AMDGPU/GlobalISel: Enable s_{and|or}n2_{b32|b64} patterns

[foad]

LGTM.

[arsenm]

87b2af814078d319dd823eb9c0a279a96ae5f349

llvm/lib/Target/AMDGPU/SOPInstructions.td

 class SOP2_64_32_32 <string opName, list<dag> pattern=[]> : SOP2_Pseudo <
   opName, (outs SReg_64:$sdst), (ins SSrc_b32:$src0, SSrc_b32:$src1),
   "$sdst, $src0, $src1", pattern
 >;
 
 class UniformUnaryFrag<SDPatternOperator Op> : PatFrag <
   (ops node:$src0),
   (Op $src0),
-  [{ return !N->isDivergent(); }]
->;
+  [{ return !N->isDivergent(); }]> {
+  // This check is unnecessary as it's captured by the result register
+  // bank constraint.
+  //
+  // FIXME: Should add a way for the emitter to recognize this is a
+  // trivially true predicate to eliminate the check.
+  let GISelPredicateCode = [{return true;}];
+}
 
 class UniformBinFrag<SDPatternOperator Op> : PatFrag <
   (ops node:$src0, node:$src1),
   (Op $src0, $src1),
   [{ return !N->isDivergent(); }]> {
   // This check is unnecessary as it's captured by the result register
   // bank constraint.
   //

llvm/test/CodeGen/AMDGPU/GlobalISel/andn2.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti < %s | FileCheck -check-prefixes=GCN,GFX6 %s
 ; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX9 %s
 
 define amdgpu_ps i32 @s_andn2_i32(i32 inreg %src0, i32 inreg %src1) {
 ; GCN-LABEL: s_andn2_i32:
 ; GCN:       ; %bb.0:
-; GCN-NEXT:    s_not_b32 s0, s3
-; GCN-NEXT:    s_and_b32 s0, s2, s0
+; GCN-NEXT:    s_andn2_b32 s0, s2, s3
 ; GCN-NEXT:    ; return to shader part epilog
   %not.src1 = xor i32 %src1, -1
   %and = and i32 %src0, %not.src1
   ret i32 %and
 }
 
 define amdgpu_ps i32 @s_andn2_i32_commute(i32 inreg %src0, i32 inreg %src1) {
 ; GCN-LABEL: s_andn2_i32_commute:
 ; GCN:       ; %bb.0:
-; GCN-NEXT:    s_not_b32 s0, s3
-; GCN-NEXT:    s_and_b32 s0, s0, s2
+; GCN-NEXT:    s_andn2_b32 s0, s2, s3
 ; GCN-NEXT:    ; return to shader part epilog
   %not.src1 = xor i32 %src1, -1
   %and = and i32 %not.src1, %src0
   ret i32 %and
 }
 
 define amdgpu_ps { i32, i32 } @s_andn2_i32_multi_use(i32 inreg %src0, i32 inreg %src1) {
 ; GCN-LABEL: s_andn2_i32_multi_use:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    s_not_b32 s1, s3
-; GCN-NEXT:    s_and_b32 s0, s2, s1
+; GCN-NEXT:    s_andn2_b32 s0, s2, s3
 ; GCN-NEXT:    ; return to shader part epilog
   %not.src1 = xor i32 %src1, -1
   %and = and i32 %src0, %not.src1
   %insert.0 = insertvalue { i32, i32 } undef, i32 %and, 0
   %insert.1 = insertvalue { i32, i32 } %insert.0, i32 %not.src1, 1
   ret { i32, i32 } %insert.1
 }
 
