AMDGPU/R600: Convert buffer_id to VTX_READ input

[jvesely]

Summary

Use Patterns to match buffer ids to vtx_reads
Add buffer id to asm string

[arsenm]

What's the reason for moving to separate patterns?

lib/Target/AMDGPU/EvergreenInstructions.td

144–145

Capitalize, indent comment and remove commented out line. The comment should probably go on the encoding bit definition

[jvesely]

In D22650#491979, @arsenm wrote:

What's the reason for moving to separate patterns?

'twas the easiest way to add it to the asm string.

[jvesely]

remove dead code and comments
change buffer_id to lower case to be consistent with other input field names

[jvesely]

ping

[jvesely]

ping

[• tstellarAMD]

LGTM.

[jvesely]

committed as r278749

lib/Target/AMDGPU/CaymanInstructions.td

   let eop = 0; // This bit is not used on Cayman.
 }
 
-class VTX_READ_cm <string name, bits<8> buffer_id, dag outs, list<dag> pattern>
-    : VTX_WORD0_cm, VTX_READ<name, buffer_id, outs, pattern> {
+class VTX_READ_cm <string name, dag outs>
+    : VTX_WORD0_cm, VTX_READ<name, outs, []> {
 
   // Static fields
   let VC_INST = 0;
   let FETCH_TYPE = 2;
   let FETCH_WHOLE_QUAD = 0;
-  let BUFFER_ID = buffer_id;
+// this is set by tablegen magic
+//  let BUFFER_ID = $BUFFER_ID;
   let SRC_REL = 0;
   // XXX: We can infer this field based on the SRC_GPR.  This would allow us
   // to store vertex addresses in any channel, not just X.

   let Inst{31-0} = Word0;
 }
 
-class VTX_READ_8_cm <bits<8> buffer_id, list<dag> pattern>
-    : VTX_READ_cm <"VTX_READ_8 $dst_gpr, $src_gpr", buffer_id,
-                   (outs R600_TReg32_X:$dst_gpr), pattern> {
+def VTX_READ_8_cm
+    : VTX_READ_cm <"VTX_READ_8 $dst_gpr, $src_gpr",
+                   (outs R600_TReg32_X:$dst_gpr)> {
 
   let DST_SEL_X = 0;
   let DST_SEL_Y = 7;   // Masked

   let DATA_FORMAT = 1; // FMT_8
 }
 
-class VTX_READ_16_cm <bits<8> buffer_id, list<dag> pattern>
-    : VTX_READ_cm <"VTX_READ_16 $dst_gpr, $src_gpr", buffer_id,
-                   (outs R600_TReg32_X:$dst_gpr), pattern> {
+def VTX_READ_16_cm
+    : VTX_READ_cm <"VTX_READ_16 $dst_gpr, $src_gpr",
+                   (outs R600_TReg32_X:$dst_gpr)> {
   let DST_SEL_X = 0;
   let DST_SEL_Y = 7;   // Masked
   let DST_SEL_Z = 7;   // Masked

 
 }
 
-class VTX_READ_32_cm <bits<8> buffer_id, list<dag> pattern>
-    : VTX_READ_cm <"VTX_READ_32 $dst_gpr, $src_gpr", buffer_id,
-                   (outs R600_TReg32_X:$dst_gpr), pattern> {
+def VTX_READ_32_cm
+    : VTX_READ_cm <"VTX_READ_32 $dst_gpr, $src_gpr",
+                   (outs R600_TReg32_X:$dst_gpr)> {
 
   let DST_SEL_X        = 0;
   let DST_SEL_Y        = 7;   // Masked

   let Constraints = "$src_gpr.ptr = $dst_gpr";
 }
 
-class VTX_READ_64_cm <bits<8> buffer_id, list<dag> pattern>
-    : VTX_READ_cm <"VTX_READ_64 $dst_gpr, $src_gpr", buffer_id,
-                   (outs R600_Reg64:$dst_gpr), pattern> {
+def VTX_READ_64_cm
+    : VTX_READ_cm <"VTX_READ_64 $dst_gpr.XY, $src_gpr",
+                   (outs R600_Reg64:$dst_gpr)> {
 
   let DST_SEL_X        = 0;
   let DST_SEL_Y        = 1;

   let DATA_FORMAT      = 0x1D; // COLOR_32_32
 }
 
-class VTX_READ_128_cm <bits<8> buffer_id, list<dag> pattern>
-    : VTX_READ_cm <"VTX_READ_128 $dst_gpr.XYZW, $src_gpr", buffer_id,
-                   (outs R600_Reg128:$dst_gpr), pattern> {
+def VTX_READ_128_cm
+    : VTX_READ_cm <"VTX_READ_128 $dst_gpr.XYZW, $src_gpr",
+                   (outs R600_Reg128:$dst_gpr)> {
 
   let DST_SEL_X        =  0;
   let DST_SEL_Y        =  1;

