aboutsummaryrefslogtreecommitdiff
path: root/lib/Target/R600/R600Instructions.td
diff options
context:
space:
mode:
authorTom Stellard <thomas.stellard@amd.com>2013-02-07 14:02:35 +0000
committerTom Stellard <thomas.stellard@amd.com>2013-02-07 14:02:35 +0000
commit1234c9be42b4ebd4b398df461123205dccf3706c (patch)
treed328469268473b89e100f52577a0dbfe6f244eb5 /lib/Target/R600/R600Instructions.td
parent2a77cf7f47900c8df55cfea794936006b494e943 (diff)
R600: Add support for SET*_DX10 instructions
These instructions compare two floating point values and return an integer true (-1) or false (0) value. When compiling code generated by the Mesa GLSL frontend, the SET*_DX10 instructions save us four instructions for most branch decisions that use floating-point comparisons. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@174609 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'lib/Target/R600/R600Instructions.td')
-rw-r--r--lib/Target/R600/R600Instructions.td52
1 files changed, 52 insertions, 0 deletions
diff --git a/lib/Target/R600/R600Instructions.td b/lib/Target/R600/R600Instructions.td
index d40e75bf9e..06196e72fa 100644
--- a/lib/Target/R600/R600Instructions.td
+++ b/lib/Target/R600/R600Instructions.td
@@ -711,6 +711,34 @@ def SNE : R600_2OP <
COND_NE))]
>;
+def SETE_DX10 : R600_2OP <
+ 0xC, "SETE_DX10",
+ [(set R600_Reg32:$dst,
+ (selectcc (f32 R600_Reg32:$src0), R600_Reg32:$src1, (i32 -1), (i32 0),
+ COND_EQ))]
+>;
+
+def SETGT_DX10 : R600_2OP <
+ 0xD, "SETGT_DX10",
+ [(set R600_Reg32:$dst,
+ (selectcc (f32 R600_Reg32:$src0), R600_Reg32:$src1, (i32 -1), (i32 0),
+ COND_GT))]
+>;
+
+def SETGE_DX10 : R600_2OP <
+ 0xE, "SETGE_DX10",
+ [(set R600_Reg32:$dst,
+ (selectcc (f32 R600_Reg32:$src0), R600_Reg32:$src1, (i32 -1), (i32 0),
+ COND_GE))]
+>;
+
+def SETNE_DX10 : R600_2OP <
+ 0xF, "SETNE_DX10",
+ [(set R600_Reg32:$dst,
+ (selectcc (f32 R600_Reg32:$src0), R600_Reg32:$src1, (i32 -1), (i32 0),
+ COND_NE))]
+>;
+
def FRACT : R600_1OP_Helper <0x10, "FRACT", AMDGPUfract>;
def TRUNC : R600_1OP_Helper <0x11, "TRUNC", int_AMDGPU_trunc>;
def CEIL : R600_1OP_Helper <0x12, "CEIL", fceil>;
@@ -1772,6 +1800,18 @@ def : Pat <
(SGE R600_Reg32:$src1, R600_Reg32:$src0)
>;
+// SETGT_DX10 reverse args
+def : Pat <
+ (selectcc (f32 R600_Reg32:$src0), R600_Reg32:$src1, -1, 0, COND_LT),
+ (SETGT_DX10 R600_Reg32:$src1, R600_Reg32:$src0)
+>;
+
+// SETGE_DX10 reverse args
+def : Pat <
+ (selectcc (f32 R600_Reg32:$src0), R600_Reg32:$src1, -1, 0, COND_LE),
+ (SETGE_DX10 R600_Reg32:$src1, R600_Reg32:$src0)
+>;
+
// SETGT_INT reverse args
def : Pat <
(selectcc (i32 R600_Reg32:$src0), R600_Reg32:$src1, -1, 0, SETLT),
@@ -1810,12 +1850,24 @@ def : Pat <
(SETE R600_Reg32:$src0, R600_Reg32:$src1)
>;
+//SETE_DX10 - 'true if ordered'
+def : Pat <
+ (selectcc (f32 R600_Reg32:$src0), R600_Reg32:$src1, -1, 0, SETO),
+ (SETE_DX10 R600_Reg32:$src0, R600_Reg32:$src1)
+>;
+
//SNE - 'true if unordered'
def : Pat <
(selectcc (f32 R600_Reg32:$src0), R600_Reg32:$src1, FP_ONE, FP_ZERO, SETUO),
(SNE R600_Reg32:$src0, R600_Reg32:$src1)
>;
+//SETNE_DX10 - 'true if ordered'
+def : Pat <
+ (selectcc (f32 R600_Reg32:$src0), R600_Reg32:$src1, -1, 0, SETUO),
+ (SETNE_DX10 R600_Reg32:$src0, R600_Reg32:$src1)
+>;
+
def : Extract_Element <f32, v4f32, R600_Reg128, 0, sel_x>;
def : Extract_Element <f32, v4f32, R600_Reg128, 1, sel_y>;
def : Extract_Element <f32, v4f32, R600_Reg128, 2, sel_z>;