AMDGPU: Overload return type of llvm.amdgcn.buffer.load.format

Summary:
Allow the selection of BUFFER_LOAD_FORMAT_x and _XY. Do this now before
the frontend patches land in Mesa. Eventually, we may want to automatically
reduce the size of loads at the LLVM IR level, which requires such overloads,
and in some cases Mesa can generate them directly.

Reviewers: tstellarAMD, arsenm

Subscribers: arsenm, llvm-commits

Differential Revision: http://reviews.llvm.org/D18255

llvm-svn: 263792

4 files changed, 81 insertions, 54 deletions

diff --git a/llvm/include/llvm/IR/IntrinsicsAMDGPU.td b/llvm/include/llvm/IR/IntrinsicsAMDGPU.td
index 5215264eb4c..4e67e361ae4 100644
--- a/llvm/include/llvm/IR/IntrinsicsAMDGPU.td
+++ b/llvm/include/llvm/IR/IntrinsicsAMDGPU.td
@@ -206,7 +206,7 @@ def int_amdgcn_image_atomic_cmpswap : Intrinsic <
   []>;
 
 def int_amdgcn_buffer_load_format : Intrinsic <
-  [llvm_v4f32_ty],
+  [llvm_anyfloat_ty],
   [llvm_v4i32_ty,     // rsrc(SGPR)
    llvm_i32_ty,       // vindex(VGPR)
    llvm_i32_ty,       // offset(SGPR/VGPR/imm)
diff --git a/llvm/lib/Target/AMDGPU/SIInstructions.td b/llvm/lib/Target/AMDGPU/SIInstructions.td
index 7b0c6eaad0e..639faeb8302 100644
--- a/llvm/lib/Target/AMDGPU/SIInstructions.td
+++ b/llvm/lib/Target/AMDGPU/SIInstructions.td
@@ -2108,45 +2108,52 @@ def : Pat <
 //===----------------------------------------------------------------------===//
 // buffer_load/store_format patterns
 //===----------------------------------------------------------------------===//
-def : Pat<
-  (int_amdgcn_buffer_load_format v4i32:$rsrc, 0,
-                                 (MUBUFIntrinsicOffset i32:$soffset,
-                                                       i16:$offset),
-                                 imm:$glc, imm:$slc),
-  (BUFFER_LOAD_FORMAT_XYZW_OFFSET $rsrc, $soffset, (as_i16imm $offset),
-    (as_i1imm $glc), (as_i1imm $slc), 0)
->;
 
-def : Pat<
-  (int_amdgcn_buffer_load_format v4i32:$rsrc, i32:$vindex,
-                                 (MUBUFIntrinsicOffset i32:$soffset,
-                                                       i16:$offset),
-                                 imm:$glc, imm:$slc),
-  (BUFFER_LOAD_FORMAT_XYZW_IDXEN $vindex, $rsrc, $soffset, (as_i16imm $offset),
-    (as_i1imm $glc), (as_i1imm $slc), 0)
->;
+multiclass MUBUF_LoadIntrinsicPat<ValueType vt, string opcode> {
+  def : Pat<
+    (vt (int_amdgcn_buffer_load_format v4i32:$rsrc, 0,
+                                       (MUBUFIntrinsicOffset i32:$soffset,
+                                                             i16:$offset),
+                                       imm:$glc, imm:$slc)),
+    (!cast<MUBUF>(opcode # _OFFSET) $rsrc, $soffset, (as_i16imm $offset),
+      (as_i1imm $glc), (as_i1imm $slc), 0)
+  >;
 
-def : Pat<
-  (int_amdgcn_buffer_load_format v4i32:$rsrc, 0,
-                                 (MUBUFIntrinsicVOffset i32:$soffset,
-                                                        i16:$offset,
-                                                        i32:$voffset),
-                                 imm:$glc, imm:$slc),
-  (BUFFER_LOAD_FORMAT_XYZW_OFFEN $voffset, $rsrc, $soffset, (as_i16imm $offset),
-    (as_i1imm $glc), (as_i1imm $slc), 0)
->;
+  def : Pat<
+    (vt (int_amdgcn_buffer_load_format v4i32:$rsrc, i32:$vindex,
+                                       (MUBUFIntrinsicOffset i32:$soffset,
+                                                             i16:$offset),
+                                       imm:$glc, imm:$slc)),
+    (!cast<MUBUF>(opcode # _IDXEN) $vindex, $rsrc, $soffset, (as_i16imm $offset),
+      (as_i1imm $glc), (as_i1imm $slc), 0)
+  >;
 