 //===----------------------------------------------------------------------===//
 // VTX Read from parameter memory space
 //===----------------------------------------------------------------------===//
-def VTX_READ_PARAM_8_cm : VTX_READ_8_cm <0,
-  [(set i32:$dst_gpr, (load_param_exti8 ADDRVTX_READ:$src_gpr))]
->;
-
-def VTX_READ_PARAM_16_cm : VTX_READ_16_cm <0,
-  [(set i32:$dst_gpr, (load_param_exti16 ADDRVTX_READ:$src_gpr))]
->;
-
-def VTX_READ_PARAM_32_cm : VTX_READ_32_cm <0,
-  [(set i32:$dst_gpr, (load_param ADDRVTX_READ:$src_gpr))]
->;
-
-def VTX_READ_PARAM_64_cm : VTX_READ_64_cm <0,
-  [(set v2i32:$dst_gpr, (load_param ADDRVTX_READ:$src_gpr))]
->;
+def : Pat<(i32:$dst_gpr (vtx_id3_az_extloadi8 ADDRVTX_READ:$src_gpr)),
+          (VTX_READ_8_cm MEMxi:$src_gpr, 3)>;
+def : Pat<(i32:$dst_gpr (vtx_id3_az_extloadi16 ADDRVTX_READ:$src_gpr)),
+          (VTX_READ_16_cm MEMxi:$src_gpr, 3)>;
+def : Pat<(i32:$dst_gpr (vtx_id3_load ADDRVTX_READ:$src_gpr)),
+          (VTX_READ_32_cm MEMxi:$src_gpr, 3)>;
+def : Pat<(v2i32:$dst_gpr (vtx_id3_load ADDRVTX_READ:$src_gpr)),
+          (VTX_READ_64_cm MEMxi:$src_gpr, 3)>;
+def : Pat<(v4i32:$dst_gpr (vtx_id3_load ADDRVTX_READ:$src_gpr)),
+          (VTX_READ_128_cm MEMxi:$src_gpr, 3)>;
 
-def VTX_READ_PARAM_128_cm : VTX_READ_128_cm <0,
-  [(set v4i32:$dst_gpr, (load_param ADDRVTX_READ:$src_gpr))]
->;
+//===----------------------------------------------------------------------===//
+// VTX Read from constant memory space
+//===----------------------------------------------------------------------===//
+def : Pat<(i32:$dst_gpr (vtx_id2_az_extloadi8 ADDRVTX_READ:$src_gpr)),
+          (VTX_READ_8_cm MEMxi:$src_gpr, 2)>;
+def : Pat<(i32:$dst_gpr (vtx_id2_az_extloadi16 ADDRVTX_READ:$src_gpr)),
+          (VTX_READ_16_cm MEMxi:$src_gpr, 2)>;
+def : Pat<(i32:$dst_gpr (vtx_id2_load ADDRVTX_READ:$src_gpr)),
+          (VTX_READ_32_cm MEMxi:$src_gpr, 2)>;
+def : Pat<(v2i32:$dst_gpr (vtx_id2_load ADDRVTX_READ:$src_gpr)),
+          (VTX_READ_64_cm MEMxi:$src_gpr, 2)>;
+def : Pat<(v4i32:$dst_gpr (vtx_id2_load ADDRVTX_READ:$src_gpr)),
+          (VTX_READ_128_cm MEMxi:$src_gpr, 2)>;
 
