[AMDGPU] Optimize _L image intrinsic to _LZ when lod is zero

Summary:
Add _L to _LZ image intrinsic table mapping to table gen.
In ISelLowering check if image intrinsic has lod and if it's equal
to zero, if so remove lod and change opcode to equivalent mapped _LZ.

Change-Id: Ie24cd7e788e2195d846c7bd256151178cbb9ec71

Subscribers: arsenm, mehdi_amini, kzhuravl, wdng, nhaehnle, yaxunl, dstuttard, tpr, t-tye, steven_wu, dexonsmith, llvm-commits

Differential Revision: https://reviews.llvm.org/D49483

llvm-svn: 338523

5 files changed, 166 insertions, 2 deletions

diff --git a/llvm/lib/Target/AMDGPU/MIMGInstructions.td b/llvm/lib/Target/AMDGPU/MIMGInstructions.td
index 1e0bc62c45a..44c2d366e46 100644
--- a/llvm/lib/Target/AMDGPU/MIMGInstructions.td
+++ b/llvm/lib/Target/AMDGPU/MIMGInstructions.td
@@ -66,6 +66,22 @@ def MIMGDimInfoTable : GenericTable {
   let PrimaryKeyName = "getMIMGDimInfo";
 }
 
+class MIMGLZMapping<MIMGBaseOpcode l, MIMGBaseOpcode lz> {
+  MIMGBaseOpcode L = l;
+  MIMGBaseOpcode LZ = lz;
+}
+
+def MIMGLZMappingTable : GenericTable {
+  let FilterClass = "MIMGLZMapping";
+  let CppTypeName = "MIMGLZMappingInfo";
+  let Fields = ["L", "LZ"];
+  GenericEnum TypeOf_L = MIMGBaseOpcode;
+  GenericEnum TypeOf_LZ = MIMGBaseOpcode;
+
+  let PrimaryKey = ["L"];
+  let PrimaryKeyName = "getMIMGLZMappingInfo";
+}
+
 class mimg <bits<7> si, bits<7> vi = si> {
   field bits<7> SI = si;
   field bits<7> VI = vi;
@@ -547,3 +563,13 @@ foreach intr = !listconcat(AMDGPUImageDimIntrinsics,
                            AMDGPUImageDimAtomicIntrinsics) in {
   def : ImageDimIntrinsicInfo<intr>;
 }
+
+// L to LZ Optimization Mapping
+def : MIMGLZMapping<IMAGE_SAMPLE_L, IMAGE_SAMPLE_LZ>;
+def : MIMGLZMapping<IMAGE_SAMPLE_C_L, IMAGE_SAMPLE_C_LZ>;
+def : MIMGLZMapping<IMAGE_SAMPLE_L_O, IMAGE_SAMPLE_LZ_O>;
+def : MIMGLZMapping<IMAGE_SAMPLE_C_L_O, IMAGE_SAMPLE_C_LZ_O>;
+def : MIMGLZMapping<IMAGE_GATHER4_L, IMAGE_GATHER4_LZ>;
+def : MIMGLZMapping<IMAGE_GATHER4_C_L, IMAGE_GATHER4_C_LZ>;
+def : MIMGLZMapping<IMAGE_GATHER4_L_O, IMAGE_GATHER4_LZ_O>;
+def : MIMGLZMapping<IMAGE_GATHER4_C_L_O, IMAGE_GATHER4_C_LZ_O>;
diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
index 993c00dae1e..25007861fd1 100644
--- a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -4555,6 +4555,9 @@ SDValue SITargetLowering::lowerImage(SDValue Op,
   const AMDGPU::MIMGBaseOpcodeInfo *BaseOpcode =
       AMDGPU::getMIMGBaseOpcodeInfo(Intr->BaseOpcode);
   const AMDGPU::MIMGDimInfo *DimInfo = AMDGPU::getMIMGDimInfo(Intr->Dim);
+  const AMDGPU::MIMGLZMappingInfo *LZMappingInfo =
+      AMDGPU::getMIMGLZMappingInfo(Intr->BaseOpcode);
+  unsigned IntrOpcode = Intr->BaseOpcode;
 
