 @@ def IMM8bitDWORD : ImmLeaf <
   }]>
 >;
 
-def IMM12bit : ImmLeaf <
-  i16,
-  [{return isUInt<12>(Imm);}]
+def as_i16imm : SDNodeXForm<imm, [{
+  return CurDAG->getTargetConstant(N->getSExtValue(), MVT::i16);
+}]>;
+
+def IMM12bit : PatLeaf <(imm),
+  [{return isUInt<12>(N->getZExtValue());}]
 >;
 
 class InlineImm <ValueType vt> : PatLeaf <(vt imm), [{

diff --git a/lib/Target/R600/SIInstructions.td b/lib/Target/R600/SIInstructions.td
index c739e2ad5729bd2352fbc53fa74ab7c573690664..b6db815d3655a879fa3fdbbd92395671147be3dd 100644 (file)
--- a/lib/Target/R600/SIInstructions.td
+++ b/lib/Target/R600/SIInstructions.td
@@ -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", []>;
@@ -1611,6 +1611,34 @@ 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>;
+
 multiclass MUBUFStore_Pattern <MUBUF Instr, ValueType vt> {
 
   def : Pat <

diff --git a/test/CodeGen/R600/load.ll b/test/CodeGen/R600/load.ll
index b03245ae87b3c5c8f3bd77531b6ba4a6313cf516..ff774ec922dbab8f5414be971b00c9a606ebc121 100644 (file)
--- a/test/CodeGen/R600/load.ll
+++ b/test/CodeGen/R600/load.ll
@@ -1,8 +1,12 @@
-; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s
+; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck --check-prefix=R600-CHECK %s
+; RUN: llc < %s -march=r600 -mcpu=SI | FileCheck --check-prefix=SI-CHECK  %s
 
 ; Load an i8 value from the global address space.
-; CHECK: VTX_READ_8 T{{[0-9]+\.X, T[0-9]+\.X}}
+; R600-CHECK: @load_i8
+; R600-CHECK: VTX_READ_8 T{{[0-9]+\.X, T[0-9]+\.X}}
 
+; SI-CHECK: @load_i8
+; SI-CHECK: BUFFER_LOAD_UBYTE VGPR{{[0-9]+}},
 define void @load_i8(i32 addrspace(1)* %out, i8 addrspace(1)* %in) {
   %1 = load i8 addrspace(1)* %in
   %2 = zext i8 %1 to i32
@@ -10,9 +14,51 @@ define void @load_i8(i32 addrspace(1)* %out, i8 addrspace(1)* %in) {
   ret void
 }
 
+; load an i32 value from the global address space.
+; R600-CHECK: @load_i32
+; R600-CHECK: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0
+
+; SI-CHECK: @load_i32
+; SI-CHECK: BUFFER_LOAD_DWORD VGPR{{[0-9]+}}
+define void @load_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) {
+entry:
+  %0 = load i32 addrspace(1)* %in
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; load a f32 value from the global address space.
+; R600-CHECK: @load_f32
+; R600-CHECK: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0
+
+; SI-CHECK: @load_f32
+; SI-CHECK: BUFFER_LOAD_DWORD VGPR{{[0-9]+}}
+define void @load_f32(float addrspace(1)* %out, float addrspace(1)* %in) {
+entry:
+  %0 = load float addrspace(1)* %in
+  store float %0, float addrspace(1)* %out
+  ret void
+}
+
+; Load an i32 value from the constant address space.
+; R600-CHECK: @load_const_addrspace_i32
+; R600-CHECK: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0
+
+; SI-CHECK: @load_const_addrspace_i32
+; SI-CHECK: S_LOAD_DWORD SGPR{{[0-9]+}}
+define void @load_const_addrspace_i32(i32 addrspace(1)* %out, i32 addrspace(2)* %in) {
+entry:
+  %0 = load i32 addrspace(2)* %in
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
 ; Load a f32 value from the constant address space.
-; CHECK: VTX_READ_32 T{{[0-9]+\.X, T[0-9]+\.X}}
+; R600-CHECK: @load_const_addrspace_f32
+; R600-CHECK: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0
 
+; SI-CHECK: @load_const_addrspace_f32
+; SI-CHECK: S_LOAD_DWORD SGPR{{[0-9]+}}
 define void @load_const_addrspace_f32(float addrspace(1)* %out, float addrspace(2)* %in) {
   %1 = load float addrspace(2)* %in
   store float %1, float addrspace(1)* %out