 //===----------------------------------------------------------------------===//
 // VTX Read from global memory space
 //===----------------------------------------------------------------------===//
-
-// 8-bit reads
-def VTX_READ_ID1_8_cm : VTX_READ_8_cm <1,
-  [(set i32:$dst_gpr, (vtx_id1_az_extloadi8 ADDRVTX_READ:$src_gpr))]
->;
-
-// 16-bit reads
-def VTX_READ_ID1_16_cm : VTX_READ_16_cm <1,
-  [(set i32:$dst_gpr, (vtx_id1_az_extloadi16 ADDRVTX_READ:$src_gpr))]
->;
-
-// 32-bit reads
-def VTX_READ_ID1_32_cm : VTX_READ_32_cm <1,
-  [(set i32:$dst_gpr, (vtx_id1_load ADDRVTX_READ:$src_gpr))]
->;
-
-// 64-bit reads
-def VTX_READ_ID1_64_cm : VTX_READ_64_cm <1,
-  [(set v2i32:$dst_gpr, (vtx_id1_load ADDRVTX_READ:$src_gpr))]
->;
-
-// 128-bit reads
-def VTX_READ_ID1_128_cm : VTX_READ_128_cm <1,
-  [(set v4i32:$dst_gpr, (vtx_id1_load ADDRVTX_READ:$src_gpr))]
->;
-
-// 8-bit reads
-def VTX_READ_ID2_8_cm : VTX_READ_8_cm <2,
-  [(set i32:$dst_gpr, (vtx_id2_az_extloadi8 ADDRVTX_READ:$src_gpr))]
->;
-
-// 16-bit reads
-def VTX_READ_ID2_16_cm : VTX_READ_16_cm <2,
-  [(set i32:$dst_gpr, (vtx_id2_az_extloadi16 ADDRVTX_READ:$src_gpr))]
->;
-
-// 32-bit reads
-def VTX_READ_ID2_32_cm : VTX_READ_32_cm <2,
-  [(set i32:$dst_gpr, (vtx_id2_load ADDRVTX_READ:$src_gpr))]
->;
-
-// 64-bit reads
-def VTX_READ_ID2_64_cm : VTX_READ_64_cm <2,
-  [(set v2i32:$dst_gpr, (vtx_id2_load ADDRVTX_READ:$src_gpr))]
->;
-
-// 128-bit reads
-def VTX_READ_ID2_128_cm : VTX_READ_128_cm <2,
-  [(set v4i32:$dst_gpr, (vtx_id2_load ADDRVTX_READ:$src_gpr))]
->;
+def : Pat<(i32:$dst_gpr (vtx_id1_az_extloadi8 ADDRVTX_READ:$src_gpr)),
+          (VTX_READ_8_cm MEMxi:$src_gpr, 1)>;
+def : Pat<(i32:$dst_gpr (vtx_id1_az_extloadi16 ADDRVTX_READ:$src_gpr)),
+          (VTX_READ_16_cm MEMxi:$src_gpr, 1)>;
+def : Pat<(i32:$dst_gpr (vtx_id1_load ADDRVTX_READ:$src_gpr)),
+          (VTX_READ_32_cm MEMxi:$src_gpr, 1)>;
+def : Pat<(v2i32:$dst_gpr (vtx_id1_load ADDRVTX_READ:$src_gpr)),
+          (VTX_READ_64_cm MEMxi:$src_gpr, 1)>;
+def : Pat<(v4i32:$dst_gpr (vtx_id1_load ADDRVTX_READ:$src_gpr)),
+          (VTX_READ_128_cm MEMxi:$src_gpr, 1)>;
 
 } // End isCayman
 

lib/Target/AMDGPU/EvergreenInstructions.td

 
 } // End usesCustomInserter = 1
 
-class VTX_READ_eg <string name, bits<8> buffer_id, dag outs, list<dag> pattern>
-    : VTX_WORD0_eg, VTX_READ<name, buffer_id, outs, pattern> {
+class VTX_READ_eg <string name, dag outs>
+    : VTX_WORD0_eg, VTX_READ<name, outs, []> {
 
   // Static fields
   let VC_INST = 0;
   let FETCH_TYPE = 2;
   let FETCH_WHOLE_QUAD = 0;
-  let BUFFER_ID = buffer_id;
+// this is set by tablegen magic
+//  let BUFFER_ID = $BUFFER_ID;
   let SRC_REL = 0;
   // XXX: We can infer this field based on the SRC_GPR.  This would allow us
   // to store vertex addresses in any channel, not just X.

