[PATCH v2 1/1] R600: Make FMIN/MAXNUM legal on all asics
Jan Vesely
jan.vesely at rutgers.edu
Sun Apr 12 09:46:02 PDT 2015
v2: Add tests
Signed-off-by: Jan Vesely <jan.vesely at rutgers.edu>
---
lib/Target/R600/AMDGPUISelLowering.cpp | 2 +
lib/Target/R600/R600Instructions.td | 5 ++
lib/Target/R600/SIISelLowering.cpp | 2 -
test/CodeGen/R600/fmaxnum.ll | 91 ++++++++++++++++++++++++++++++++++
test/CodeGen/R600/fminnum.ll | 91 ++++++++++++++++++++++++++++++++++
5 files changed, 189 insertions(+), 2 deletions(-)
diff --git a/lib/Target/R600/AMDGPUISelLowering.cpp b/lib/Target/R600/AMDGPUISelLowering.cpp
index 33bcc80..2c56959 100644
--- a/lib/Target/R600/AMDGPUISelLowering.cpp
+++ b/lib/Target/R600/AMDGPUISelLowering.cpp
@@ -126,6 +126,8 @@ AMDGPUTargetLowering::AMDGPUTargetLowering(TargetMachine &TM,
setOperationAction(ISD::FFLOOR, MVT::f32, Legal);
setOperationAction(ISD::FRINT, MVT::f32, Legal);
setOperationAction(ISD::FTRUNC, MVT::f32, Legal);
+ setOperationAction(ISD::FMINNUM, MVT::f32, Legal);
+ setOperationAction(ISD::FMAXNUM, MVT::f32, Legal);
setOperationAction(ISD::FROUND, MVT::f32, Custom);
setOperationAction(ISD::FROUND, MVT::f64, Custom);
diff --git a/lib/Target/R600/R600Instructions.td b/lib/Target/R600/R600Instructions.td
index 291fb04..7126c82 100644
--- a/lib/Target/R600/R600Instructions.td
+++ b/lib/Target/R600/R600Instructions.td
@@ -683,6 +683,11 @@ def MUL_IEEE : R600_2OP_Helper <0x2, "MUL_IEEE", fmul>;
// TODO: Do these actually match the regular fmin/fmax behavior?
def MAX : R600_2OP_Helper <0x3, "MAX", AMDGPUfmax_legacy>;
def MIN : R600_2OP_Helper <0x4, "MIN", AMDGPUfmin_legacy>;
+// According to https://msdn.microsoft.com/en-us/library/windows/desktop/cc308050%28v=vs.85%29.aspx
+// DX10 min/max returns the other operand if one is NaN,
+// this matches http://llvm.org/docs/LangRef.html#llvm-minnum-intrinsic
+def MAX_DX10 : R600_2OP_Helper <0x5, "MAX_DX10", fmaxnum>;
+def MIN_DX10 : R600_2OP_Helper <0x6, "MIN_DX10", fminnum>;
// For the SET* instructions there is a naming conflict in TargetSelectionDAG.td,
// so some of the instruction names don't match the asm string.