 define amdgpu_ps { i32, i32 } @s_andn2_i32_multi_foldable_use(i32 inreg %src0, i32 inreg %src1, i32 inreg %src2) {
 ; GCN-LABEL: s_andn2_i32_multi_foldable_use:
 ; GCN:       ; %bb.0:
-; GCN-NEXT:    s_not_b32 s1, s4
-; GCN-NEXT:    s_and_b32 s0, s2, s1
-; GCN-NEXT:    s_and_b32 s1, s3, s1
+; GCN-NEXT:    s_andn2_b32 s0, s2, s4
+; GCN-NEXT:    s_andn2_b32 s1, s3, s4
 ; GCN-NEXT:    ; return to shader part epilog
   %not.src2 = xor i32 %src2, -1
   %and0 = and i32 %src0, %not.src2
   %and1 = and i32 %src1, %not.src2
   %insert.0 = insertvalue { i32, i32 } undef, i32 %and0, 0
   %insert.1 = insertvalue { i32, i32 } %insert.0, i32 %and1, 1
   ret { i32, i32 } %insert.1
 }
; GCN-NEXT:    ; return to shader part epilog
   %and = and i32 %src0, %not.src1
   %cast = bitcast i32 %and to float
   ret float %cast
 }
 
 define amdgpu_ps i64 @s_andn2_i64(i64 inreg %src0, i64 inreg %src1) {
 ; GCN-LABEL: s_andn2_i64:
 ; GCN:       ; %bb.0:
-; GCN-NEXT:    s_not_b64 s[0:1], s[4:5]
-; GCN-NEXT:    s_and_b64 s[0:1], s[2:3], s[0:1]
+; GCN-NEXT:    s_andn2_b64 s[0:1], s[2:3], s[4:5]
 ; GCN-NEXT:    ; return to shader part epilog
   %not.src1 = xor i64 %src1, -1
   %and = and i64 %src0, %not.src1
   ret i64 %and
 }
 
 define amdgpu_ps i64 @s_andn2_i64_commute(i64 inreg %src0, i64 inreg %src1) {
 ; GCN-LABEL: s_andn2_i64_commute:
 ; GCN:       ; %bb.0:
-; GCN-NEXT:    s_not_b64 s[0:1], s[4:5]
-; GCN-NEXT:    s_and_b64 s[0:1], s[0:1], s[2:3]
+; GCN-NEXT:    s_andn2_b64 s[0:1], s[2:3], s[4:5]
 ; GCN-NEXT:    ; return to shader part epilog
   %not.src1 = xor i64 %src1, -1
   %and = and i64 %not.src1, %src0
   ret i64 %and
 }
 
 define amdgpu_ps { i64, i64 } @s_andn2_i64_multi_foldable_use(i64 inreg %src0, i64 inreg %src1, i64 inreg %src2) {
 ; GCN-LABEL: s_andn2_i64_multi_foldable_use:
 ; GCN:       ; %bb.0:
-; GCN-NEXT:    s_not_b64 s[6:7], s[6:7]
-; GCN-NEXT:    s_and_b64 s[0:1], s[2:3], s[6:7]
-; GCN-NEXT:    s_and_b64 s[2:3], s[4:5], s[6:7]
+; GCN-NEXT:    s_andn2_b64 s[0:1], s[2:3], s[6:7]
+; GCN-NEXT:    s_andn2_b64 s[2:3], s[4:5], s[6:7]
 ; GCN-NEXT:    ; return to shader part epilog
   %not.src2 = xor i64 %src2, -1
   %and0 = and i64 %src0, %not.src2
   %and1 = and i64 %src1, %not.src2
   %insert.0 = insertvalue { i64, i64 } undef, i64 %and0, 0
   %insert.1 = insertvalue { i64, i64 } %insert.0, i64 %and1, 1
   ret { i64, i64 } %insert.1
 }
 