   let Inst{31-0} = Word0;
 }
 
-class VTX_READ_8_eg <bits<8> buffer_id, list<dag> pattern>
-    : VTX_READ_eg <"VTX_READ_8 $dst_gpr, $src_gpr", buffer_id,
-                   (outs R600_TReg32_X:$dst_gpr), pattern> {
+def VTX_READ_8_eg
+    : VTX_READ_eg <"VTX_READ_8 $dst_gpr, $src_gpr",
+                   (outs R600_TReg32_X:$dst_gpr)> {
 
   let MEGA_FETCH_COUNT = 1;
   let DST_SEL_X = 0;

   let DATA_FORMAT = 1; // FMT_8
 }
 
-class VTX_READ_16_eg <bits<8> buffer_id, list<dag> pattern>
-    : VTX_READ_eg <"VTX_READ_16 $dst_gpr, $src_gpr", buffer_id,
-                   (outs R600_TReg32_X:$dst_gpr), pattern> {
+def VTX_READ_16_eg
+    : VTX_READ_eg <"VTX_READ_16 $dst_gpr, $src_gpr",
+                   (outs R600_TReg32_X:$dst_gpr)> {
   let MEGA_FETCH_COUNT = 2;
   let DST_SEL_X = 0;
   let DST_SEL_Y = 7;   // Masked

 
 }
 
-class VTX_READ_32_eg <bits<8> buffer_id, list<dag> pattern>
-    : VTX_READ_eg <"VTX_READ_32 $dst_gpr, $src_gpr", buffer_id,
-                   (outs R600_TReg32_X:$dst_gpr), pattern> {
+def VTX_READ_32_eg
+    : VTX_READ_eg <"VTX_READ_32 $dst_gpr, $src_gpr",
+                   (outs R600_TReg32_X:$dst_gpr)> {
 
   let MEGA_FETCH_COUNT = 4;
   let DST_SEL_X        = 0;

   let Constraints = "$src_gpr.ptr = $dst_gpr";
 }
 
-class VTX_READ_64_eg <bits<8> buffer_id, list<dag> pattern>
-    : VTX_READ_eg <"VTX_READ_64 $dst_gpr.XY, $src_gpr", buffer_id,
-                   (outs R600_Reg64:$dst_gpr), pattern> {
+def VTX_READ_64_eg
+    : VTX_READ_eg <"VTX_READ_64 $dst_gpr.XY, $src_gpr",
+                   (outs R600_Reg64:$dst_gpr)> {
 
   let MEGA_FETCH_COUNT = 8;
   let DST_SEL_X        = 0;

   let DATA_FORMAT      = 0x1D; // COLOR_32_32
 }
 
-class VTX_READ_128_eg <bits<8> buffer_id, list<dag> pattern>
-    : VTX_READ_eg <"VTX_READ_128 $dst_gpr.XYZW, $src_gpr", buffer_id,
-                   (outs R600_Reg128:$dst_gpr), pattern> {
+def VTX_READ_128_eg
+    : VTX_READ_eg <"VTX_READ_128 $dst_gpr.XYZW, $src_gpr",
+                   (outs R600_Reg128:$dst_gpr)> {
 
   let MEGA_FETCH_COUNT = 16;
   let DST_SEL_X        =  0;

 //===----------------------------------------------------------------------===//
 // VTX Read from parameter memory space
 //===----------------------------------------------------------------------===//
+def : Pat<(i32:$dst_gpr (vtx_id3_az_extloadi8 ADDRVTX_READ:$src_gpr)),
+          (VTX_READ_8_eg MEMxi:$src_gpr, 3)>;
+def : Pat<(i32:$dst_gpr (vtx_id3_az_extloadi16 ADDRVTX_READ:$src_gpr)),
+          (VTX_READ_16_eg MEMxi:$src_gpr, 3)>;
+def : Pat<(i32:$dst_gpr (vtx_id3_load ADDRVTX_READ:$src_gpr)),
+          (VTX_READ_32_eg MEMxi:$src_gpr, 3)>;
+def : Pat<(v2i32:$dst_gpr (vtx_id3_load ADDRVTX_READ:$src_gpr)),
+          (VTX_READ_64_eg MEMxi:$src_gpr, 3)>;
+def : Pat<(v4i32:$dst_gpr (vtx_id3_load ADDRVTX_READ:$src_gpr)),
+          (VTX_READ_128_eg MEMxi:$src_gpr, 3)>;
 
