Merging r200743:

------------------------------------------------------------------------
r200743 | michel.daenzer | 2014-02-03 23:12:38 -0800 (Mon, 03 Feb 2014) | 11 lines

R600/SI: Fix fneg for 0.0

V_ADD_F32 with source modifier does not produce -0.0 for this. Just
manipulate the sign bit directly instead.

Also add a pattern for (fneg (fabs ...)).

Fixes a bunch of bit encoding piglit tests with radeonsi.

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/branches/release_34@204643 91177308-0d34-0410-b5e6-96231b3b80d8

4 files changed, 87 insertions, 18 deletions

diff --git a/lib/Target/R600/SIInstructions.td b/lib/Target/R600/SIInstructions.td
index 6d38481cd2..ae641164b4 100644
--- a/lib/Target/R600/SIInstructions.td
+++ b/lib/Target/R600/SIInstructions.td
@@ -1668,16 +1668,30 @@ def : Pat <
    0 /* ABS */, 1 /* CLAMP */, 0 /* OMOD */, 0 /* NEG */)
 >;
 
+/********** ================================ **********/
+/********** Floating point absolute/negative **********/
+/********** ================================ **********/
+
+// Manipulate the sign bit directly, as e.g. using the source negation modifier
+// in V_ADD_F32_e64 $src, 0, [...] does not result in -0.0 for $src == +0.0,
+// breaking the piglit *s-floatBitsToInt-neg* tests
+
+// TODO: Look into not implementing isFNegFree/isFAbsFree for SI, and possibly
+// removing these patterns
+
+def : Pat <
+  (fneg (fabs f32:$src)),
+  (V_OR_B32_e32 $src, (V_MOV_B32_e32 0x80000000)) /* Set sign bit */
+>;
+
 def : Pat <
   (fabs f32:$src),
-  (V_ADD_F32_e64 $src, (i32 0 /* SRC1 */),
-   1 /* ABS */, 0 /* CLAMP */, 0 /* OMOD */, 0 /* NEG */)
+  (V_AND_B32_e32 $src, (V_MOV_B32_e32 0x7fffffff)) /* Clear sign bit */
 >;
 
 def : Pat <
   (fneg f32:$src),
-  (V_ADD_F32_e64 $src, (i32 0 /* SRC1 */),
-   0 /* ABS */, 0 /* CLAMP */, 0 /* OMOD */, 1 /* NEG */)
+  (V_XOR_B32_e32 $src, (V_MOV_B32_e32 0x80000000)) /* Toggle sign bit */
 >;
 
 /********** ================== **********/
diff --git a/test/CodeGen/R600/fabs.ll b/test/CodeGen/R600/fabs.ll
index a5f5df96b5..2cd3a4f604 100644
--- a/test/CodeGen/R600/fabs.ll
+++ b/test/CodeGen/R600/fabs.ll
@@ -9,7 +9,7 @@
 ; R600-CHECK-NOT: AND
 ; R600-CHECK: |PV.{{[XYZW]}}|
 ; SI-CHECK-LABEL: @fabs_free
-; SI-CHECK: V_ADD_F32_e64 v{{[0-9]}}, s{{[0-9]}}, 0, 1, 0, 0, 0
+; SI-CHECK: V_AND_B32
 
