R600: Properly set COUNT_3 bit in TEX clause initiating inst for pre EG gen.

Fixes rv7x0 bug in Heaven reported here:
https://bugs.freedesktop.org/show_bug.cgi?id=64257

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@184116 91177308-0d34-0410-b5e6-96231b3b80d8

2 files changed, 60 insertions, 14 deletions

diff --git a/lib/Target/R600/R600Instructions.td b/lib/Target/R600/R600Instructions.td
index 27b0214b00..83d735f28c 100644
--- a/lib/Target/R600/R600Instructions.td
+++ b/lib/Target/R600/R600Instructions.td
@@ -575,14 +575,16 @@ class CF_WORD0_R600 {
 class CF_CLAUSE_R600 <bits<7> inst, dag ins, string AsmPrint> : AMDGPUInst <(outs),
 ins, AsmPrint, [] >, CF_WORD0_R600, CF_WORD1_R600 {
   field bits<64> Inst;
+  bits<4> CNT;
 
   let CF_INST = inst;
   let BARRIER = 1;
   let CF_CONST = 0;
   let VALID_PIXEL_MODE = 0;
   let COND = 0;
+  let COUNT = CNT{2-0};
   let CALL_COUNT = 0;
-  let COUNT_3 = 0;
+  let COUNT_3 = CNT{3};
   let END_OF_PROGRAM = 0;
   let WHOLE_QUAD_MODE = 0;
 
@@ -1162,52 +1164,52 @@ let Predicates = [isR600] in {
   }
   defm : SteamOutputExportPattern<R600_ExportBuf, 0x20, 0x21, 0x22, 0x23>;
 