-def : Pat<
-  (int_amdgcn_buffer_load_format v4i32:$rsrc, i32:$vindex,
-                                 (MUBUFIntrinsicVOffset i32:$soffset,
-                                                        i16:$offset,
-                                                        i32:$voffset),
-                                 imm:$glc, imm:$slc),
-  (BUFFER_LOAD_FORMAT_XYZW_BOTHEN
-    (REG_SEQUENCE VReg_64, $vindex, sub0, $voffset, sub1),
-    $rsrc, $soffset, (as_i16imm $offset),
-    (as_i1imm $glc), (as_i1imm $slc), 0)
->;
+  def : Pat<
+    (vt (int_amdgcn_buffer_load_format v4i32:$rsrc, 0,
+                                       (MUBUFIntrinsicVOffset i32:$soffset,
+                                                              i16:$offset,
+                                                              i32:$voffset),
+                                       imm:$glc, imm:$slc)),
+    (!cast<MUBUF>(opcode # _OFFEN) $voffset, $rsrc, $soffset, (as_i16imm $offset),
+      (as_i1imm $glc), (as_i1imm $slc), 0)
+  >;
+
+  def : Pat<
+    (vt (int_amdgcn_buffer_load_format v4i32:$rsrc, i32:$vindex,
+                                       (MUBUFIntrinsicVOffset i32:$soffset,
+                                                              i16:$offset,
+                                                              i32:$voffset),
+                                       imm:$glc, imm:$slc)),
+    (!cast<MUBUF>(opcode # _BOTHEN)
+      (REG_SEQUENCE VReg_64, $vindex, sub0, $voffset, sub1),
+      $rsrc, $soffset, (as_i16imm $offset),
+      (as_i1imm $glc), (as_i1imm $slc), 0)
+  >;
+}
+
+defm : MUBUF_LoadIntrinsicPat<f32, "BUFFER_LOAD_FORMAT_X">;
+defm : MUBUF_LoadIntrinsicPat<v2f32, "BUFFER_LOAD_FORMAT_XY">;
+defm : MUBUF_LoadIntrinsicPat<v4f32, "BUFFER_LOAD_FORMAT_XYZW">;
 
 def : Pat<
   (int_amdgcn_buffer_store_format v4f32:$vdata, v4i32:$rsrc, 0,
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.buffer.load.format.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.buffer.load.format.ll
index c6222f426b3..ae2a84b945e 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.buffer.load.format.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.buffer.load.format.ll
@@ -8,9 +8,9 @@
 ;CHECK: s_waitcnt
 define {<4 x float>, <4 x float>, <4 x float>} @buffer_load(<4 x i32> inreg) #0 {
 main_body:
-  %data = call <4 x float> @llvm.amdgcn.buffer.load.format(<4 x i32> %0, i32 0, i32 0, i1 0, i1 0)
-  %data_glc = call <4 x float> @llvm.amdgcn.buffer.load.format(<4 x i32> %0, i32 0, i32 0, i1 1, i1 0)
-  %data_slc = call <4 x float> @llvm.amdgcn.buffer.load.format(<4 x i32> %0, i32 0, i32 0, i1 0, i1 1)
+  %data = call <4 x float> @llvm.amdgcn.buffer.load.format.v4f32(<4 x i32> %0, i32 0, i32 0, i1 0, i1 0)
+  %data_glc = call <4 x float> @llvm.amdgcn.buffer.load.format.v4f32(<4 x i32> %0, i32 0, i32 0, i1 1, i1 0)
+  %data_slc = call <4 x float> @llvm.amdgcn.buffer.load.format.v4f32(<4 x i32> %0, i32 0, i32 0, i1 0, i1 1)
   %r0 = insertvalue {<4 x float>, <4 x float>, <4 x float>} undef, <4 x float> %data, 0
   %r1 = insertvalue {<4 x float>, <4 x float>, <4 x float>} %r0, <4 x float> %data_glc, 1
   %r2 = insertvalue {<4 x float>, <4 x float>, <4 x float>} %r1, <4 x float> %data_slc, 2
@@ -22,7 +22,7 @@ main_body:
 ;CHECK: s_waitcnt
 define <4 x float> @buffer_load_immoffs(<4 x i32> inreg) #0 {
 main_body:
-  %data = call <4 x float> @llvm.amdgcn.buffer.load.format(<4 x i32> %0, i32 0, i32 42, i1 0, i1 0)
+  %data = call <4 x float> @llvm.amdgcn.buffer.load.format.v4f32(<4 x i32> %0, i32 0, i32 42, i1 0, i1 0)
   ret <4 x float> %data
 }
 