 define void @fabs_free(float addrspace(1)* %out, i32 %in) {
 entry:
@@ -23,8 +23,8 @@ entry:
 ; R600-CHECK: |{{(PV|T[0-9])\.[XYZW]}}|
 ; R600-CHECK: |{{(PV|T[0-9])\.[XYZW]}}|
 ; SI-CHECK-LABEL: @fabs_v2
-; SI-CHECK: V_ADD_F32_e64 v{{[0-9]}}, s{{[0-9]}}, 0, 1, 0, 0, 0
-; SI-CHECK: V_ADD_F32_e64 v{{[0-9]}}, s{{[0-9]}}, 0, 1, 0, 0, 0
+; SI-CHECK: V_AND_B32
+; SI-CHECK: V_AND_B32
 define void @fabs_v2(<2 x float> addrspace(1)* %out, <2 x float> %in) {
 entry:
   %0 = call <2 x float> @llvm.fabs.v2f32(<2 x float> %in)
@@ -38,10 +38,10 @@ entry:
 ; R600-CHECK: |{{(PV|T[0-9])\.[XYZW]}}|
 ; R600-CHECK: |{{(PV|T[0-9])\.[XYZW]}}|
 ; SI-CHECK-LABEL: @fabs_v4
-; SI-CHECK: V_ADD_F32_e64 v{{[0-9]}}, s{{[0-9]}}, 0, 1, 0, 0, 0
-; SI-CHECK: V_ADD_F32_e64 v{{[0-9]}}, s{{[0-9]}}, 0, 1, 0, 0, 0
-; SI-CHECK: V_ADD_F32_e64 v{{[0-9]}}, s{{[0-9]}}, 0, 1, 0, 0, 0
-; SI-CHECK: V_ADD_F32_e64 v{{[0-9]}}, s{{[0-9]}}, 0, 1, 0, 0, 0
+; SI-CHECK: V_AND_B32
+; SI-CHECK: V_AND_B32
+; SI-CHECK: V_AND_B32
+; SI-CHECK: V_AND_B32
 define void @fabs_v4(<4 x float> addrspace(1)* %out, <4 x float> %in) {
 entry:
   %0 = call <4 x float> @llvm.fabs.v4f32(<4 x float> %in)
diff --git a/test/CodeGen/R600/fneg-fabs.ll b/test/CodeGen/R600/fneg-fabs.ll
new file mode 100644
index 0000000000..d95e1311bc
--- /dev/null
+++ b/test/CodeGen/R600/fneg-fabs.ll
@@ -0,0 +1,55 @@
+; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s --check-prefix=R600-CHECK
+; RUN: llc < %s -march=r600 -mcpu=SI -verify-machineinstrs | FileCheck %s --check-prefix=SI-CHECK
+
+; DAGCombiner will transform:
+; (fabs (f32 bitcast (i32 a))) => (f32 bitcast (and (i32 a), 0x7FFFFFFF))
+; unless isFabsFree returns true
+
+; R600-CHECK-LABEL: @fneg_fabs_free
+; R600-CHECK-NOT: AND
+; R600-CHECK: |PV.{{[XYZW]}}|
+; R600-CHECK: -PV
+; SI-CHECK-LABEL: @fneg_fabs_free
+; SI-CHECK: V_OR_B32
+
+define void @fneg_fabs_free(float addrspace(1)* %out, i32 %in) {
+entry:
+  %0 = bitcast i32 %in to float
+  %1 = call float @fabs(float %0)
+  %2 = fsub float -0.000000e+00, %1
+  store float %2, float addrspace(1)* %out
+  ret void
+}
+
+; R600-CHECK-LABEL: @fneg_fabs_v2
+; R600-CHECK: |{{(PV|T[0-9])\.[XYZW]}}|
+; R600-CHECK: -PV
+; R600-CHECK: |{{(PV|T[0-9])\.[XYZW]}}|
+; R600-CHECK: -PV
+; SI-CHECK-LABEL: @fneg_fabs_v2
+; SI-CHECK: V_OR_B32
+; SI-CHECK: V_OR_B32
+define void @fneg_fabs_v2(<2 x float> addrspace(1)* %out, <2 x float> %in) {
+entry:
+  %0 = call <2 x float> @llvm.fabs.v2f32(<2 x float> %in)
+  %1 = fsub <2 x float> <float -0.000000e+00, float -0.000000e+00>, %0
+  store <2 x float> %1, <2 x float> addrspace(1)* %out
+  ret void
+}
+
+; SI-CHECK-LABEL: @fneg_fabs_v4
+; SI-CHECK: V_OR_B32
+; SI-CHECK: V_OR_B32
+; SI-CHECK: V_OR_B32
+; SI-CHECK: V_OR_B32
+define void @fneg_fabs_v4(<4 x float> addrspace(1)* %out, <4 x float> %in) {
+entry:
+  %0 = call <4 x float> @llvm.fabs.v4f32(<4 x float> %in)
+  %1 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %0
+  store <4 x float> %1, <4 x float> addrspace(1)* %out
+  ret void
+}
+
+declare float @fabs(float ) readnone
+declare <2 x float> @llvm.fabs.v2f32(<2 x float> ) readnone
+declare <4 x float> @llvm.fabs.v4f32(<4 x float> ) readnone
diff --git a/test/CodeGen/R600/fneg.ll b/test/CodeGen/R600/fneg.ll
index 9446aa8ea9..f4e6be6246 100644
--- a/test/CodeGen/R600/fneg.ll
+++ b/test/CodeGen/R600/fneg.ll
@@ -4,7 +4,7 @@
 ; R600-CHECK-LABEL: @fneg
 ; R600-CHECK: -PV
 ; SI-CHECK-LABEL: @fneg
-; SI-CHECK: V_ADD_F32_e64 v{{[0-9]}}, s{{[0-9]}}, 0, 0, 0, 0, 1
+; SI-CHECK: V_XOR_B32
 define void @fneg(float addrspace(1)* %out, float %in) {
 entry:
   %0 = fsub float -0.000000e+00, %in
@@ -16,8 +16,8 @@ entry:
 ; R600-CHECK: -PV
 ; R600-CHECK: -PV
 ; SI-CHECK-LABEL: @fneg_v2
-; SI-CHECK: V_ADD_F32_e64 v{{[0-9]}}, s{{[0-9]}}, 0, 0, 0, 0, 1
-; SI-CHECK: V_ADD_F32_e64 v{{[0-9]}}, s{{[0-9]}}, 0, 0, 0, 0, 1
+; SI-CHECK: V_XOR_B32
+; SI-CHECK: V_XOR_B32
 define void @fneg_v2(<2 x float> addrspace(1)* nocapture %out, <2 x float> %in) {
 entry:
   %0 = fsub <2 x float> <float -0.000000e+00, float -0.000000e+00>, %in
@@ -31,10 +31,10 @@ entry:
 ; R600-CHECK: -PV
 ; R600-CHECK: -PV
 ; SI-CHECK-LABEL: @fneg_v4
-; SI-CHECK: V_ADD_F32_e64 v{{[0-9]}}, s{{[0-9]}}, 0, 0, 0, 0, 1
-; SI-CHECK: V_ADD_F32_e64 v{{[0-9]}}, s{{[0-9]}}, 0, 0, 0, 0, 1
-; SI-CHECK: V_ADD_F32_e64 v{{[0-9]}}, s{{[0-9]}}, 0, 0, 0, 0, 1
-; SI-CHECK: V_ADD_F32_e64 v{{[0-9]}}, s{{[0-9]}}, 0, 0, 0, 0, 1
+; SI-CHECK: V_XOR_B32
+; SI-CHECK: V_XOR_B32
+; SI-CHECK: V_XOR_B32
+; SI-CHECK: V_XOR_B32
 define void @fneg_v4(<4 x float> addrspace(1)* nocapture %out, <4 x float> %in) {
 entry:
   %0 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %in