[AArch64][CostModel] Add NFC tests for extractelement cost

[sushgokh]

A successive patch aims to reduce the extractelement cost where the only user(s) is fmul instruction.

[llvmbot]

@llvm/pr-subscribers-llvm-analysis

Author: Sushant Gokhale (sushgokh)

Changes

A successive patch aims to reduce the extractelement cost where the only user(s) is fmul instruction.

---

Full diff: https://github.com/llvm/llvm-project/pull/108941.diff

1 Files Affected:

• (added) llvm/test/Analysis/CostModel/AArch64/extract_float.ll (+207)

diff --git a/llvm/test/Analysis/CostModel/AArch64/extract_float.ll b/llvm/test/Analysis/CostModel/AArch64/extract_float.ll
new file mode 100644
index 00000000000000..b354b452cb140d
--- /dev/null
+++ b/llvm/test/Analysis/CostModel/AArch64/extract_float.ll
@@ -0,0 +1,207 @@
+; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py UTC_ARGS: --version 5
+; RUN: opt < %s -passes="print<cost-model>" 2>&1 -disable-output -mtriple=aarch64-unknown-linux \
+; RUN:       -mattr=-fullfp16 | FileCheck %s --check-prefixes=CHECK,NOFP16
+; RUN: opt < %s -passes="print<cost-model>" 2>&1 -disable-output -mtriple=aarch64-unknown-linux \
+; RUN:       -mattr=+fullfp16 | FileCheck %s --check-prefixes=CHECK,FULLFP16
+
+; res = lane 0 * lane 1
+define double @extract_case1(<2 x double> %a) {
+; CHECK-LABEL: 'extract_case1'
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %0 = extractelement <2 x double> %a, i32 0
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %1 = extractelement <2 x double> %a, i32 1
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %res = fmul double %0, %1
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %res
+entry:
+  %1 = extractelement <2 x double> %a, i32 0
+  %2 = extractelement <2 x double> %a, i32 1
+  %res = fmul double %1, %2
+  ret double %res
+}
+
+; res = lane 1 * lane 1
+define double @extract_case2(<2 x double> %a) {
+; CHECK-LABEL: 'extract_case2'
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %0 = extractelement <2 x double> %a, i32 1
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %res = fmul double %0, %0
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %res
+entry:
+  %1 = extractelement <2 x double> %a, i32 1
+  %res = fmul double %1, %1
+  ret double %res
+}
+
+; res = lane 0 * lane 0
+define double @extract_case3(<2 x double> %a) {
+; CHECK-LABEL: 'extract_case3'
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %0 = extractelement <2 x double> %a, i32 0
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %res = fmul double %0, %0
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %res
+entry:
+  %1 = extractelement <2 x double> %a, i32 0
+  %res = fmul double %1, %1
+  ret double %res
+}
+
+; res = lane 0 * scalar
+define double @extract_case4(<2 x double> %a, double %b) {
+; CHECK-LABEL: 'extract_case4'
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %0 = extractelement <2 x double> %a, i32 0
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %res = fmul double %0, %b
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %res
+entry:
+  %1 = extractelement <2 x double> %a, i32 0
+  %res = fmul double %1, %b
+  ret double %res
+}
+
+; res = lane 1 * scalar
+define double @extract_case5(<2 x double> %a, double %b) {
+; CHECK-LABEL: 'extract_case5'
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %0 = extractelement <2 x double> %a, i32 1
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %res = fmul double %0, %b
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %res
+entry:
+  %1 = extractelement <2 x double> %a, i32 1
+  %res = fmul double %1, %b
+  ret double %res
+}
+
+; Input vector = <3 x double> (i.e. odd length vector)
+; res = lane 0 * lane 1
+define double @extract_case6(<3 x double> %a) {
+; CHECK-LABEL: 'extract_case6'
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %0 = extractelement <3 x double> %a, i32 0
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %1 = extractelement <3 x double> %a, i32 1
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %res = fmul double %0, %1
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %res
+entry:
+  %1 = extractelement <3 x double> %a, i32 0
+  %2 = extractelement <3 x double> %a, i32 1
+  %res = fmul double %1, %2
+  ret double %res
+}
+
+; res = lane 1 * lane 2
+; Extract from lane 2 is equivalent to extract from lane 0 of other 128-bit
+; register. But for other register sizes, this is not the case.
+define double @extract_case7(<4 x double> %a) {
+; CHECK-LABEL: 'extract_case7'
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %0 = extractelement <4 x double> %a, i32 1
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %1 = extractelement <4 x double> %a, i32 2
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %res = fmul double %0, %1
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %res
+entry:
+  %1 = extractelement <4 x double> %a, i32 1
+  %2 = extractelement <4 x double> %a, i32 2
+  %res = fmul double %1, %2
+  ret double %res
+}
+
+; res = lane 0 * lane 1
+; Additional insert of extract from lane 1.
+define double @extract_case8(<2 x double> %a) {
+; CHECK-LABEL: 'extract_case8'
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %0 = extractelement <2 x double> %a, i32 0
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %1 = extractelement <2 x double> %a, i32 1