-def VTX_READ_PARAM_8_eg : VTX_READ_8_eg <3,
-  [(set i32:$dst_gpr, (load_param_exti8 ADDRVTX_READ:$src_gpr))]
->;
-
-def VTX_READ_PARAM_16_eg : VTX_READ_16_eg <3,
-  [(set i32:$dst_gpr, (load_param_exti16 ADDRVTX_READ:$src_gpr))]
->;
-
-def VTX_READ_PARAM_32_eg : VTX_READ_32_eg <3,
-  [(set i32:$dst_gpr, (load_param ADDRVTX_READ:$src_gpr))]
->;
-
-def VTX_READ_PARAM_64_eg : VTX_READ_64_eg <3,
-  [(set v2i32:$dst_gpr, (load_param ADDRVTX_READ:$src_gpr))]
->;
-
-def VTX_READ_PARAM_128_eg : VTX_READ_128_eg <3,
-  [(set v4i32:$dst_gpr, (load_param ADDRVTX_READ:$src_gpr))]
->;
+//===----------------------------------------------------------------------===//
+// VTX Read from constant memory space
+//===----------------------------------------------------------------------===//
+def : Pat<(i32:$dst_gpr (vtx_id2_az_extloadi8 ADDRVTX_READ:$src_gpr)),
+          (VTX_READ_8_eg MEMxi:$src_gpr, 2)>;
+def : Pat<(i32:$dst_gpr (vtx_id2_az_extloadi16 ADDRVTX_READ:$src_gpr)),
+          (VTX_READ_16_eg MEMxi:$src_gpr, 2)>;
+def : Pat<(i32:$dst_gpr (vtx_id2_load ADDRVTX_READ:$src_gpr)),
+          (VTX_READ_32_eg MEMxi:$src_gpr, 2)>;
+def : Pat<(v2i32:$dst_gpr (vtx_id2_load ADDRVTX_READ:$src_gpr)),
+          (VTX_READ_64_eg MEMxi:$src_gpr, 2)>;
+def : Pat<(v4i32:$dst_gpr (vtx_id2_load ADDRVTX_READ:$src_gpr)),
+          (VTX_READ_128_eg MEMxi:$src_gpr, 2)>;
 
 //===----------------------------------------------------------------------===//
 // VTX Read from global memory space
 //===----------------------------------------------------------------------===//
-
-// 8-bit reads
-def VTX_READ_ID1_8_eg : VTX_READ_8_eg <1,
-  [(set i32:$dst_gpr, (vtx_id1_az_extloadi8 ADDRVTX_READ:$src_gpr))]
->;
-
-// 16-bit reads
-def VTX_READ_ID1_16_eg : VTX_READ_16_eg <1,
-  [(set i32:$dst_gpr, (vtx_id1_az_extloadi16 ADDRVTX_READ:$src_gpr))]
->;
-
-// 32-bit reads
-def VTX_READ_ID1_32_eg : VTX_READ_32_eg <1,
-  [(set i32:$dst_gpr, (vtx_id1_load ADDRVTX_READ:$src_gpr))]
->;
-
-// 64-bit reads
-def VTX_READ_ID1_64_eg : VTX_READ_64_eg <1,
-  [(set v2i32:$dst_gpr, (vtx_id1_load ADDRVTX_READ:$src_gpr))]
->;
-
-// 128-bit reads
-def VTX_READ_ID1_128_eg : VTX_READ_128_eg <1,
-  [(set v4i32:$dst_gpr, (vtx_id1_load ADDRVTX_READ:$src_gpr))]
->;
-
-// 8-bit reads
-def VTX_READ_ID2_8_eg : VTX_READ_8_eg <2,
-  [(set i32:$dst_gpr, (vtx_id2_az_extloadi8 ADDRVTX_READ:$src_gpr))]
->;
-
-// 16-bit reads
-def VTX_READ_ID2_16_eg : VTX_READ_16_eg <2,
-  [(set i32:$dst_gpr, (vtx_id2_az_extloadi16 ADDRVTX_READ:$src_gpr))]
->;
-
-// 32-bit reads
-def VTX_READ_ID2_32_eg : VTX_READ_32_eg <2,
-  [(set i32:$dst_gpr, (vtx_id2_load ADDRVTX_READ:$src_gpr))]
->;
-
-// 64-bit reads
-def VTX_READ_ID2_64_eg : VTX_READ_64_eg <2,
-  [(set v2i32:$dst_gpr, (vtx_id2_load ADDRVTX_READ:$src_gpr))]
->;
-
-// 128-bit reads
-def VTX_READ_ID2_128_eg : VTX_READ_128_eg <2,
-  [(set v4i32:$dst_gpr, (vtx_id2_load ADDRVTX_READ:$src_gpr))]
->;
+def : Pat<(i32:$dst_gpr (vtx_id1_az_extloadi8 ADDRVTX_READ:$src_gpr)),
+          (VTX_READ_8_eg MEMxi:$src_gpr, 1)>;
+def : Pat<(i32:$dst_gpr (vtx_id1_az_extloadi16 ADDRVTX_READ:$src_gpr)),
+          (VTX_READ_16_eg MEMxi:$src_gpr, 1)>;
+def : Pat<(i32:$dst_gpr (vtx_id1_load ADDRVTX_READ:$src_gpr)),
+          (VTX_READ_32_eg MEMxi:$src_gpr, 1)>;
+def : Pat<(v2i32:$dst_gpr (vtx_id1_load ADDRVTX_READ:$src_gpr)),
+          (VTX_READ_64_eg MEMxi:$src_gpr, 1)>;
+def : Pat<(v4i32:$dst_gpr (vtx_id1_load ADDRVTX_READ:$src_gpr)),
+          (VTX_READ_128_eg MEMxi:$src_gpr, 1)>;
 
