[CostModel][X86] Add basic GFNI target test coverage for shift/rotate costs

Author: RKSimon

llvm/test/Analysis/CostModel/X86/fshl-codesize.ll

@@ -15,6 +15,7 @@
 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mcpu=slm | FileCheck %s --check-prefixes=CHECK,SSE42
 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mcpu=goldmont | FileCheck %s --check-prefixes=CHECK,SSE42
 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mcpu=btver2 | FileCheck %s --check-prefixes=CHECK,AVX,AVX1
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mcpu=tigerlake | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512GFNI
 
 ; Verify the cost of vector logical shift right instructions.
 
@@ -240,6 +241,10 @@ define <8 x i16> @var_shift_v8i16(<8 x i16> %a, <8 x i16> %b) {
 ; AVX512BW-LABEL: 'var_shift_v8i16'
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i16> %a, %b
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX512GFNI-LABEL: 'var_shift_v8i16'
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i16> %a, %b
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
 ;
   %shift = shl <8 x i16> %a, %b
   ret <8 x i16> %shift
@@ -273,6 +278,10 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
 ; AVX512BW-LABEL: 'var_shift_v16i16'
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i16> %a, %b
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX512GFNI-LABEL: 'var_shift_v16i16'
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i16> %a, %b
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
 ;
   %shift = shl <16 x i16> %a, %b
   ret <16 x i16> %shift
@@ -306,6 +315,10 @@ define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
 ; AVX512BW-LABEL: 'var_shift_v32i16'
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i16> %a, %b
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512GFNI-LABEL: 'var_shift_v32i16'
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i16> %a, %b
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
 ;
   %shift = shl <32 x i16> %a, %b
   ret <32 x i16> %shift
@@ -335,6 +348,10 @@ define <16 x i8> @var_shift_v16i8(<16 x i8> %a, <16 x i8> %b) {
 ; AVX512BW-LABEL: 'var_shift_v16i8'
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <16 x i8> %a, %b
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512GFNI-LABEL: 'var_shift_v16i8'
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <16 x i8> %a, %b
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
 ;
   %shift = shl <16 x i8> %a, %b
   ret <16 x i8> %shift
@@ -397,6 +414,10 @@ define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
 ; AVX512BW-LABEL: 'var_shift_v64i8'
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %shift = shl <64 x i8> %a, %b
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512GFNI-LABEL: 'var_shift_v64i8'
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %shift = shl <64 x i8> %a, %b
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
 ;
   %shift = shl <64 x i8> %a, %b
   ret <64 x i8> %shift
@@ -762,6 +783,12 @@ define <32 x i16> @splatvar_shift_v32i16(<32 x i16> %a, i16 %b) {
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <32 x i16> %insert, <32 x i16> undef, <32 x i32> zeroinitializer
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <32 x i16> %a, %splat
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512GFNI-LABEL: 'splatvar_shift_v32i16'
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <32 x i16> undef, i16 %b, i32 0
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <32 x i16> %insert, <32 x i16> undef, <32 x i32> zeroinitializer
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <32 x i16> %a, %splat
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
 ;
   %insert = insertelement <32 x i16> undef, i16 %b, i32 0
   %splat = shufflevector <32 x i16> %insert, <32 x i16> undef, <32 x i32> zeroinitializer
@@ -915,6 +942,12 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %shift = shl <64 x i8> %a, %splat
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512GFNI-LABEL: 'splatvar_shift_v64i8'
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %shift = shl <64 x i8> %a, %splat
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
 ;
   %insert = insertelement <64 x i8> undef, i8 %b, i32 0
   %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
@@ -1186,6 +1219,10 @@ define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
 ; AVX512BW-LABEL: 'constant_shift_v32i16'
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512GFNI-LABEL: 'constant_shift_v32i16'
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
 ;
   %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
   ret <32 x i16> %shift
@@ -1215,6 +1252,10 @@ define <16 x i8> @constant_shift_v16i8(<16 x i8> %a) {
 ; AVX512BW-LABEL: 'constant_shift_v16i8'
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512GFNI-LABEL: 'constant_shift_v16i8'
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
 ;
   %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
   ret <16 x i8> %shift
@@ -1277,6 +1318,10 @@ define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
 ; AVX512BW-LABEL: 'constant_shift_v64i8'
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512GFNI-LABEL: 'constant_shift_v64i8'
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
 ;
   %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
   ret <64 x i8> %shift
@@ -1477,6 +1522,10 @@ define <16 x i16> @splatconstant_shift_v16i16(<16 x i16> %a) {
 ; AVX512BW-LABEL: 'splatconstant_shift_v16i16'
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX512GFNI-LABEL: 'splatconstant_shift_v16i16'
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
 ;
   %shift = shl <16 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
   ret <16 x i16> %shift
@@ -1514,6 +1563,10 @@ define <32 x i16> @splatconstant_shift_v32i16(<32 x i16> %a) {
 ; AVX512BW-LABEL: 'splatconstant_shift_v32i16'
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512GFNI-LABEL: 'splatconstant_shift_v32i16'
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
 ;
   %shift = shl <32 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
   ret <32 x i16> %shift
@@ -1613,6 +1666,10 @@ define <64 x i8> @splatconstant_shift_v64i8(<64 x i8> %a) {
 ; AVX512BW-LABEL: 'splatconstant_shift_v64i8'
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512GFNI-LABEL: 'splatconstant_shift_v64i8'
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
 ;
   %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
   ret <64 x i8> %shift