R600/SI: Add intrinsic for texture image loading

Patch by: Michel Dänzer

Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@181267 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 46 insertions, 10 deletions

diff --git a/lib/Target/R600/SIInstrInfo.td b/lib/Target/R600/SIInstrInfo.td
index aafc331f25..c8aecb707e 100644
--- a/lib/Target/R600/SIInstrInfo.td
+++ b/lib/Target/R600/SIInstrInfo.td
@@ -335,7 +335,22 @@ class MTBUF_Load_Helper <bits<3> op, string asm, RegisterClass regClass> : MTBUF
   let mayStore = 0;
 }
 
-class MIMG_Load_Helper <bits<7> op, string asm> : MIMG <
+class MIMG_NoSampler_Helper <bits<7> op, string asm> : MIMG <
+  op,
+  (outs VReg_128:$vdata),
+  (ins i32imm:$dmask, i1imm:$unorm, i1imm:$glc, i1imm:$da, i1imm:$r128,
+       i1imm:$tfe, i1imm:$lwe, i1imm:$slc, unknown:$vaddr,
+       SReg_256:$srsrc),
+  asm#" $vdata, $dmask, $unorm, $glc, $da, $r128,"
+     #" $tfe, $lwe, $slc, $vaddr, $srsrc",
+  []> {
+  let SSAMP = 0;
+  let mayLoad = 1;
+  let mayStore = 0;
+  let hasPostISelHook = 1;
+}
+
+class MIMG_Sampler_Helper <bits<7> op, string asm> : MIMG <
   op,
   (outs VReg_128:$vdata),
   (ins i32imm:$dmask, i1imm:$unorm, i1imm:$glc, i1imm:$da, i1imm:$r128,
@@ -382,7 +397,7 @@ def getCommuteOrig : InstrMapping {
 
 // Test if the supplied opcode is an MIMG instruction
 def isMIMG : InstrMapping {
-  let FilterClass = "MIMG_Load_Helper";
+  let FilterClass = "MIMG";
   let RowFields = ["Inst"];
   let ColFields = ["Size"];
   let KeyCol = ["8"];
diff --git a/lib/Target/R600/SIInstructions.td b/lib/Target/R600/SIInstructions.td
index 985d5ed84f..132b6d1643 100644
--- a/lib/Target/R600/SIInstructions.td
+++ b/lib/Target/R600/SIInstructions.td
@@ -495,7 +495,7 @@ defm S_BUFFER_LOAD_DWORDX16 : SMRD_Helper <
 //def S_MEMTIME : SMRD_ <0x0000001e, "S_MEMTIME", []>;
 //def S_DCACHE_INV : SMRD_ <0x0000001f, "S_DCACHE_INV", []>;
 //def IMAGE_LOAD : MIMG_NoPattern_ <"IMAGE_LOAD", 0x00000000>;
-//def IMAGE_LOAD_MIP : MIMG_NoPattern_ <"IMAGE_LOAD_MIP", 0x00000001>;
+def IMAGE_LOAD_MIP : MIMG_NoSampler_Helper <0x00000001, "IMAGE_LOAD_MIP">;
 //def IMAGE_LOAD_PCK : MIMG_NoPattern_ <"IMAGE_LOAD_PCK", 0x00000002>;
 //def IMAGE_LOAD_PCK_SGN : MIMG_NoPattern_ <"IMAGE_LOAD_PCK_SGN", 0x00000003>;
 //def IMAGE_LOAD_MIP_PCK : MIMG_NoPattern_ <"IMAGE_LOAD_MIP_PCK", 0x00000004>;
@@ -522,20 +522,20 @@ defm S_BUFFER_LOAD_DWORDX16 : SMRD_Helper <
 //def IMAGE_ATOMIC_FCMPSWAP : MIMG_NoPattern_ <"IMAGE_ATOMIC_FCMPSWAP", 0x0000001d>;
 //def IMAGE_ATOMIC_FMIN : MIMG_NoPattern_ <"IMAGE_ATOMIC_FMIN", 0x0000001e>;
 //def IMAGE_ATOMIC_FMAX : MIMG_NoPattern_ <"IMAGE_ATOMIC_FMAX", 0x0000001f>;
-def IMAGE_SAMPLE : MIMG_Load_Helper <0x00000020, "IMAGE_SAMPLE">; 
+def IMAGE_SAMPLE : MIMG_Sampler_Helper <0x00000020, "IMAGE_SAMPLE">; 
 //def IMAGE_SAMPLE_CL : MIMG_NoPattern_ <"IMAGE_SAMPLE_CL", 0x00000021>;
-def IMAGE_SAMPLE_D : MIMG_Load_Helper <0x00000022, "IMAGE_SAMPLE_D">;
+def IMAGE_SAMPLE_D : MIMG_Sampler_Helper <0x00000022, "IMAGE_SAMPLE_D">;
 //def IMAGE_SAMPLE_D_CL : MIMG_NoPattern_ <"IMAGE_SAMPLE_D_CL", 0x00000023>;
-def IMAGE_SAMPLE_L : MIMG_Load_Helper <0x00000024, "IMAGE_SAMPLE_L">;
-def IMAGE_SAMPLE_B : MIMG_Load_Helper <0x00000025, "IMAGE_SAMPLE_B">;
+def IMAGE_SAMPLE_L : MIMG_Sampler_Helper <0x00000024, "IMAGE_SAMPLE_L">;
+def IMAGE_SAMPLE_B : MIMG_Sampler_Helper <0x00000025, "IMAGE_SAMPLE_B">;
 //def IMAGE_SAMPLE_B_CL : MIMG_NoPattern_ <"IMAGE_SAMPLE_B_CL", 0x00000026>;
 //def IMAGE_SAMPLE_LZ : MIMG_NoPattern_ <"IMAGE_SAMPLE_LZ", 0x00000027>;
-def IMAGE_SAMPLE_C : MIMG_Load_Helper <0x00000028, "IMAGE_SAMPLE_C">;
+def IMAGE_SAMPLE_C : MIMG_Sampler_Helper <0x00000028, "IMAGE_SAMPLE_C">;
 //def IMAGE_SAMPLE_C_CL : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_CL", 0x00000029>;
 //def IMAGE_SAMPLE_C_D : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_D", 0x0000002a>;
 //def IMAGE_SAMPLE_C_D_CL : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_D_CL", 0x0000002b>;
-def IMAGE_SAMPLE_C_L : MIMG_Load_Helper <0x0000002c, "IMAGE_SAMPLE_C_L">;
-def IMAGE_SAMPLE_C_B : MIMG_Load_Helper <0x0000002d, "IMAGE_SAMPLE_C_B">;
+def IMAGE_SAMPLE_C_L : MIMG_Sampler_Helper <0x0000002c, "IMAGE_SAMPLE_C_L">;
+def IMAGE_SAMPLE_C_B : MIMG_Sampler_Helper <0x0000002d, "IMAGE_SAMPLE_C_B">;
 //def IMAGE_SAMPLE_C_B_CL : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_B_CL", 0x0000002e>;
 //def IMAGE_SAMPLE_C_LZ : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_LZ", 0x0000002f>;
 //def IMAGE_SAMPLE_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_O", 0x00000030>;
@@ -1281,6 +1281,25 @@ defm : SamplePatterns<v4i32>;
 defm : SamplePatterns<v8i32>;
 defm : SamplePatterns<v16i32>;
 
+/* int_SI_imageload for texture fetches consuming varying address parameters */
+class ImageLoadPattern<Intrinsic name, MIMG opcode, ValueType addr_type> : Pat <
+    (name addr_type:$addr, v32i8:$rsrc, imm),
+    (opcode 0xf, 0, 0, 0, 0, 0, 0, 0, $addr, $rsrc)
+>;
+
+class ImageLoadArrayPattern<Intrinsic name, MIMG opcode, ValueType addr_type> : Pat <
+    (name addr_type:$addr, v32i8:$rsrc, TEX_ARRAY),
+    (opcode 0xf, 0, 0, 1, 0, 0, 0, 0, $addr, $rsrc)
+>;
+
+multiclass ImageLoadPatterns<ValueType addr_type> {
+  def : ImageLoadPattern <int_SI_imageload, IMAGE_LOAD_MIP, addr_type>;
+  def : ImageLoadArrayPattern <int_SI_imageload, IMAGE_LOAD_MIP, addr_type>;
+}
+
+defm : ImageLoadPatterns<v2i32>;
+defm : ImageLoadPatterns<v4i32>;
+
 /********** ============================================ **********/
 /********** Extraction, Insertion, Building and Casting  **********/
 /********** ============================================ **********/
diff --git a/lib/Target/R600/SIIntrinsics.td b/lib/Target/R600/SIIntrinsics.td
index 16d9d812af..c2fb8843af 100644
--- a/lib/Target/R600/SIIntrinsics.td
+++ b/lib/Target/R600/SIIntrinsics.td
@@ -25,6 +25,8 @@ let TargetPrefix = "SI", isTarget = 1 in {
   def int_SI_sampleb : Sample;
   def int_SI_samplel : Sample;
 
+  def int_SI_imageload : Intrinsic <[llvm_v4i32_ty], [llvm_anyvector_ty, llvm_v32i8_ty, llvm_i32_ty], [IntrNoMem]>;
+
   /* Interpolation Intrinsics */
 
   def int_SI_fs_constant : Intrinsic <[llvm_float_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;