 define amdgpu_ps { i64, i64 } @s_andn2_i64_multi_use(i64 inreg %src0, i64 inreg %src1) {
 ; GCN-LABEL: s_andn2_i64_multi_use:
 ; GCN:       ; %bb.0:
-; GCN-NEXT:    s_not_b64 s[4:5], s[4:5]
-; GCN-NEXT:    s_and_b64 s[0:1], s[2:3], s[4:5]
-; GCN-NEXT:    s_mov_b32 s2, s4
-; GCN-NEXT:    s_mov_b32 s3, s5
+; GCN-NEXT:    s_not_b64 s[6:7], s[4:5]
+; GCN-NEXT:    s_andn2_b64 s[0:1], s[2:3], s[4:5]
+; GCN-NEXT:    s_mov_b32 s2, s6
+; GCN-NEXT:    s_mov_b32 s3, s7
 ; GCN-NEXT:    ; return to shader part epilog
   %not.src1 = xor i64 %src1, -1
   %and = and i64 %src0, %not.src1
   %insert.0 = insertvalue { i64, i64 } undef, i64 %and, 0
   %insert.1 = insertvalue { i64, i64 } %insert.0, i64 %not.src1, 1
   ret { i64, i64 } %insert.1
 }
 

llvm/test/CodeGen/AMDGPU/GlobalISel/orn2.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti < %s | FileCheck -check-prefixes=GCN,GFX6 %s
 ; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX9 %s
 
 define amdgpu_ps i32 @s_orn2_i32(i32 inreg %src0, i32 inreg %src1) {
 ; GCN-LABEL: s_orn2_i32:
 ; GCN:       ; %bb.0:
-; GCN-NEXT:    s_not_b32 s0, s3
-; GCN-NEXT:    s_or_b32 s0, s2, s0
+; GCN-NEXT:    s_orn2_b32 s0, s2, s3
 ; GCN-NEXT:    ; return to shader part epilog
   %not.src1 = xor i32 %src1, -1
   %or = or i32 %src0, %not.src1
   ret i32 %or
 }
 
 define amdgpu_ps i32 @s_orn2_i32_commute(i32 inreg %src0, i32 inreg %src1) {
 ; GCN-LABEL: s_orn2_i32_commute:
 ; GCN:       ; %bb.0:
-; GCN-NEXT:    s_not_b32 s0, s3
-; GCN-NEXT:    s_or_b32 s0, s0, s2
+; GCN-NEXT:    s_orn2_b32 s0, s2, s3
 ; GCN-NEXT:    ; return to shader part epilog
   %not.src1 = xor i32 %src1, -1
   %or = or i32 %not.src1, %src0
   ret i32 %or
 }
 
 define amdgpu_ps { i32, i32 } @s_orn2_i32_multi_use(i32 inreg %src0, i32 inreg %src1) {
 ; GCN-LABEL: s_orn2_i32_multi_use:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    s_not_b32 s1, s3
-; GCN-NEXT:    s_or_b32 s0, s2, s1
+; GCN-NEXT:    s_orn2_b32 s0, s2, s3
 ; GCN-NEXT:    ; return to shader part epilog
   %not.src1 = xor i32 %src1, -1
   %or = or i32 %src0, %not.src1
   %insert.0 = insertvalue { i32, i32 } undef, i32 %or, 0
   %insert.1 = insertvalue { i32, i32 } %insert.0, i32 %not.src1, 1
   ret { i32, i32 } %insert.1
 }
 
 define amdgpu_ps { i32, i32 } @s_orn2_i32_multi_foldable_use(i32 inreg %src0, i32 inreg %src1, i32 inreg %src2) {
 ; GCN-LABEL: s_orn2_i32_multi_foldable_use:
 ; GCN:       ; %bb.0:
-; GCN-NEXT:    s_not_b32 s1, s4
-; GCN-NEXT:    s_or_b32 s0, s2, s1
-; GCN-NEXT:    s_or_b32 s1, s3, s1
+; GCN-NEXT:    s_orn2_b32 s0, s2, s4
+; GCN-NEXT:    s_orn2_b32 s1, s3, s4
 ; GCN-NEXT:    ; return to shader part epilog
   %not.src2 = xor i32 %src2, -1
   %or0 = or i32 %src0, %not.src2
   %or1 = or i32 %src1, %not.src2
   %insert.0 = insertvalue { i32, i32 } undef, i32 %or0, 0
   %insert.1 = insertvalue { i32, i32 } %insert.0, i32 %or1, 1
   ret { i32, i32 } %insert.1
 }
; GCN-NEXT:    ; return to shader part epilog
   %or = or i32 %src0, %not.src1
   %cast = bitcast i32 %or to float
   ret float %cast
 }
 