-  def CF_TC_R600 : CF_CLAUSE_R600<1, (ins i32imm:$ADDR, i32imm:$COUNT),
-  "TEX $COUNT @$ADDR"> {
+  def CF_TC_R600 : CF_CLAUSE_R600<1, (ins i32imm:$ADDR, i32imm:$CNT),
+  "TEX $CNT @$ADDR"> {
     let POP_COUNT = 0;
   }
-  def CF_VC_R600 : CF_CLAUSE_R600<2, (ins i32imm:$ADDR, i32imm:$COUNT),
-  "VTX $COUNT @$ADDR"> {
+  def CF_VC_R600 : CF_CLAUSE_R600<2, (ins i32imm:$ADDR, i32imm:$CNT),
+  "VTX $CNT @$ADDR"> {
     let POP_COUNT = 0;
   }
   def WHILE_LOOP_R600 : CF_CLAUSE_R600<6, (ins i32imm:$ADDR),
   "LOOP_START_DX10 @$ADDR"> {
     let POP_COUNT = 0;
-    let COUNT = 0;
+    let CNT = 0;
   }
   def END_LOOP_R600 : CF_CLAUSE_R600<5, (ins i32imm:$ADDR), "END_LOOP @$ADDR"> {
     let POP_COUNT = 0;
-    let COUNT = 0;
+    let CNT = 0;
   }
   def LOOP_BREAK_R600 : CF_CLAUSE_R600<9, (ins i32imm:$ADDR),
   "LOOP_BREAK @$ADDR"> {
     let POP_COUNT = 0;
-    let COUNT = 0;
+    let CNT = 0;
   }
   def CF_CONTINUE_R600 : CF_CLAUSE_R600<8, (ins i32imm:$ADDR),
   "CONTINUE @$ADDR"> {
     let POP_COUNT = 0;
-    let COUNT = 0;
+    let CNT = 0;
   }
   def CF_JUMP_R600 : CF_CLAUSE_R600<10, (ins i32imm:$ADDR, i32imm:$POP_COUNT),
   "JUMP @$ADDR POP:$POP_COUNT"> {
-    let COUNT = 0;
+    let CNT = 0;
   }
   def CF_ELSE_R600 : CF_CLAUSE_R600<13, (ins i32imm:$ADDR, i32imm:$POP_COUNT),
   "ELSE @$ADDR POP:$POP_COUNT"> {
-    let COUNT = 0;
+    let CNT = 0;
   }
   def CF_CALL_FS_R600 : CF_CLAUSE_R600<19, (ins), "CALL_FS"> {
     let ADDR = 0;
-    let COUNT = 0;
+    let CNT = 0;
     let POP_COUNT = 0;
   }
   def POP_R600 : CF_CLAUSE_R600<14, (ins i32imm:$ADDR, i32imm:$POP_COUNT),
   "POP @$ADDR POP:$POP_COUNT"> {
-    let COUNT = 0;
+    let CNT = 0;
   }
   def CF_END_R600 : CF_CLAUSE_R600<0, (ins), "CF_END"> {
-    let COUNT = 0;
+    let CNT = 0;
     let POP_COUNT = 0;
     let ADDR = 0;
     let END_OF_PROGRAM = 1;
diff --git a/test/CodeGen/R600/rv7x0_count3.ll b/test/CodeGen/R600/rv7x0_count3.ll
new file mode 100644
index 0000000000..474d6ba902
--- /dev/null
+++ b/test/CodeGen/R600/rv7x0_count3.ll
@@ -0,0 +1,44 @@
+; RUN: llc < %s -march=r600 -show-mc-encoding  -mcpu=rv710 | FileCheck %s
+
+; CHECK: TEX 9 @4 ;  encoding: [0x04,0x00,0x00,0x00,0x00,0x04,0x88,0x80]
+
+define void @test(<4 x float> addrspace(1)* %out, <4 x float> addrspace(1)* %in) {
+   %1 = call float @llvm.R600.load.input(i32 4)
+   %2 = call float @llvm.R600.load.input(i32 5)
+   %3 = call float @llvm.R600.load.input(i32 6)
+   %4 = call float @llvm.R600.load.input(i32 7)
+   %5 = insertelement <4 x float> undef, float %1, i32 0
+   %6 = insertelement <4 x float> %5, float %2, i32 1
+   %7 = insertelement <4 x float> %6, float %3, i32 2
+   %8 = insertelement <4 x float> %7, float %4, i32 3
+   %9 = call <4 x float> @llvm.AMDGPU.tex(<4 x float> %8, i32 0, i32 0, i32 1)
+   %10 = call <4 x float> @llvm.AMDGPU.tex(<4 x float> %8, i32 1, i32 0, i32 1)
+   %11 = call <4 x float> @llvm.AMDGPU.tex(<4 x float> %8, i32 2, i32 0, i32 1)
+   %12 = call <4 x float> @llvm.AMDGPU.tex(<4 x float> %8, i32 3, i32 0, i32 1)
+   %13 = call <4 x float> @llvm.AMDGPU.tex(<4 x float> %8, i32 4, i32 0, i32 1)
+   %14 = call <4 x float> @llvm.AMDGPU.tex(<4 x float> %8, i32 5, i32 0, i32 1)
+   %15 = call <4 x float> @llvm.AMDGPU.tex(<4 x float> %8, i32 6, i32 0, i32 1)
+   %16 = call <4 x float> @llvm.AMDGPU.tex(<4 x float> %8, i32 7, i32 0, i32 1)
+   %17 = call <4 x float> @llvm.AMDGPU.tex(<4 x float> %8, i32 8, i32 0, i32 1)
+   %18 = call <4 x float> @llvm.AMDGPU.tex(<4 x float> %8, i32 9, i32 0, i32 1)
+   %19 = fadd <4 x float> %9, %10
+   %20 = fadd <4 x float> %19, %11
+   %21 = fadd <4 x float> %20, %12
+   %22 = fadd <4 x float> %21, %13
+   %23 = fadd <4 x float> %22, %14
+   %24 = fadd <4 x float> %23, %15
+   %25 = fadd <4 x float> %24, %16
+   %26 = fadd <4 x float> %25, %17
+   %27 = fadd <4 x float> %26, %18
+   call void @llvm.R600.store.swizzle(<4 x float> %27, i32 0, i32 2)
+   ret void
+}
+
+declare <4 x float> @llvm.AMDGPU.tex(<4 x float>, i32, i32, i32) readnone
+
+; Function Attrs: readnone
+declare float @llvm.R600.load.input(i32) #1
+
+
+declare void @llvm.R600.store.swizzle(<4 x float>, i32, i32)
+attributes #1 = { readnone }