@@ -35,9 +35,9 @@ main_body:
 ;CHECK: s_waitcnt
 define <4 x float> @buffer_load_immoffs_large(<4 x i32> inreg) #0 {
 main_body:
-  %d.0 = call <4 x float> @llvm.amdgcn.buffer.load.format(<4 x i32> %0, i32 0, i32 4156, i1 0, i1 0)
-  %d.1 = call <4 x float> @llvm.amdgcn.buffer.load.format(<4 x i32> %0, i32 0, i32 36860, i1 0, i1 0)
-  %d.2 = call <4 x float> @llvm.amdgcn.buffer.load.format(<4 x i32> %0, i32 0, i32 36864, i1 0, i1 0)
+  %d.0 = call <4 x float> @llvm.amdgcn.buffer.load.format.v4f32(<4 x i32> %0, i32 0, i32 4156, i1 0, i1 0)
+  %d.1 = call <4 x float> @llvm.amdgcn.buffer.load.format.v4f32(<4 x i32> %0, i32 0, i32 36860, i1 0, i1 0)
+  %d.2 = call <4 x float> @llvm.amdgcn.buffer.load.format.v4f32(<4 x i32> %0, i32 0, i32 36864, i1 0, i1 0)
   %d.3 = fadd <4 x float> %d.0, %d.1
   %data = fadd <4 x float> %d.2, %d.3
   ret <4 x float> %data
@@ -51,8 +51,8 @@ main_body:
 ;CHECK: s_waitcnt
 define <4 x float> @buffer_load_immoffs_reuse(<4 x i32> inreg) #0 {
 main_body:
-  %d.0 = call <4 x float> @llvm.amdgcn.buffer.load.format(<4 x i32> %0, i32 0, i32 4160, i1 0, i1 0)
-  %d.1 = call <4 x float> @llvm.amdgcn.buffer.load.format(<4 x i32> %0, i32 0, i32 4176, i1 0, i1 0)
+  %d.0 = call <4 x float> @llvm.amdgcn.buffer.load.format.v4f32(<4 x i32> %0, i32 0, i32 4160, i1 0, i1 0)
+  %d.1 = call <4 x float> @llvm.amdgcn.buffer.load.format.v4f32(<4 x i32> %0, i32 0, i32 4176, i1 0, i1 0)
   %data = fadd <4 x float> %d.0, %d.1
   ret <4 x float> %data
 }
@@ -62,7 +62,7 @@ main_body:
 ;CHECK: s_waitcnt
 define <4 x float> @buffer_load_idx(<4 x i32> inreg, i32) #0 {
 main_body:
-  %data = call <4 x float> @llvm.amdgcn.buffer.load.format(<4 x i32> %0, i32 %1, i32 0, i1 0, i1 0)
+  %data = call <4 x float> @llvm.amdgcn.buffer.load.format.v4f32(<4 x i32> %0, i32 %1, i32 0, i1 0, i1 0)
   ret <4 x float> %data
 }
 
@@ -71,7 +71,7 @@ main_body:
 ;CHECK: s_waitcnt
 define <4 x float> @buffer_load_ofs(<4 x i32> inreg, i32) #0 {
 main_body:
-  %data = call <4 x float> @llvm.amdgcn.buffer.load.format(<4 x i32> %0, i32 0, i32 %1, i1 0, i1 0)
+  %data = call <4 x float> @llvm.amdgcn.buffer.load.format.v4f32(<4 x i32> %0, i32 0, i32 %1, i1 0, i1 0)
   ret <4 x float> %data
 }
 
@@ -81,7 +81,7 @@ main_body:
 define <4 x float> @buffer_load_ofs_imm(<4 x i32> inreg, i32) #0 {
 main_body:
   %ofs = add i32 %1, 58
-  %data = call <4 x float> @llvm.amdgcn.buffer.load.format(<4 x i32> %0, i32 0, i32 %ofs, i1 0, i1 0)
+  %data = call <4 x float> @llvm.amdgcn.buffer.load.format.v4f32(<4 x i32> %0, i32 0, i32 %ofs, i1 0, i1 0)
   ret <4 x float> %data
 }
 