diff --git a/lib/Target/R600/SIISelLowering.cpp b/lib/Target/R600/SIISelLowering.cpp
index 0a6166d..43507d8 100644
--- a/lib/Target/R600/SIISelLowering.cpp
+++ b/lib/Target/R600/SIISelLowering.cpp
@@ -76,8 +76,6 @@ SITargetLowering::SITargetLowering(TargetMachine &TM,
setOperationAction(ISD::FSIN, MVT::f32, Custom);
setOperationAction(ISD::FCOS, MVT::f32, Custom);
- setOperationAction(ISD::FMINNUM, MVT::f32, Legal);
- setOperationAction(ISD::FMAXNUM, MVT::f32, Legal);
setOperationAction(ISD::FMINNUM, MVT::f64, Legal);
setOperationAction(ISD::FMAXNUM, MVT::f64, Legal);
diff --git a/test/CodeGen/R600/fmaxnum.ll b/test/CodeGen/R600/fmaxnum.ll
index c105598..3029bd02 100644
--- a/test/CodeGen/R600/fmaxnum.ll
+++ b/test/CodeGen/R600/fmaxnum.ll
@@ -11,6 +11,9 @@ declare double @llvm.maxnum.f64(double, double)
; FUNC-LABEL: @test_fmax_f32
; SI: v_max_f32_e32
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT:T[0-9]+\.[XYZW]]]
+; EG: MAX_DX10 {{.*}}[[OUT]]
define void @test_fmax_f32(float addrspace(1)* %out, float %a, float %b) nounwind {
%val = call float @llvm.maxnum.f32(float %a, float %b) #0
store float %val, float addrspace(1)* %out, align 4
@@ -20,6 +23,10 @@ define void @test_fmax_f32(float addrspace(1)* %out, float %a, float %b) nounwin
; FUNC-LABEL: @test_fmax_v2f32
; SI: v_max_f32_e32
; SI: v_max_f32_e32
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT:T[0-9]+]]
+; EG: MAX_DX10 {{.*}}[[OUT]]
+; EG: MAX_DX10 {{.*}}[[OUT]]
define void @test_fmax_v2f32(<2 x float> addrspace(1)* %out, <2 x float> %a, <2 x float> %b) nounwind {
%val = call <2 x float> @llvm.maxnum.v2f32(<2 x float> %a, <2 x float> %b) #0
store <2 x float> %val, <2 x float> addrspace(1)* %out, align 8
@@ -31,6 +38,12 @@ define void @test_fmax_v2f32(<2 x float> addrspace(1)* %out, <2 x float> %a, <2
; SI: v_max_f32_e32
; SI: v_max_f32_e32
; SI: v_max_f32_e32
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT:T[0-9]+]]
+; EG: MAX_DX10 {{.*}}[[OUT]]
+; EG: MAX_DX10 {{.*}}[[OUT]]
+; EG: MAX_DX10 {{.*}}[[OUT]]
+; EG: MAX_DX10 {{.*}}[[OUT]]
define void @test_fmax_v4f32(<4 x float> addrspace(1)* %out, <4 x float> %a, <4 x float> %b) nounwind {
%val = call <4 x float> @llvm.maxnum.v4f32(<4 x float> %a, <4 x float> %b) #0
store <4 x float> %val, <4 x float> addrspace(1)* %out, align 16
@@ -46,6 +59,17 @@ define void @test_fmax_v4f32(<4 x float> addrspace(1)* %out, <4 x float> %a, <4
; SI: v_max_f32_e32
; SI: v_max_f32_e32
; SI: v_max_f32_e32
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT1:T[0-9]+]]
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT2:T[0-9]+]]
+; EG-DAG: MAX_DX10 {{.*}}[[OUT1]].X
+; EG-DAG: MAX_DX10 {{.*}}[[OUT1]].Y
+; EG-DAG: MAX_DX10 {{.*}}[[OUT1]].Z
+; EG-DAG: MAX_DX10 {{.*}}[[OUT1]].W
+; EG-DAG: MAX_DX10 {{.*}}[[OUT2]].X
+; EG-DAG: MAX_DX10 {{.*}}[[OUT2]].Y
+; EG-DAG: MAX_DX10 {{.*}}[[OUT2]].Z
+; EG-DAG: MAX_DX10 {{.*}}[[OUT2]].W
define void @test_fmax_v8f32(<8 x float> addrspace(1)* %out, <8 x float> %a, <8 x float> %b) nounwind {
%val = call <8 x float> @llvm.maxnum.v8f32(<8 x float> %a, <8 x float> %b) #0
store <8 x float> %val, <8 x float> addrspace(1)* %out, align 32