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %2 = insertelement <2 x double> %a, double %1, i32 0
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %3 = call double @llvm.vector.reduce.fmul.v2f64(double 0.000000e+00, <2 x double> %2)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %4 = fmul double %0, %1
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %5 = fmul double %3, %4
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %5
+entry:
+  %1 = extractelement <2 x double> %a, i32 0
+  %2 = extractelement <2 x double> %a, i32 1
+  %3 = insertelement <2 x double> %a, double %2, i32 0
+  %4 = call double @llvm.vector.reduce.fmul.v2f64(double 0.0, <2 x double> %3)
+  %5 = fmul double %1, %2
+  %6 = fmul double %4, %5
+  ret double %6
+}
+
+; res = lane 0 * lane 1
+; Additional insert of extract from lane 1.
+define double @extract_case9(<2 x double> %a) {
+; CHECK-LABEL: 'extract_case9'
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %0 = extractelement <2 x double> %a, i32 0
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %1 = extractelement <2 x double> %a, i32 1
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %2 = insertelement <2 x double> %a, double %1, i32 0
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %3 = call double @llvm.vector.reduce.fadd.v2f64(double 0.000000e+00, <2 x double> %2)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %4 = fmul double %0, %1
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %5 = fmul double %3, %4
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %5
+entry:
+  %1 = extractelement <2 x double> %a, i32 0
+  %2 = extractelement <2 x double> %a, i32 1
+  %3 = insertelement <2 x double> %a, double %2, i32 0
+  %4 = call double @llvm.vector.reduce.fadd.v2f64(double 0.0, <2 x double> %3)
+  %5 = fmul double %1, %2
+  %6 = fmul double %4, %5
+  ret double %6
+}
+
+; res = lane 0 * lane 1
+; Extract from lane 1 passed as function param.
+define double @extract_case10(<4 x double> %a) {
+; CHECK-LABEL: 'extract_case10'
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %0 = extractelement <4 x double> %a, i32 0
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %1 = extractelement <4 x double> %a, i32 1
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: call void @foo(double %1)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %2 = fmul double %0, %1
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %2
+entry:
+  %1 = extractelement <4 x double> %a, i32 0
+  %2 = extractelement <4 x double> %a, i32 1
+  call void @foo(double %2)
+  %3 = fmul double %1, %2
+  ret double %3
+}
+
+; res = lane 0 * lane 1
+define half @extract_case11(<2 x half> %a) {
+; NOFP16-LABEL: 'extract_case11'
+; NOFP16-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %0 = extractelement <2 x half> %a, i32 0
+; NOFP16-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %1 = extractelement <2 x half> %a, i32 1
+; NOFP16-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %res = fmul half %0, %1
+; NOFP16-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret half %res
+; FULLFP16-LABEL: 'extract_case11'
+; FULLFP16-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %0 = extractelement <2 x half> %a, i32 0
+; FULLFP16-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %1 = extractelement <2 x half> %a, i32 1
+; FULLFP16-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %res = fmul half %0, %1
+; FULLFP16-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret half %res
+entry:
+  %1 = extractelement <2 x half> %a, i32 0
+  %2 = extractelement <2 x half> %a, i32 1
+  %res = fmul half %1, %2
+  ret half %res
+}
+
+; res = lane 0 * lane 1
+define float @extract_case12(<2 x float> %a) {
+; CHECK-LABEL: 'extract_case12'
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %0 = extractelement <2 x float> %a, i32 0
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %1 = extractelement <2 x float> %a, i32 1
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %res = fmul float %0, %1
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %res
+entry:
+  %1 = extractelement <2 x float> %a, i32 0
+  %2 = extractelement <2 x float> %a, i32 1
+  %res = fmul float %1, %2
+  ret float %res
+}
+
+; res = lane 0 + lane 1
+; Use of bin-op other than fmul.
+define double @extract_case13(<2 x double> %a) {
+; CHECK-LABEL: 'extract_case13'
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %0 = extractelement <2 x double> %a, i32 0
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %1 = extractelement <2 x double> %a, i32 1
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %res = fadd double %0, %1
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %res
+entry:
+  %1 = extractelement <2 x double> %a, i32 0
+  %2 = extractelement <2 x double> %a, i32 1
+  %res = fadd double %1, %2
+  ret double %res
+}
+
+declare void @foo(double)

[davemgreen]

Some of the half tests look like they should already be higher for fp16. It looks like you are hoping to adjust the cost of fmul lane operations?

These look OK to me, Hopefully the full patch won't require any adjustments. LGTM.

[sushgokh]

Some of the half tests look like they should already be higher for fp16. It looks like you are hoping to adjust the cost of fmul lane operations?
They should be higher for non-fp16 since convert to float(32 bits) take place. The actual patch would try to reduce the cost of extracts from lane!=0