 define amdgpu_ps i64 @s_orn2_i64(i64 inreg %src0, i64 inreg %src1) {
 ; GCN-LABEL: s_orn2_i64:
 ; GCN:       ; %bb.0:
-; GCN-NEXT:    s_not_b64 s[0:1], s[4:5]
-; GCN-NEXT:    s_or_b64 s[0:1], s[2:3], s[0:1]
+; GCN-NEXT:    s_orn2_b64 s[0:1], s[2:3], s[4:5]
 ; GCN-NEXT:    ; return to shader part epilog
   %not.src1 = xor i64 %src1, -1
   %or = or i64 %src0, %not.src1
   ret i64 %or
 }
 
 define amdgpu_ps i64 @s_orn2_i64_commute(i64 inreg %src0, i64 inreg %src1) {
 ; GCN-LABEL: s_orn2_i64_commute:
 ; GCN:       ; %bb.0:
-; GCN-NEXT:    s_not_b64 s[0:1], s[4:5]
-; GCN-NEXT:    s_or_b64 s[0:1], s[0:1], s[2:3]
+; GCN-NEXT:    s_orn2_b64 s[0:1], s[2:3], s[4:5]
 ; GCN-NEXT:    ; return to shader part epilog
   %not.src1 = xor i64 %src1, -1
   %or = or i64 %not.src1, %src0
   ret i64 %or
 }
 
 define amdgpu_ps { i64, i64 } @s_orn2_i64_multi_foldable_use(i64 inreg %src0, i64 inreg %src1, i64 inreg %src2) {
 ; GCN-LABEL: s_orn2_i64_multi_foldable_use:
 ; GCN:       ; %bb.0:
-; GCN-NEXT:    s_not_b64 s[6:7], s[6:7]
-; GCN-NEXT:    s_or_b64 s[0:1], s[2:3], s[6:7]
-; GCN-NEXT:    s_or_b64 s[2:3], s[4:5], s[6:7]
+; GCN-NEXT:    s_orn2_b64 s[0:1], s[2:3], s[6:7]
+; GCN-NEXT:    s_orn2_b64 s[2:3], s[4:5], s[6:7]
 ; GCN-NEXT:    ; return to shader part epilog
   %not.src2 = xor i64 %src2, -1
   %or0 = or i64 %src0, %not.src2
   %or1 = or i64 %src1, %not.src2
   %insert.0 = insertvalue { i64, i64 } undef, i64 %or0, 0
   %insert.1 = insertvalue { i64, i64 } %insert.0, i64 %or1, 1
   ret { i64, i64 } %insert.1
 }
 
 define amdgpu_ps { i64, i64 } @s_orn2_i64_multi_use(i64 inreg %src0, i64 inreg %src1) {
 ; GCN-LABEL: s_orn2_i64_multi_use:
 ; GCN:       ; %bb.0:
-; GCN-NEXT:    s_not_b64 s[4:5], s[4:5]
-; GCN-NEXT:    s_or_b64 s[0:1], s[2:3], s[4:5]
-; GCN-NEXT:    s_mov_b32 s2, s4
-; GCN-NEXT:    s_mov_b32 s3, s5
+; GCN-NEXT:    s_not_b64 s[6:7], s[4:5]
+; GCN-NEXT:    s_orn2_b64 s[0:1], s[2:3], s[4:5]
+; GCN-NEXT:    s_mov_b32 s2, s6
+; GCN-NEXT:    s_mov_b32 s3, s7
 ; GCN-NEXT:    ; return to shader part epilog
   %not.src1 = xor i64 %src1, -1
   %or = or i64 %src0, %not.src1
   %insert.0 = insertvalue { i64, i64 } undef, i64 %or, 0
   %insert.1 = insertvalue { i64, i64 } %insert.0, i64 %not.src1, 1
   ret { i64, i64 } %insert.1
 }