 } // End Predicates = [isEG]
 

lib/Target/AMDGPU/R600Instructions.td

 
 }
 
-class VTX_READ <string name, bits<8> buffer_id, dag outs, list<dag> pattern>
-    : InstR600ISA <outs, (ins MEMxi:$src_gpr), !strconcat("  ", name), pattern>,
+class VTX_READ <string name, dag outs, list<dag> pattern>
+    : InstR600ISA <outs, (ins MEMxi:$src_gpr, i8imm:$BUFFER_ID), !strconcat("  ", name, ", #$BUFFER_ID"), pattern>,
       VTX_WORD1_GPR {
 
   // Static fields

             (cast<LoadSDNode>(N)->getAddressSpace() == AMDGPUAS::PARAM_I_ADDRESS); }]
 >;
 
-def load_param : LoadParamFrag<load>;
-def load_param_exti8 : LoadParamFrag<az_extloadi8>;
-def load_param_exti16 : LoadParamFrag<az_extloadi16>;
+def vtx_id3_az_extloadi8 : LoadParamFrag<az_extloadi8>;
+def vtx_id3_az_extloadi16 : LoadParamFrag<az_extloadi16>;
+def vtx_id3_load : LoadParamFrag<load>;
 
 class LoadVtxId1 <PatFrag load> : PatFrag <
   (ops node:$ptr), (load node:$ptr), [{

test/CodeGen/AMDGPU/vertex-fetch-encoding.ll

-; RUN: llc < %s -march=r600 -show-mc-encoding -mcpu=barts | FileCheck --check-prefix=NI %s
-; RUN: llc < %s -march=r600 -show-mc-encoding -mcpu=cayman | FileCheck --check-prefix=CM %s
+; RUN: llc < %s -march=r600 -show-mc-encoding -mcpu=cypress | FileCheck --check-prefix=EG --check-prefix=FUNC %s
+; RUN: llc < %s -march=r600 -show-mc-encoding -mcpu=barts | FileCheck --check-prefix=EG --check-prefix=FUNC %s
+; RUN: llc < %s -march=r600 -show-mc-encoding -mcpu=cayman | FileCheck --check-prefix=CM --check-prefix=FUNC %s
 
-; NI: {{^}}vtx_fetch32:
-; NI: VTX_READ_32 T[[GPR:[0-9]]].X, T[[GPR]].X, 0 ; encoding: [0x40,0x01,0x0[[GPR]],0x10,0x0[[GPR]],0xf0,0x5f,0x13,0x00,0x00,0x08,0x00
-; CM: {{^}}vtx_fetch32:
-; CM: VTX_READ_32 T[[GPR:[0-9]]].X, T[[GPR]].X, 0 ; encoding: [0x40,0x01,0x0[[GPR]],0x00,0x0[[GPR]],0xf0,0x5f,0x13,0x00,0x00,0x00,0x00
+; FUNC-LABEL: {{^}}vtx_fetch32:
+; EG: VTX_READ_32 T[[GPR:[0-9]]].X, T[[GPR]].X, 0, #1 ; encoding: [0x40,0x01,0x0[[GPR]],0x10,0x0[[GPR]],0xf0,0x5f,0x13,0x00,0x00,0x08,0x00
+; CM: VTX_READ_32 T[[GPR:[0-9]]].X, T[[GPR]].X, 0, #1 ; encoding: [0x40,0x01,0x0[[GPR]],0x00,0x0[[GPR]],0xf0,0x5f,0x13,0x00,0x00,0x00,0x00
 