@@ -90,7 +90,7 @@ main_body:
 ;CHECK: s_waitcnt
 define <4 x float> @buffer_load_both(<4 x i32> inreg, i32, i32) #0 {
 main_body:
-  %data = call <4 x float> @llvm.amdgcn.buffer.load.format(<4 x i32> %0, i32 %1, i32 %2, i1 0, i1 0)
+  %data = call <4 x float> @llvm.amdgcn.buffer.load.format.v4f32(<4 x i32> %0, i32 %1, i32 %2, i1 0, i1 0)
   ret <4 x float> %data
 }
 
@@ -100,11 +100,31 @@ main_body:
 ;CHECK: s_waitcnt
 define <4 x float> @buffer_load_both_reversed(<4 x i32> inreg, i32, i32) #0 {
 main_body:
-  %data = call <4 x float> @llvm.amdgcn.buffer.load.format(<4 x i32> %0, i32 %2, i32 %1, i1 0, i1 0)
+  %data = call <4 x float> @llvm.amdgcn.buffer.load.format.v4f32(<4 x i32> %0, i32 %2, i32 %1, i1 0, i1 0)
   ret <4 x float> %data
 }
 
-declare <4 x float> @llvm.amdgcn.buffer.load.format(<4 x i32>, i32, i32, i1, i1) #1
+;CHECK-LABEL: {{^}}buffer_load_x:
+;CHECK: buffer_load_format_x v0, s[0:3], 0
+;CHECK: s_waitcnt
+define float @buffer_load_x(<4 x i32> inreg %rsrc) #0 {
+main_body:
+  %data = call float @llvm.amdgcn.buffer.load.format.f32(<4 x i32> %rsrc, i32 0, i32 0, i1 0, i1 0)
+  ret float %data
+}
+
+;CHECK-LABEL: {{^}}buffer_load_xy:
+;CHECK: buffer_load_format_xy v[0:1], s[0:3], 0
+;CHECK: s_waitcnt
+define <2 x float> @buffer_load_xy(<4 x i32> inreg %rsrc) #0 {
+main_body:
+  %data = call <2 x float> @llvm.amdgcn.buffer.load.format.v2f32(<4 x i32> %rsrc, i32 0, i32 0, i1 0, i1 0)
+  ret <2 x float> %data
+}
+
+declare float @llvm.amdgcn.buffer.load.format.f32(<4 x i32>, i32, i32, i1, i1) #1
+declare <2 x float> @llvm.amdgcn.buffer.load.format.v2f32(<4 x i32>, i32, i32, i1, i1) #1
+declare <4 x float> @llvm.amdgcn.buffer.load.format.v4f32(<4 x i32>, i32, i32, i1, i1) #1
 
 attributes #0 = { "ShaderType"="0" }
 attributes #1 = { nounwind readonly }
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.buffer.store.format.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.buffer.store.format.ll
index 7e254efdcca..3867aad6dc9 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.buffer.store.format.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.buffer.store.format.ll
@@ -65,13 +65,13 @@ main_body:
 define void @buffer_store_wait(<4 x i32> inreg, <4 x float>, i32, i32, i32) #0 {
 main_body:
   call void @llvm.amdgcn.buffer.store.format.v4f32(<4 x float> %1, <4 x i32> %0, i32 %2, i32 0, i1 0, i1 0)
-  %data = call <4 x float> @llvm.amdgcn.buffer.load.format(<4 x i32> %0, i32 %3, i32 0, i1 0, i1 0)
+  %data = call <4 x float> @llvm.amdgcn.buffer.load.format.v4f32(<4 x i32> %0, i32 %3, i32 0, i1 0, i1 0)
   call void @llvm.amdgcn.buffer.store.format.v4f32(<4 x float> %data, <4 x i32> %0, i32 %4, i32 0, i1 0, i1 0)
   ret void
 }
 
 declare void @llvm.amdgcn.buffer.store.format.v4f32(<4 x float>, <4 x i32>, i32, i32, i1, i1) #1
-declare <4 x float> @llvm.amdgcn.buffer.load.format(<4 x i32>, i32, i32, i1, i1) #2
+declare <4 x float> @llvm.amdgcn.buffer.load.format.v4f32(<4 x i32>, i32, i32, i1, i1) #2
 
 attributes #0 = { "ShaderType"="0" }
 attributes #1 = { nounwind }