@@ -69,6 +93,27 @@ define void @test_fmax_v8f32(<8 x float> addrspace(1)* %out, <8 x float> %a, <8
; SI: v_max_f32_e32
; SI: v_max_f32_e32
; SI: v_max_f32_e32
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT1:T[0-9]+]]
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT2:T[0-9]+]]
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT3:T[0-9]+]]
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT4:T[0-9]+]]
+; EG-DAG: MAX_DX10 {{.*}}[[OUT1]].X
+; EG-DAG: MAX_DX10 {{.*}}[[OUT1]].Y
+; EG-DAG: MAX_DX10 {{.*}}[[OUT1]].Z
+; EG-DAG: MAX_DX10 {{.*}}[[OUT1]].W
+; EG-DAG: MAX_DX10 {{.*}}[[OUT2]].X
+; EG-DAG: MAX_DX10 {{.*}}[[OUT2]].Y
+; EG-DAG: MAX_DX10 {{.*}}[[OUT2]].Z
+; EG-DAG: MAX_DX10 {{.*}}[[OUT2]].W
+; EG-DAG: MAX_DX10 {{.*}}[[OUT3]].X
+; EG-DAG: MAX_DX10 {{.*}}[[OUT3]].Y
+; EG-DAG: MAX_DX10 {{.*}}[[OUT3]].Z
+; EG-DAG: MAX_DX10 {{.*}}[[OUT3]].W
+; EG-DAG: MAX_DX10 {{.*}}[[OUT4]].X
+; EG-DAG: MAX_DX10 {{.*}}[[OUT4]].Y
+; EG-DAG: MAX_DX10 {{.*}}[[OUT4]].Z
+; EG-DAG: MAX_DX10 {{.*}}[[OUT4]].W
define void @test_fmax_v16f32(<16 x float> addrspace(1)* %out, <16 x float> %a, <16 x float> %b) nounwind {
%val = call <16 x float> @llvm.maxnum.v16f32(<16 x float> %a, <16 x float> %b) #0
store <16 x float> %val, <16 x float> addrspace(1)* %out, align 64
@@ -79,6 +124,10 @@ define void @test_fmax_v16f32(<16 x float> addrspace(1)* %out, <16 x float> %a,
; SI-NOT: v_max_f32_e32
; SI: v_mov_b32_e32 [[REG:v[0-9]+]], 2.0
; SI: buffer_store_dword [[REG]]
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT:T[0-9]+\.[XYZW]]]
+; EG-NOT: MAX_DX10
+; EG: MOV {{.*}}[[OUT]], literal.{{[xy]}}
define void @constant_fold_fmax_f32(float addrspace(1)* %out) nounwind {
%val = call float @llvm.maxnum.f32(float 1.0, float 2.0) #0
store float %val, float addrspace(1)* %out, align 4
@@ -89,6 +138,11 @@ define void @constant_fold_fmax_f32(float addrspace(1)* %out) nounwind {
; SI-NOT: v_max_f32_e32
; SI: v_mov_b32_e32 [[REG:v[0-9]+]], 0x7fc00000
; SI: buffer_store_dword [[REG]]
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT:T[0-9]+\.[XYZW]]]
+; EG-NOT: MAX_DX10
+; EG: MOV {{.*}}[[OUT]], literal.{{[xy]}}
+; EG: 2143289344(nan)
define void @constant_fold_fmax_f32_nan_nan(float addrspace(1)* %out) nounwind {
%val = call float @llvm.maxnum.f32(float 0x7FF8000000000000, float 0x7FF8000000000000) #0
store float %val, float addrspace(1)* %out, align 4
@@ -99,6 +153,10 @@ define void @constant_fold_fmax_f32_nan_nan(float addrspace(1)* %out) nounwind {
; SI-NOT: v_max_f32_e32
; SI: v_mov_b32_e32 [[REG:v[0-9]+]], 1.0
; SI: buffer_store_dword [[REG]]
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT:T[0-9]+\.[XYZW]]]
+; EG-NOT: MAX_DX10
+; EG: MOV {{.*}}[[OUT]], literal.{{[xy]}}
define void @constant_fold_fmax_f32_val_nan(float addrspace(1)* %out) nounwind {
%val = call float @llvm.maxnum.f32(float 1.0, float 0x7FF8000000000000) #0
store float %val, float addrspace(1)* %out, align 4
@@ -109,6 +167,10 @@ define void @constant_fold_fmax_f32_val_nan(float addrspace(1)* %out) nounwind {
; SI-NOT: v_max_f32_e32
; SI: v_mov_b32_e32 [[REG:v[0-9]+]], 1.0
; SI: buffer_store_dword [[REG]]
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT:T[0-9]+\.[XYZW]]]
+; EG-NOT: MAX_DX10
+; EG: MOV {{.*}}[[OUT]], literal.{{[xy]}}
define void @constant_fold_fmax_f32_nan_val(float addrspace(1)* %out) nounwind {
%val = call float @llvm.maxnum.f32(float 0x7FF8000000000000, float 1.0) #0
store float %val, float addrspace(1)* %out, align 4
@@ -119,6 +181,10 @@ define void @constant_fold_fmax_f32_nan_val(float addrspace(1)* %out) nounwind {
; SI-NOT: v_max_f32_e32
; SI: v_mov_b32_e32 [[REG:v[0-9]+]], 0
; SI: buffer_store_dword [[REG]]
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT:T[0-9]+\.[XYZW]]]
+; EG-NOT: MAX_DX10
+; EG: MOV {{.*}}[[OUT]], literal.{{[xy]}}
define void @constant_fold_fmax_f32_p0_p0(float addrspace(1)* %out) nounwind {
%val = call float @llvm.maxnum.f32(float 0.0, float 0.0) #0
store float %val, float addrspace(1)* %out, align 4
@@ -129,6 +195,10 @@ define void @constant_fold_fmax_f32_p0_p0(float addrspace(1)* %out) nounwind {
; SI-NOT: v_max_f32_e32
; SI: v_mov_b32_e32 [[REG:v[0-9]+]], 0
; SI: buffer_store_dword [[REG]]
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT:T[0-9]+\.[XYZW]]]
+; EG-NOT: MAX_DX10
+; EG: MOV {{.*}}[[OUT]], literal.{{[xy]}}
define void @constant_fold_fmax_f32_p0_n0(float addrspace(1)* %out) nounwind {
%val = call float @llvm.maxnum.f32(float 0.0, float -0.0) #0
store float %val, float addrspace(1)* %out, align 4
@@ -139,6 +209,10 @@ define void @constant_fold_fmax_f32_p0_n0(float addrspace(1)* %out) nounwind {
; SI-NOT: v_max_f32_e32
; SI: v_mov_b32_e32 [[REG:v[0-9]+]], 0x80000000
; SI: buffer_store_dword [[REG]]
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT:T[0-9]+\.[XYZW]]]
+; EG-NOT: MAX_DX10
+; EG: MOV {{.*}}[[OUT]], literal.{{[xy]}}
define void @constant_fold_fmax_f32_n0_p0(float addrspace(1)* %out) nounwind {
%val = call float @llvm.maxnum.f32(float -0.0, float 0.0) #0
store float %val, float addrspace(1)* %out, align 4
@@ -149,6 +223,10 @@ define void @constant_fold_fmax_f32_n0_p0(float addrspace(1)* %out) nounwind {
; SI-NOT: v_max_f32_e32
; SI: v_mov_b32_e32 [[REG:v[0-9]+]], 0x80000000
; SI: buffer_store_dword [[REG]]
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT:T[0-9]+\.[XYZW]]]
+; EG-NOT: MAX_DX10
+; EG: MOV {{.*}}[[OUT]], literal.{{[xy]}}
define void @constant_fold_fmax_f32_n0_n0(float addrspace(1)* %out) nounwind {
%val = call float @llvm.maxnum.f32(float -0.0, float -0.0) #0
store float %val, float addrspace(1)* %out, align 4
@@ -157,6 +235,10 @@ define void @constant_fold_fmax_f32_n0_n0(float addrspace(1)* %out) nounwind {
; FUNC-LABEL: @fmax_var_immediate_f32
; SI: v_max_f32_e64 {{v[0-9]+}}, 2.0, {{s[0-9]+}}
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT:T[0-9]+\.[XYZW]]]
+; EG-NOT: MAX_DX10
+; EG: MOV {{.*}}[[OUT]], literal.{{[xy]}}
define void @fmax_var_immediate_f32(float addrspace(1)* %out, float %a) nounwind {
%val = call float @llvm.maxnum.f32(float %a, float 2.0) #0
store float %val, float addrspace(1)* %out, align 4
@@ -165,6 +247,9 @@ define void @fmax_var_immediate_f32(float addrspace(1)* %out, float %a) nounwind
; FUNC-LABEL: @fmax_immediate_var_f32
; SI: v_max_f32_e64 {{v[0-9]+}}, 2.0, {{s[0-9]+}}
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT:T[0-9]+\.[XYZW]]]
+; EG: MAX_DX10 {{.*}}[[OUT]], {{KC0\[[0-9]\].[XYZW]}}, literal.{{[xy]}}
define void @fmax_immediate_var_f32(float addrspace(1)* %out, float %a) nounwind {
%val = call float @llvm.maxnum.f32(float 2.0, float %a) #0
store float %val, float addrspace(1)* %out, align 4
@@ -174,6 +259,9 @@ define void @fmax_immediate_var_f32(float addrspace(1)* %out, float %a) nounwind
; FUNC-LABEL: @fmax_var_literal_f32
; SI: v_mov_b32_e32 [[REG:v[0-9]+]], 0x42c60000
; SI: v_max_f32_e32 {{v[0-9]+}}, {{s[0-9]+}}, [[REG]]
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT:T[0-9]+\.[XYZW]]]
+; EG: MAX_DX10 {{.*}}[[OUT]], {{KC0\[[0-9]\].[XYZW]}}, literal.{{[xy]}}
define void @fmax_var_literal_f32(float addrspace(1)* %out, float %a) nounwind {
%val = call float @llvm.maxnum.f32(float %a, float 99.0) #0
store float %val, float addrspace(1)* %out, align 4
@@ -183,6 +271,9 @@ define void @fmax_var_literal_f32(float addrspace(1)* %out, float %a) nounwind {
; FUNC-LABEL: @fmax_literal_var_f32
; SI: v_mov_b32_e32 [[REG:v[0-9]+]], 0x42c60000
; SI: v_max_f32_e32 {{v[0-9]+}}, {{s[0-9]+}}, [[REG]]
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT:T[0-9]+\.[XYZW]]]
+; EG: MAX_DX10 {{.*}}[[OUT]], {{KC0\[[0-9]\].[XYZW]}}, literal.{{[xy]}}
define void @fmax_literal_var_f32(float addrspace(1)* %out, float %a) nounwind {
%val = call float @llvm.maxnum.f32(float 99.0, float %a) #0
store float %val, float addrspace(1)* %out, align 4
diff --git a/test/CodeGen/R600/fminnum.ll b/test/CodeGen/R600/fminnum.ll
index 6b93b83..92666fe 100644
--- a/test/CodeGen/R600/fminnum.ll
+++ b/test/CodeGen/R600/fminnum.ll
@@ -1,5 +1,6 @@
; RUN: llc -march=amdgcn -mcpu=SI < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
; RUN: llc -march=amdgcn -mcpu=tonga < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
+; RUN: llc -march=r600 -mcpu=cypress < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
declare float @llvm.minnum.f32(float, float) #0
declare <2 x float> @llvm.minnum.v2f32(<2 x float>, <2 x float>) #0
@@ -9,6 +10,9 @@ declare <16 x float> @llvm.minnum.v16f32(<16 x float>, <16 x float>) #0
; FUNC-LABEL: @test_fmin_f32
; SI: v_min_f32_e32
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT:T[0-9]+\.[XYZW]]]
+; EG: MIN_DX10 {{.*}}[[OUT]]
define void @test_fmin_f32(float addrspace(1)* %out, float %a, float %b) nounwind {
%val = call float @llvm.minnum.f32(float %a, float %b) #0
store float %val, float addrspace(1)* %out, align 4
@@ -18,6 +22,10 @@ define void @test_fmin_f32(float addrspace(1)* %out, float %a, float %b) nounwin
; FUNC-LABEL: @test_fmin_v2f32
; SI: v_min_f32_e32
; SI: v_min_f32_e32
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT:T[0-9]+]]
+; EG: MIN_DX10 {{.*}}[[OUT]]
+; EG: MIN_DX10 {{.*}}[[OUT]]
define void @test_fmin_v2f32(<2 x float> addrspace(1)* %out, <2 x float> %a, <2 x float> %b) nounwind {
%val = call <2 x float> @llvm.minnum.v2f32(<2 x float> %a, <2 x float> %b) #0
