[PATCH] D75392: [AMDGPU] Fix the gfx10 scheduling model for f32 conversions

[Jay Foad via Phabricator via llvm-commits]

This revision was automatically updated to reflect the committed changes.
foad marked an inline comment as done.
Closed by commit rGc8f0d27ef37c: [AMDGPU] Fix the gfx10 scheduling model for f32 conversions (authored by foad).

Changed prior to commit:
  https://reviews.llvm.org/D75392?vs=247385&id=249475#toc

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D75392/new/

https://reviews.llvm.org/D75392

Files:
  llvm/lib/Target/AMDGPU/SISchedule.td
  llvm/lib/Target/AMDGPU/VOP1Instructions.td


Index: llvm/lib/Target/AMDGPU/VOP1Instructions.td
===================================================================
--- llvm/lib/Target/AMDGPU/VOP1Instructions.td
+++ llvm/lib/Target/AMDGPU/VOP1Instructions.td
@@ -190,7 +190,7 @@
 defm V_CVT_F64_U32 : VOP1Inst <"v_cvt_f64_u32", VOP1_F64_I32, uint_to_fp>;
 } // End SchedRW = [WriteDoubleCvt]
 
-let SchedRW = [WriteQuarterRate32] in {
+let SchedRW = [WriteFloatCvt] in {
 defm V_CVT_F32_I32 : VOP1Inst <"v_cvt_f32_i32", VOP1_F32_I32, sint_to_fp>;
 defm V_CVT_F32_U32 : VOP1Inst <"v_cvt_f32_u32", VOP1_F32_I32, uint_to_fp>;
 defm V_CVT_U32_F32 : VOP1Inst <"v_cvt_u32_f32", VOP_I32_F32, fp_to_uint>;
@@ -202,7 +202,7 @@
 defm V_CVT_RPI_I32_F32 : VOP1Inst <"v_cvt_rpi_i32_f32", VOP_I32_F32, cvt_rpi_i32_f32>;
 defm V_CVT_FLR_I32_F32 : VOP1Inst <"v_cvt_flr_i32_f32", VOP_I32_F32, cvt_flr_i32_f32>;
 defm V_CVT_OFF_F32_I4 : VOP1Inst  <"v_cvt_off_f32_i4", VOP1_F32_I32>;
-} // End SchedRW = [WriteQuarterRate32]
+} // End SchedRW = [WriteFloatCvt]
 
 defm V_CVT_F32_UBYTE0 : VOP1Inst <"v_cvt_f32_ubyte0", VOP1_F32_I32, AMDGPUcvt_f32_ubyte0>;
 defm V_CVT_F32_UBYTE1 : VOP1Inst <"v_cvt_f32_ubyte1", VOP1_F32_I32, AMDGPUcvt_f32_ubyte1>;
Index: llvm/lib/Target/AMDGPU/SISchedule.td
===================================================================
--- llvm/lib/Target/AMDGPU/SISchedule.td
+++ llvm/lib/Target/AMDGPU/SISchedule.td
@@ -29,6 +29,7 @@
 
 // Vector ALU instructions
 def Write32Bit         : SchedWrite;
+def WriteFloatCvt      : SchedWrite;
 def WriteQuarterRate32 : SchedWrite;
 
 def WriteFloatFMA   : SchedWrite;
@@ -126,6 +127,7 @@
 
   def : HWVALUWriteRes<Write32Bit,         1>;
   def : HWVALUWriteRes<Write64Bit,         2>;
+  def : HWVALUWriteRes<WriteFloatCvt,      4>;
   def : HWVALUWriteRes<WriteQuarterRate32, 4>;
   def : HWVALUWriteRes<Write2PassMAI,      2>;
   def : HWVALUWriteRes<Write8PassMAI,      8>;
@@ -185,6 +187,7 @@
 // The latency values are 1 / (operations / cycle).
 // Add 1 stall cycle for VGPR read.
 def : HWWriteRes<Write32Bit,         [HWVALU, HWRC],   5>;
+def : HWWriteRes<WriteFloatCvt,      [HWVALU, HWRC],   5>;
 def : HWWriteRes<Write64Bit,         [HWVALU, HWRC],   9>;
 def : HWWriteRes<WriteQuarterRate32, [HWVALU, HWRC],   17>;
 def : HWWriteRes<WriteFloatFMA,      [HWVALU, HWRC],   5>;


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D75392.249475.patch
Type: text/x-patch
Size: 2295 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20200310/264a7375/attachment.bin>