R600: Add pattern for SHA-256 Ma function

This can be optimized using the BFI_INT instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@181033 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/R600/AMDGPUInstructions.td b/lib/Target/R600/AMDGPUInstructions.td
index 83e1359950da5caf667962854f04143027dd93bb..b44d2482d5e301fddd21b2c9e9f33b041a2b7f96 100644 (file)
--- a/lib/Target/R600/AMDGPUInstructions.td
+++ b/lib/Target/R600/AMDGPUInstructions.td
@@ -276,6 +276,14 @@ multiclass BFIPatterns <Instruction BFI_INT> {
 
 }
 
+// SHA-256 Ma patterns
+
+// ((x & z) | (y & (x | z))) -> BFI_INT (XOR x, y), z, y
+class SHA256MaPattern <Instruction BFI_INT, Instruction XOR> : Pat <
+  (or (and i32:$x, i32:$z), (and i32:$y, (or i32:$x, i32:$z))),
+  (BFI_INT (XOR i32:$x, i32:$y), i32:$z, i32:$y)
+>;
+
 include "R600Instructions.td"
 
 include "SIInstrInfo.td"

diff --git a/lib/Target/R600/R600Instructions.td b/lib/Target/R600/R600Instructions.td
index 6bae4b3079cfcb01f7196480a1efbb780968cc9c..0466c187ef09c595f477dd9b5678937f215eebd1 100644 (file)
--- a/lib/Target/R600/R600Instructions.td
+++ b/lib/Target/R600/R600Instructions.td
@@ -1669,6 +1669,9 @@ let hasSideEffects = 1 in {
 
   def : Pat<(fp_to_uint f32:$src0), (FLT_TO_UINT_eg (TRUNC $src0))>;
 
+  // SHA-256 Patterns
+  def : SHA256MaPattern <BFI_INT_eg, XOR_INT>;
+
   def EG_ExportSwz : ExportSwzInst {
     let Word1{19-16} = 0; // BURST_COUNT
     let Word1{20} = 1; // VALID_PIXEL_MODE

diff --git a/lib/Target/R600/SIInstructions.td b/lib/Target/R600/SIInstructions.td
index 3ff4548354fe8a4b03a06b7a16d9574b81b378e5..fcae109fea7c85bd0ac611377b0050fa32fea719 100644 (file)
--- a/lib/Target/R600/SIInstructions.td
+++ b/lib/Target/R600/SIInstructions.td
@@ -1592,4 +1592,10 @@ def : Pat<
   (V_CMP_U_F32_e64 $src0, $src1)
 >;
 
+//============================================================================//
+// Miscellaneous Optimization Patterns
+//============================================================================//
+
+def : SHA256MaPattern <V_BFI_B32, V_XOR_B32_e32>;
+
 } // End isSI predicate

diff --git a/test/CodeGen/R600/bfi_int.ll b/test/CodeGen/R600/bfi_int.ll
index c9015a6e78aea9f74ed09a478af6b5f987f45f24..4244dcf3c77bdc33b4b234103e2909fd30884f75 100644 (file)
--- a/test/CodeGen/R600/bfi_int.ll
+++ b/test/CodeGen/R600/bfi_int.ll
@@ -32,3 +32,21 @@ entry:
   store i32 %2, i32 addrspace(1)* %out
   ret void
 }
+
+; SHA-256 Ma function
+; ((x & z) | (y & (x | z)))
+; R600-CHECK: @bfi_sha256_ma
+; R600-CHECK: XOR_INT * [[DST:T[0-9]+\.[XYZW]]], {{T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; R600-CHECK: BFI_INT * {{T[0-9]+\.[XYZW]}}, {{[[DST]]|PV.x}}, {{T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; SI-CHECK: V_XOR_B32_e32 [[DST:VGPR[0-9]+]], {{VGPR[0-9]+, VGPR[0-9]+}}
+; SI-CHECK: V_BFI_B32 {{VGPR[0-9]+}}, [[DST]], {{VGPR[0-9]+, VGPR[0-9]+}}
+
+define void @bfi_sha256_ma(i32 addrspace(1)* %out, i32 %x, i32 %y, i32 %z) {
+entry:
+  %0 = and i32 %x, %z
+  %1 = or i32 %x, %z
+  %2 = and i32 %y, %1
+  %3 = or i32 %0, %2
+  store i32 %3, i32 addrspace(1)* %out
+  ret void
+}