store <2 x float> %val, <2 x float> addrspace(1)* %out, align 8
@@ -29,6 +37,12 @@ define void @test_fmin_v2f32(<2 x float> addrspace(1)* %out, <2 x float> %a, <2
; SI: v_min_f32_e32
; SI: v_min_f32_e32
; SI: v_min_f32_e32
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT:T[0-9]+]]
+; EG: MIN_DX10 {{.*}}[[OUT]]
+; EG: MIN_DX10 {{.*}}[[OUT]]
+; EG: MIN_DX10 {{.*}}[[OUT]]
+; EG: MIN_DX10 {{.*}}[[OUT]]
define void @test_fmin_v4f32(<4 x float> addrspace(1)* %out, <4 x float> %a, <4 x float> %b) nounwind {
%val = call <4 x float> @llvm.minnum.v4f32(<4 x float> %a, <4 x float> %b) #0
store <4 x float> %val, <4 x float> addrspace(1)* %out, align 16
@@ -44,6 +58,17 @@ define void @test_fmin_v4f32(<4 x float> addrspace(1)* %out, <4 x float> %a, <4
; SI: v_min_f32_e32
; SI: v_min_f32_e32
; SI: v_min_f32_e32
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT1:T[0-9]+]]
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT2:T[0-9]+]]
+; EG-DAG: MIN_DX10 {{.*}}[[OUT1]].X
+; EG-DAG: MIN_DX10 {{.*}}[[OUT1]].Y
+; EG-DAG: MIN_DX10 {{.*}}[[OUT1]].Z
+; EG-DAG: MIN_DX10 {{.*}}[[OUT1]].W
+; EG-DAG: MIN_DX10 {{.*}}[[OUT2]].X
+; EG-DAG: MIN_DX10 {{.*}}[[OUT2]].Y
+; EG-DAG: MIN_DX10 {{.*}}[[OUT2]].Z
+; EG-DAG: MIN_DX10 {{.*}}[[OUT2]].W
define void @test_fmin_v8f32(<8 x float> addrspace(1)* %out, <8 x float> %a, <8 x float> %b) nounwind {
%val = call <8 x float> @llvm.minnum.v8f32(<8 x float> %a, <8 x float> %b) #0
store <8 x float> %val, <8 x float> addrspace(1)* %out, align 32
@@ -67,6 +92,27 @@ define void @test_fmin_v8f32(<8 x float> addrspace(1)* %out, <8 x float> %a, <8
; SI: v_min_f32_e32
; SI: v_min_f32_e32
; SI: v_min_f32_e32
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT1:T[0-9]+]]
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT2:T[0-9]+]]
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT3:T[0-9]+]]
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT4:T[0-9]+]]
+; EG-DAG: MIN_DX10 {{.*}}[[OUT1]].X
+; EG-DAG: MIN_DX10 {{.*}}[[OUT1]].Y
+; EG-DAG: MIN_DX10 {{.*}}[[OUT1]].Z
+; EG-DAG: MIN_DX10 {{.*}}[[OUT1]].W
+; EG-DAG: MIN_DX10 {{.*}}[[OUT2]].X
+; EG-DAG: MIN_DX10 {{.*}}[[OUT2]].Y
+; EG-DAG: MIN_DX10 {{.*}}[[OUT2]].Z
+; EG-DAG: MIN_DX10 {{.*}}[[OUT2]].W
+; EG-DAG: MIN_DX10 {{.*}}[[OUT3]].X
+; EG-DAG: MIN_DX10 {{.*}}[[OUT3]].Y
+; EG-DAG: MIN_DX10 {{.*}}[[OUT3]].Z
+; EG-DAG: MIN_DX10 {{.*}}[[OUT3]].W
+; EG-DAG: MIN_DX10 {{.*}}[[OUT4]].X
+; EG-DAG: MIN_DX10 {{.*}}[[OUT4]].Y
+; EG-DAG: MIN_DX10 {{.*}}[[OUT4]].Z
+; EG-DAG: MIN_DX10 {{.*}}[[OUT4]].W
define void @test_fmin_v16f32(<16 x float> addrspace(1)* %out, <16 x float> %a, <16 x float> %b) nounwind {
%val = call <16 x float> @llvm.minnum.v16f32(<16 x float> %a, <16 x float> %b) #0
store <16 x float> %val, <16 x float> addrspace(1)* %out, align 64
@@ -77,6 +123,10 @@ define void @test_fmin_v16f32(<16 x float> addrspace(1)* %out, <16 x float> %a,
; SI-NOT: v_min_f32_e32
; SI: v_mov_b32_e32 [[REG:v[0-9]+]], 1.0
; SI: buffer_store_dword [[REG]]
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT:T[0-9]+\.[XYZW]]]
+; EG-NOT: MIN_DX10
+; EG: MOV {{.*}}[[OUT]], literal.{{[xy]}}