 define void @vtx_fetch32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) {
-entry:
-  %0 = load i32, i32 addrspace(1)* %in
-  store i32 %0, i32 addrspace(1)* %out
+  %v = load i32, i32 addrspace(1)* %in
+  store i32 %v, i32 addrspace(1)* %out
   ret void
 }
 
-; NI: {{^}}vtx_fetch128:
-; NI: VTX_READ_128 T[[DST:[0-9]]].XYZW, T[[SRC:[0-9]]].X, 0 ; encoding: [0x40,0x01,0x0[[SRC]],0x40,0x0[[DST]],0x10,0x8d,0x18,0x00,0x00,0x08,0x00
-; XXX: Add a case for Cayman when v4i32 stores are supported.
+; FUNC-LABEL: {{^}}vtx_fetch128:
+; EG: VTX_READ_128 T[[DST:[0-9]]].XYZW, T[[SRC:[0-9]]].X, 0, #1 ; encoding: [0x40,0x01,0x0[[SRC]],0x40,0x0[[DST]],0x10,0x8d,0x18,0x00,0x00,0x08,0x00
+; CM: VTX_READ_128 T[[DST:[0-9]]].XYZW, T[[SRC:[0-9]]].X, 0, #1 ; encoding: [0x40,0x01,0x0[[SRC]],0x00,0x0[[DST]],0x10,0x8d,0x18,0x00,0x00,0x00,0x00
 
 define void @vtx_fetch128(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in) {
-entry:
-  %0 = load <4 x i32>, <4 x i32> addrspace(1)* %in
-  store <4 x i32> %0, <4 x i32> addrspace(1)* %out
+  %v = load <4 x i32>, <4 x i32> addrspace(1)* %in
+  store <4 x i32> %v, <4 x i32> addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}vtx_fetch32_id3:
+; EG: VTX_READ_32 T[[GPR:[0-9]]].X, T[[GPR]].X, 0, #3 ; encoding: [0x40,0x03,0x0[[GPR]],0x10,0x0[[GPR]],0xf0,0x5f,0x13,0x00,0x00,0x08,0x00
+; CM: VTX_READ_32 T[[GPR:[0-9]]].X, T[[GPR]].X, 0, #3 ; encoding: [0x40,0x03,0x0[[GPR]],0x00,0x0[[GPR]],0xf0,0x5f,0x13,0x00,0x00,0x00,0x00
+
+define void @vtx_fetch32_id3(i32 addrspace(1)* %out, i32 addrspace(7)* %in) {
+  %v = load i32, i32 addrspace(7)* %in
+  store i32 %v, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}vtx_fetch32_id2:
+; EG: VTX_READ_32 T[[GPR:[0-9]]].X, T[[GPR]].X, 0, #2 ; encoding: [0x40,0x02,0x0[[GPR]],0x10,0x0[[GPR]],0xf0,0x5f,0x13,0x00,0x00,0x08,0x00
+; CM: VTX_READ_32 T[[GPR:[0-9]]].X, T[[GPR]].X, 0, #2 ; encoding: [0x40,0x02,0x0[[GPR]],0x00,0x0[[GPR]],0xf0,0x5f,0x13,0x00,0x00,0x00,0x00
+
+@t = internal addrspace(2) constant [4 x i32] [i32 0, i32 1, i32 2, i32 3]
+
+define void @vtx_fetch32_id2(i32 addrspace(1)* %out, i32 %in) {
+  %a = getelementptr inbounds [4 x i32], [4 x i32] addrspace(2)* @t, i32 0, i32 %in
+  %v = load i32, i32 addrspace(2)* %a
+  store i32 %v, i32 addrspace(1)* %out
   ret void
 }