   SmallVector<EVT, 2> ResultTypes(Op->value_begin(), Op->value_end());
   bool IsD16 = false;
@@ -4640,6 +4643,18 @@ SDValue SITargetLowering::lowerImage(SDValue Op,
   SmallVector<SDValue, 4> VAddrs;
   for (unsigned i = 0; i < NumVAddrs; ++i)
     VAddrs.push_back(Op.getOperand(AddrIdx + i));
+
+  // Optimize _L to _LZ when _L is zero
+  if (LZMappingInfo) {
+    if (auto ConstantLod =
+         dyn_cast<ConstantFPSDNode>(VAddrs[NumVAddrs-1].getNode())) {
+      if (ConstantLod->isZero() || ConstantLod->isNegative()) {
+        IntrOpcode = LZMappingInfo->LZ;  // set new opcode to _lz variant of _l
+        VAddrs.pop_back();               // remove 'lod'
+      }
+    }
+  }
+
   SDValue VAddr = getBuildDwordsVector(DAG, DL, VAddrs);
 
   SDValue True = DAG.getTargetConstant(1, DL, MVT::i1);
@@ -4699,10 +4714,10 @@ SDValue SITargetLowering::lowerImage(SDValue Op,
   int Opcode = -1;
 
   if (Subtarget->getGeneration() >= AMDGPUSubtarget::VOLCANIC_ISLANDS)
-    Opcode = AMDGPU::getMIMGOpcode(Intr->BaseOpcode, AMDGPU::MIMGEncGfx8,
+    Opcode = AMDGPU::getMIMGOpcode(IntrOpcode, AMDGPU::MIMGEncGfx8,
                                    NumVDataDwords, NumVAddrDwords);
   if (Opcode == -1)
-    Opcode = AMDGPU::getMIMGOpcode(Intr->BaseOpcode, AMDGPU::MIMGEncGfx6,
+    Opcode = AMDGPU::getMIMGOpcode(IntrOpcode, AMDGPU::MIMGEncGfx6,
                                    NumVDataDwords, NumVAddrDwords);
   assert(Opcode != -1);
 
diff --git a/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp b/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp
index 3fd3c75874a..4eba1938231 100644
--- a/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp
@@ -110,6 +110,7 @@ struct MIMGInfo {
 #define GET_MIMGBaseOpcodesTable_IMPL
 #define GET_MIMGDimInfoTable_IMPL
 #define GET_MIMGInfoTable_IMPL
+#define GET_MIMGLZMappingTable_IMPL
 #include "AMDGPUGenSearchableTables.inc"
 
 int getMIMGOpcode(unsigned BaseOpcode, unsigned MIMGEncoding,
diff --git a/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h b/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h
index 70681c27169..5b7af8268cd 100644
--- a/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h
+++ b/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h
@@ -42,6 +42,7 @@ namespace AMDGPU {
 #define GET_MIMGBaseOpcode_DECL
 #define GET_MIMGDim_DECL
 #define GET_MIMGEncoding_DECL
+#define GET_MIMGLZMapping_DECL
 #include "AMDGPUGenSearchableTables.inc"
 
 namespace IsaInfo {
@@ -211,6 +212,14 @@ struct MIMGDimInfo {
 LLVM_READONLY
 const MIMGDimInfo *getMIMGDimInfo(unsigned Dim);
 