define void @constant_fold_fmin_f32(float addrspace(1)* %out) nounwind {
%val = call float @llvm.minnum.f32(float 1.0, float 2.0) #0
store float %val, float addrspace(1)* %out, align 4
@@ -87,6 +137,11 @@ define void @constant_fold_fmin_f32(float addrspace(1)* %out) nounwind {
; SI-NOT: v_min_f32_e32
; SI: v_mov_b32_e32 [[REG:v[0-9]+]], 0x7fc00000
; SI: buffer_store_dword [[REG]]
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT:T[0-9]+\.[XYZW]]]
+; EG-NOT: MIN_DX10
+; EG: MOV {{.*}}[[OUT]], literal.{{[xy]}}
+; EG: 2143289344(nan)
define void @constant_fold_fmin_f32_nan_nan(float addrspace(1)* %out) nounwind {
%val = call float @llvm.minnum.f32(float 0x7FF8000000000000, float 0x7FF8000000000000) #0
store float %val, float addrspace(1)* %out, align 4
@@ -97,6 +152,10 @@ define void @constant_fold_fmin_f32_nan_nan(float addrspace(1)* %out) nounwind {
; SI-NOT: v_min_f32_e32
; SI: v_mov_b32_e32 [[REG:v[0-9]+]], 1.0
; SI: buffer_store_dword [[REG]]
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT:T[0-9]+\.[XYZW]]]
+; EG-NOT: MIN_DX10
+; EG: MOV {{.*}}[[OUT]], literal.{{[xy]}}
define void @constant_fold_fmin_f32_val_nan(float addrspace(1)* %out) nounwind {
%val = call float @llvm.minnum.f32(float 1.0, float 0x7FF8000000000000) #0
store float %val, float addrspace(1)* %out, align 4
@@ -107,6 +166,10 @@ define void @constant_fold_fmin_f32_val_nan(float addrspace(1)* %out) nounwind {
; SI-NOT: v_min_f32_e32
; SI: v_mov_b32_e32 [[REG:v[0-9]+]], 1.0
; SI: buffer_store_dword [[REG]]
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT:T[0-9]+\.[XYZW]]]
+; EG-NOT: MIN_DX10
+; EG: MOV {{.*}}[[OUT]], literal.{{[xy]}}
define void @constant_fold_fmin_f32_nan_val(float addrspace(1)* %out) nounwind {
%val = call float @llvm.minnum.f32(float 0x7FF8000000000000, float 1.0) #0
store float %val, float addrspace(1)* %out, align 4
@@ -117,6 +180,10 @@ define void @constant_fold_fmin_f32_nan_val(float addrspace(1)* %out) nounwind {
; SI-NOT: v_min_f32_e32
; SI: v_mov_b32_e32 [[REG:v[0-9]+]], 0
; SI: buffer_store_dword [[REG]]
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT:T[0-9]+\.[XYZW]]]
+; EG-NOT: MIN_DX10
+; EG: MOV {{.*}}[[OUT]], literal.{{[xy]}}
define void @constant_fold_fmin_f32_p0_p0(float addrspace(1)* %out) nounwind {
%val = call float @llvm.minnum.f32(float 0.0, float 0.0) #0
store float %val, float addrspace(1)* %out, align 4
@@ -127,6 +194,10 @@ define void @constant_fold_fmin_f32_p0_p0(float addrspace(1)* %out) nounwind {
; SI-NOT: v_min_f32_e32
; SI: v_mov_b32_e32 [[REG:v[0-9]+]], 0
; SI: buffer_store_dword [[REG]]
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT:T[0-9]+\.[XYZW]]]
+; EG-NOT: MIN_DX10
+; EG: MOV {{.*}}[[OUT]], literal.{{[xy]}}
define void @constant_fold_fmin_f32_p0_n0(float addrspace(1)* %out) nounwind {
%val = call float @llvm.minnum.f32(float 0.0, float -0.0) #0
store float %val, float addrspace(1)* %out, align 4
@@ -137,6 +208,10 @@ define void @constant_fold_fmin_f32_p0_n0(float addrspace(1)* %out) nounwind {
; SI-NOT: v_min_f32_e32
; SI: v_mov_b32_e32 [[REG:v[0-9]+]], 0x80000000
; SI: buffer_store_dword [[REG]]
