From a42cc1f883f7a4aabd7f29e2fb842a51a213e37d Mon Sep 17 00:00:00 2001
From: Martin Jambor <mjambor@suse.cz>
Date: Mon, 25 May 2020 16:07:48 +0200
Subject: [PATCH] BRIG FE testsuite: Fix dump scan patterns in packed.hsail
 test

Starting with r11-165-eb72dc663e9 which converted DECL_GIMPLE_REG_P to
DECL_NOT_GIMPLE_REG_P we have failing BRIG testcase:

-PASS: packed.hsail.brig scan-tree-dump gimple "_[0-9]+ = q2 \\+ q3;"
-PASS: packed.hsail.brig scan-tree-dump gimple "= VEC_PERM_EXPR <new_output.[0-9]+_[0-9]+, [a-z0-9_]+, { 0, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31 }>;"
+FAIL: packed.hsail.brig scan-tree-dump gimple "_[0-9]+ = q2 \\+ q3;"
+FAIL: packed.hsail.brig scan-tree-dump gimple "= VEC_PERM_EXPR <new_output.[0-9]+_[0-9]+, [a-z0-9_]+, { 0, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31 }>;"

because the gimplifier is now smarter and generates nicer code, which
however, does not match the regexp in the testsuite:

--- before/packed.hsail.brig.005t.gimple        2020-05-12 17:59:26.434305513 +0200
+++ after/packed.hsail.brig.005t.gimple 2020-05-12 17:52:34.477055987 +0200
@@ -109,277 +109,267 @@
   q2 = q1 + _24;
   _25 = VEC_PERM_EXPR <q1, q1, { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }>;
   q3 = q2 + _25;
-  _26 = q2 + q3;
-  new_output.11 = _26;
-  new_output.21_27 = new_output.11;
-  _28 = VEC_PERM_EXPR <new_output.21_27, q4, { 0, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31 }>;
-  s_output.12 = _28;
+  new_output.11 = q2 + q3;
+  s_output.12 = VEC_PERM_EXPR <new_output.11, q4, { 0, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31 }>;
   q4 = s_output.12;

I have looked at the SSA dump and verified that the variable in
question is a gimple register because it gets its SSA name.  I have
not looked into why the gimplifier previously though it had to go
through the additional temporaries though.

Tested with make -k check-brig.

	* brig.dg/test/gimple/packed.hsail: Fix scan dump patterns.
---
 gcc/testsuite/ChangeLog                        | 4 ++++
 gcc/testsuite/brig.dg/test/gimple/packed.hsail | 4 ++--
 2 files changed, 6 insertions(+), 2 deletions(-)

diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog
index dfb92ec5748..58521437e2d 100644
--- a/gcc/testsuite/ChangeLog
+++ b/gcc/testsuite/ChangeLog
@@ -1,3 +1,7 @@
+2020-05-25  Martin Jambor  <mjambor@suse.cz>
+
+	* brig.dg/test/gimple/packed.hsail: Fix scan dump patterns.
+
 2020-05-25  Richard Biener  <rguenther@suse.de>
 
 	PR tree-optimization/95308
diff --git a/gcc/testsuite/brig.dg/test/gimple/packed.hsail b/gcc/testsuite/brig.dg/test/gimple/packed.hsail
index 9137490b2ab..1e2bb53de0d 100644
--- a/gcc/testsuite/brig.dg/test/gimple/packed.hsail
+++ b/gcc/testsuite/brig.dg/test/gimple/packed.hsail
@@ -61,10 +61,10 @@ prog kernel &Kernel(kernarg_u64 %input_ptr, kernarg_u64 %output_ptr)
 /* For the add_ss we assume performing the computation over the whole vector is cheaper than */
 /* extracting the scalar and performing a scalar operation. This aims to stay in the vector
 /* datapath as long as possible. */
-/* { dg-final { scan-tree-dump "_\[0-9\]+ = q2 \\\+ q3;" "gimple" } } */
+/* { dg-final { scan-tree-dump "new_output.\[0-9\]+ = q2 \\\+ q3;" "gimple" } } */
 
 /* Insert the lowest element of the result to the lowest element of the result register. */
-/* { dg-final { scan-tree-dump "= VEC_PERM_EXPR <new_output.\[0-9\]+_\[0-9\]+, \[a-z0-9_\]+, { 0, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31 }>;" "gimple" } } */
+/* { dg-final { scan-tree-dump "= VEC_PERM_EXPR <new_output.\[0-9\]+, \[a-z0-9_\]+, { 0, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31 }>;" "gimple" } } */
 
 /* FIXME */
 /* { dg-final { scan-tree-dump "q4 = \(VIEW_CONVERT_EXPR<uint128_t>\\\()?s_output.\[0-9\]+\(_\[0-9\]+\)*\\\)?;" "gimple" } } */
-- 
2.30.2