+struct MIMGLZMappingInfo {
+  MIMGBaseOpcode L;
+  MIMGBaseOpcode LZ;
+};
+
+LLVM_READONLY
+const MIMGLZMappingInfo *getMIMGLZMappingInfo(unsigned L);
+
 LLVM_READONLY
 int getMIMGOpcode(unsigned BaseOpcode, unsigned MIMGEncoding,
                   unsigned VDataDwords, unsigned VAddrDwords);
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.ltolz.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.ltolz.ll
new file mode 100644
index 00000000000..2d66a0be069
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.ltolz.ll
@@ -0,0 +1,113 @@
+; RUN: llc -march=amdgcn -mcpu=verde -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN %s
+; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN %s
+
+
+; GCN-LABEL: {{^}}sample_l_1d:
+; GCN: image_sample_lz v[0:3], v0, s[0:7], s[8:11] dmask:0xf{{$}}
+define amdgpu_ps <4 x float> @sample_l_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s, float %lod) {
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.l.1d.v4f32.f32(i32 15, float %s, float 0.0, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+; GCN-LABEL: {{^}}sample_l_2d:
+; GCN: image_sample_lz v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf{{$}}
+define amdgpu_ps <4 x float> @sample_l_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s, float %t, float %lod) {
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.l.2d.v4f32.f32(i32 15, float %s, float %t, float -0.0, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+; GCN-LABEL: {{^}}sample_c_l_1d:
+; GCN: image_sample_c_lz v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf{{$}}
+define amdgpu_ps <4 x float> @sample_c_l_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, float %s, float %lod) {
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.l.1d.v4f32.f32(i32 15, float %zcompare, float %s, float -2.0, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+; GCN-LABEL: {{^}}sample_c_l_2d:
+; GCN: image_sample_c_lz v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf{{$}}
+define amdgpu_ps <4 x float> @sample_c_l_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, float %s, float %t, float %lod) {
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.l.2d.v4f32.f32(i32 15, float %zcompare, float %s, float %t, float 0.0, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+; GCN-LABEL: {{^}}sample_l_o_1d:
+; GCN: image_sample_lz_o v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf{{$}}
+define amdgpu_ps <4 x float> @sample_l_o_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, i32 %offset, float %s, float %lod) {
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.l.o.1d.v4f32.f32(i32 15, i32 %offset, float %s, float 0.0, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+; GCN-LABEL: {{^}}sample_l_o_2d:
+; GCN: image_sample_lz_o v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf{{$}}
+define amdgpu_ps <4 x float> @sample_l_o_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, i32 %offset, float %s, float %t, float %lod) {
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.l.o.2d.v4f32.f32(i32 15, i32 %offset, float %s, float %t, float 0.0, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+; GCN-LABEL: {{^}}sample_c_l_o_1d:
+; GCN: image_sample_c_lz_o v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf{{$}}
+define amdgpu_ps <4 x float> @sample_c_l_o_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, i32 %offset, float %zcompare, float %s, float %lod) {
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.l.o.1d.v4f32.f32(i32 15, i32 %offset, float %zcompare, float %s, float 0.0, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+; GCN-LABEL: {{^}}sample_c_l_o_2d:
+; GCN: image_sample_c_lz_o v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf{{$}}
+define amdgpu_ps <4 x float> @sample_c_l_o_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, i32 %offset, float %zcompare, float %s, float %t, float %lod) {
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.l.o.2d.v4f32.f32(i32 15, i32 %offset, float %zcompare, float %s, float %t, float 0.0, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+; GCN-LABEL: {{^}}gather4_l_2d:
+; GCN: image_gather4_lz v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf{{$}}
+define amdgpu_ps <4 x float> @gather4_l_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s, float %t, float %lod) {
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.gather4.l.2d.v4f32.f32(i32 15, float %s, float %t, float 0.0, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+; GCN-LABEL: {{^}}gather4_c_l_2d:
+; GCN: image_gather4_c_lz v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf{{$}}
+define amdgpu_ps <4 x float> @gather4_c_l_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, float %s, float %t, float %lod) {
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.gather4.c.l.2d.v4f32.f32(i32 15, float %zcompare, float %s, float %t, float 0.0, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+; GCN-LABEL: {{^}}gather4_l_o_2d:
+; GCN: image_gather4_lz_o v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf{{$}}
+define amdgpu_ps <4 x float> @gather4_l_o_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, i32 %offset, float %s, float %t, float %lod) {
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.gather4.l.o.2d.v4f32.f32(i32 15, i32 %offset, float %s, float %t, float 0.0, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+; GCN-LABEL: {{^}}gather4_c_l_o_2d:
+; GCN: image_gather4_c_lz_o v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf{{$}}
+define amdgpu_ps <4 x float> @gather4_c_l_o_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, i32 %offset, float %zcompare, float %s, float %t, float %lod) {
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.gather4.c.l.o.2d.v4f32.f32(i32 15, i32 %offset, float %zcompare, float %s, float %t, float 0.0, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+declare <4 x float> @llvm.amdgcn.image.sample.l.1d.v4f32.f32(i32, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.l.2d.v4f32.f32(i32, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.l.1d.v4f32.f32(i32, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.l.2d.v4f32.f32(i32, float, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.l.o.1d.v4f32.f32(i32, i32, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.l.o.2d.v4f32.f32(i32, i32, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.l.o.1d.v4f32.f32(i32, i32, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.l.o.2d.v4f32.f32(i32, i32, float, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+
+declare <4 x float> @llvm.amdgcn.image.gather4.l.2d.v4f32.f32(i32, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.gather4.c.l.2d.v4f32.f32(i32, float, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.gather4.l.o.2d.v4f32.f32(i32, i32, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.gather4.c.l.o.2d.v4f32.f32(i32, i32, float, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1