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT:T[0-9]+\.[XYZW]]]
+; EG-NOT: MIN_DX10
+; EG: MOV {{.*}}[[OUT]], literal.{{[xy]}}
define void @constant_fold_fmin_f32_n0_p0(float addrspace(1)* %out) nounwind {
%val = call float @llvm.minnum.f32(float -0.0, float 0.0) #0
store float %val, float addrspace(1)* %out, align 4
@@ -147,6 +222,10 @@ define void @constant_fold_fmin_f32_n0_p0(float addrspace(1)* %out) nounwind {
; SI-NOT: v_min_f32_e32
; SI: v_mov_b32_e32 [[REG:v[0-9]+]], 0x80000000
; SI: buffer_store_dword [[REG]]
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT:T[0-9]+\.[XYZW]]]
+; EG-NOT: MIN_DX10
+; EG: MOV {{.*}}[[OUT]], literal.{{[xy]}}
define void @constant_fold_fmin_f32_n0_n0(float addrspace(1)* %out) nounwind {
%val = call float @llvm.minnum.f32(float -0.0, float -0.0) #0
store float %val, float addrspace(1)* %out, align 4
@@ -155,6 +234,9 @@ define void @constant_fold_fmin_f32_n0_n0(float addrspace(1)* %out) nounwind {
; FUNC-LABEL: @fmin_var_immediate_f32
; SI: v_min_f32_e64 {{v[0-9]+}}, 2.0, {{s[0-9]+}}
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT:T[0-9]+\.[XYZW]]]
+; EG: MIN_DX10 {{.*}}[[OUT]], {{KC0\[[0-9]\].[XYZW]}}, literal.{{[xy]}}
define void @fmin_var_immediate_f32(float addrspace(1)* %out, float %a) nounwind {
%val = call float @llvm.minnum.f32(float %a, float 2.0) #0
store float %val, float addrspace(1)* %out, align 4
@@ -163,6 +245,9 @@ define void @fmin_var_immediate_f32(float addrspace(1)* %out, float %a) nounwind
; FUNC-LABEL: @fmin_immediate_var_f32
; SI: v_min_f32_e64 {{v[0-9]+}}, 2.0, {{s[0-9]+}}
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT:T[0-9]+\.[XYZW]]]
+; EG: MIN_DX10 {{.*}}[[OUT]], {{KC0\[[0-9]\].[XYZW]}}, literal.{{[xy]}}
define void @fmin_immediate_var_f32(float addrspace(1)* %out, float %a) nounwind {
%val = call float @llvm.minnum.f32(float 2.0, float %a) #0
store float %val, float addrspace(1)* %out, align 4
@@ -172,6 +257,9 @@ define void @fmin_immediate_var_f32(float addrspace(1)* %out, float %a) nounwind
; FUNC-LABEL: @fmin_var_literal_f32
; SI: v_mov_b32_e32 [[REG:v[0-9]+]], 0x42c60000
; SI: v_min_f32_e32 {{v[0-9]+}}, {{s[0-9]+}}, [[REG]]
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT:T[0-9]+\.[XYZW]]]
+; EG: MIN_DX10 {{.*}}[[OUT]], {{KC0\[[0-9]\].[XYZW]}}, literal.{{[xy]}}
define void @fmin_var_literal_f32(float addrspace(1)* %out, float %a) nounwind {
%val = call float @llvm.minnum.f32(float %a, float 99.0) #0
store float %val, float addrspace(1)* %out, align 4
@@ -181,6 +269,9 @@ define void @fmin_var_literal_f32(float addrspace(1)* %out, float %a) nounwind {
; FUNC-LABEL: @fmin_literal_var_f32
; SI: v_mov_b32_e32 [[REG:v[0-9]+]], 0x42c60000
; SI: v_min_f32_e32 {{v[0-9]+}}, {{s[0-9]+}}, [[REG]]
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[OUT:T[0-9]+\.[XYZW]]]
+; EG: MIN_DX10 {{.*}}[[OUT]], {{KC0\[[0-9]\].[XYZW]}}, literal.{{[xy]}}
define void @fmin_literal_var_f32(float addrspace(1)* %out, float %a) nounwind {
%val = call float @llvm.minnum.f32(float 99.0, float %a) #0
store float %val, float addrspace(1)* %out, align 4
--
2.1.0
More information about the llvm-commits
mailing list