Mark the SSE and MMX load instructions that

X86InstrInfo::isReallyTriviallyReMaterializable knows how to handle
with the isReMaterializable flag so that it is given a chance to handle
them. Without hoisting constant-pool loads from loops this isn't very
visible, though it does keep CodeGen/X86/constant-pool-remat-0.ll from
making a copy of the constant pool on the stack.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@40736 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86InstrMMX.td b/lib/Target/X86/X86InstrMMX.td
index bf34fbb01dada2e38e45c51bf1f95ad7008968ce..604eea4fd9e212e46f276fc0d48d558093f33918 100644 (file)
--- a/lib/Target/X86/X86InstrMMX.td
+++ b/lib/Target/X86/X86InstrMMX.td
@@ -157,6 +157,7 @@ def MMX_FEMMS : MMXI<0x0E, RawFrm, (outs), (ins), "femms", [(int_x86_mmx_femms)]
 // Data Transfer Instructions
 def MMX_MOVD64rr : MMXI<0x6E, MRMSrcReg, (outs VR64:$dst), (ins GR32:$src),
                         "movd\t{$src, $dst|$dst, $src}", []>;
+let isReMaterializable = 1 in
 def MMX_MOVD64rm : MMXI<0x6E, MRMSrcMem, (outs VR64:$dst), (ins i32mem:$src),
                         "movd\t{$src, $dst|$dst, $src}", []>;
 def MMX_MOVD64mr : MMXI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR64:$src),
@@ -167,6 +168,7 @@ def MMX_MOVD64to64rr : MMXRI<0x6E, MRMSrcReg, (outs VR64:$dst), (ins GR64:$src),
 
 def MMX_MOVQ64rr : MMXI<0x6F, MRMSrcReg, (outs VR64:$dst), (ins VR64:$src),
                         "movq\t{$src, $dst|$dst, $src}", []>;
+let isReMaterializable = 1 in
 def MMX_MOVQ64rm : MMXI<0x6F, MRMSrcMem, (outs VR64:$dst), (ins i64mem:$src),
                         "movq\t{$src, $dst|$dst, $src}",
                         [(set VR64:$dst, (load_mmx addr:$src))]>;

diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index d004232f2869bcac377daa8562d51bbc60d4e4bd..2773cac7c448b0043652ac2a7eae298132be8ae2 100644 (file)
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -280,6 +280,7 @@ let usesCustomDAGSchedInserter = 1 in {  // Expanded by the scheduler.
 // Move Instructions
 def MOVSSrr : SSI<0x10, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
                   "movss\t{$src, $dst|$dst, $src}", []>;
+let isReMaterializable = 1 in
 def MOVSSrm : SSI<0x10, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
                   "movss\t{$src, $dst|$dst, $src}",
                   [(set FR32:$dst, (loadf32 addr:$src))]>;
@@ -569,6 +570,7 @@ defm MIN : sse1_fp_binop_rm<0x5D, "min", X86fmin,
 // Move Instructions
 def MOVAPSrr : PSI<0x28, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                    "movaps\t{$src, $dst|$dst, $src}", []>;
+let isReMaterializable = 1 in
 def MOVAPSrm : PSI<0x28, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
                    "movaps\t{$src, $dst|$dst, $src}",
                    [(set VR128:$dst, (alignedloadv4f32 addr:$src))]>;
@@ -936,6 +938,7 @@ def MOVZSS2PSrm : SSI<0x10, MRMSrcMem, (outs VR128:$dst), (ins f32mem:$src),
 // Move Instructions
 def MOVSDrr : SDI<0x10, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
                   "movsd\t{$src, $dst|$dst, $src}", []>;
+let isReMaterializable = 1 in
 def MOVSDrm : SDI<0x10, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
                   "movsd\t{$src, $dst|$dst, $src}",
                   [(set FR64:$dst, (loadf64 addr:$src))]>;
@@ -1228,6 +1231,7 @@ defm MIN : sse2_fp_binop_rm<0x5D, "min", X86fmin,
 // Move Instructions
 def MOVAPDrr : PDI<0x28, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                    "movapd\t{$src, $dst|$dst, $src}", []>;
+let isReMaterializable = 1 in
 def MOVAPDrm : PDI<0x28, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
                    "movapd\t{$src, $dst|$dst, $src}",
                    [(set VR128:$dst, (alignedloadv2f64 addr:$src))]>;

diff --git a/test/CodeGen/X86/constant-pool-remat-0.ll b/test/CodeGen/X86/constant-pool-remat-0.ll
new file mode 100644 (file)
index 0000000..d5af99e
--- /dev/null
+++ b/test/CodeGen/X86/constant-pool-remat-0.ll
@@ -0,0 +1,10 @@
+; RUN: llvm-as < %s | llc -march=x86-64 | grep LCPI | wc -l | grep 3
+
+declare float @qux(float %y)
+
+define float @array(float %a) {
+  %n = mul float %a, 9.0
+  %m = call float @qux(float %n)
+  %o = mul float %m, 9.0
+  ret float %o
+}