aaron/ispc
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Workaround change to linker behavior in LLVM 3.1

Browse Source 
Now, the Linker::LinkModules() call doesn't link in any functions
marked as 'internal', which is problematic, since we'd like to have
just about all of the builtins marked as internal so that they are
eliminated after they've been inlined when they are in fact used.

This change removes all of the internal qualifiers in the builtins
and adds a lSetInternalFunctions() routine to builtins.cpp that
sets this property on the functions that need it after they've
been linked in by LinkModules().
This commit is contained in:
Matt Pharr
2011-11-05 16:57:26 -07:00
parent b0d476fcdc
commit cabe358c0a
13 changed files with 760 additions and 447 deletions
Download Patch File Download Diff File Expand all files Collapse all files

42
builtins-sse4-common.ll
View File

@@ -34,7 +34,7 @@
declare <4 x float> @llvm.x86.sse41.round.ss(<4 x float>, <4 x float>, i32) nounwind readnone
define internal float @__round_uniform_float(float) nounwind readonly alwaysinline {
define float @__round_uniform_float(float) nounwind readonly alwaysinline {
; roundss, round mode nearest 0b00 | don't signal precision exceptions 0b1000 = 8
; the roundss intrinsic is a total mess--docs say:
;
@@ -58,7 +58,7 @@ define internal float @__round_uniform_float(float) nounwind readonly alwaysinli
ret float %rs
}
define internal float @__floor_uniform_float(float) nounwind readonly alwaysinline {
define float @__floor_uniform_float(float) nounwind readonly alwaysinline {
; see above for round_ss instrinsic discussion...
%xi = insertelement <4 x float> undef, float %0, i32 0
; roundps, round down 0b01 | don't signal precision exceptions 0b1010 = 9
@@ -67,7 +67,7 @@ define internal float @__floor_uniform_float(float) nounwind readonly alwaysinli
ret float %rs
}
define internal float @__ceil_uniform_float(float) nounwind readonly alwaysinline {
define float @__ceil_uniform_float(float) nounwind readonly alwaysinline {
; see above for round_ss instrinsic discussion...
%xi = insertelement <4 x float> undef, float %0, i32 0
; roundps, round up 0b10 | don't signal precision exceptions 0b1010 = 10
@@ -81,14 +81,14 @@ define internal float @__ceil_uniform_float(float) nounwind readonly alwaysinlin
declare <2 x double> @llvm.x86.sse41.round.sd(<2 x double>, <2 x double>, i32) nounwind readnone
define internal double @__round_uniform_double(double) nounwind readonly alwaysinline {
define double @__round_uniform_double(double) nounwind readonly alwaysinline {
%xi = insertelement <2 x double> undef, double %0, i32 0
%xr = call <2 x double> @llvm.x86.sse41.round.sd(<2 x double> %xi, <2 x double> %xi, i32 8)
%rs = extractelement <2 x double> %xr, i32 0
ret double %rs
}
define internal double @__floor_uniform_double(double) nounwind readonly alwaysinline {
define double @__floor_uniform_double(double) nounwind readonly alwaysinline {
; see above for round_ss instrinsic discussion...
%xi = insertelement <2 x double> undef, double %0, i32 0
; roundpd, round down 0b01 | don't signal precision exceptions 0b1001 = 9
@@ -97,7 +97,7 @@ define internal double @__floor_uniform_double(double) nounwind readonly alwaysi
ret double %rs
}
define internal double @__ceil_uniform_double(double) nounwind readonly alwaysinline {
define double @__ceil_uniform_double(double) nounwind readonly alwaysinline {
; see above for round_ss instrinsic discussion...
%xi = insertelement <2 x double> undef, double %0, i32 0
; roundps, round up 0b10 | don't signal precision exceptions 0b1010 = 10
@@ -111,7 +111,7 @@ define internal double @__ceil_uniform_double(double) nounwind readonly alwaysin
declare <4 x float> @llvm.x86.sse.rcp.ss(<4 x float>) nounwind readnone
define internal float @__rcp_uniform_float(float) nounwind readonly alwaysinline {
define float @__rcp_uniform_float(float) nounwind readonly alwaysinline {
; do the rcpss call
%vecval = insertelement <4 x float> undef, float %0, i32 0
%call = call <4 x float> @llvm.x86.sse.rcp.ss(<4 x float> %vecval)
@@ -130,7 +130,7 @@ define internal float @__rcp_uniform_float(float) nounwind readonly alwaysinline
declare <4 x float> @llvm.x86.sse.rsqrt.ss(<4 x float>) nounwind readnone
define internal float @__rsqrt_uniform_float(float) nounwind readonly alwaysinline {
define float @__rsqrt_uniform_float(float) nounwind readonly alwaysinline {
; uniform float is = extract(__rsqrt_u(v), 0);
%v = insertelement <4 x float> undef, float %0, i32 0
%vis = call <4 x float> @llvm.x86.sse.rsqrt.ss(<4 x float> %v)
@@ -152,7 +152,7 @@ define internal float @__rsqrt_uniform_float(float) nounwind readonly alwaysinli
declare <4 x float> @llvm.x86.sse.sqrt.ss(<4 x float>) nounwind readnone
define internal float @__sqrt_uniform_float(float) nounwind readonly alwaysinline {
define float @__sqrt_uniform_float(float) nounwind readonly alwaysinline {
sse_unary_scalar(ret, 4, float, @llvm.x86.sse.sqrt.ss, %0)
ret float %ret
}
@@ -163,7 +163,7 @@ define internal float @__sqrt_uniform_float(float) nounwind readonly alwaysinlin
declare void @llvm.x86.sse.stmxcsr(i8 *) nounwind
declare void @llvm.x86.sse.ldmxcsr(i8 *) nounwind
define internal void @__fastmath() nounwind alwaysinline {
define void @__fastmath() nounwind alwaysinline {
%ptr = alloca i32
%ptr8 = bitcast i32 * %ptr to i8 *
call void @llvm.x86.sse.stmxcsr(i8 * %ptr8)
@@ -182,12 +182,12 @@ define internal void @__fastmath() nounwind alwaysinline {
declare <4 x float> @llvm.x86.sse.max.ss(<4 x float>, <4 x float>) nounwind readnone
declare <4 x float> @llvm.x86.sse.min.ss(<4 x float>, <4 x float>) nounwind readnone
define internal float @__max_uniform_float(float, float) nounwind readonly alwaysinline {
define float @__max_uniform_float(float, float) nounwind readonly alwaysinline {
sse_binary_scalar(ret, 4, float, @llvm.x86.sse.max.ss, %0, %1)
ret float %ret
}
define internal float @__min_uniform_float(float, float) nounwind readonly alwaysinline {
define float @__min_uniform_float(float, float) nounwind readonly alwaysinline {
sse_binary_scalar(ret, 4, float, @llvm.x86.sse.min.ss, %0, %1)
ret float %ret
}
@@ -197,7 +197,7 @@ define internal float @__min_uniform_float(float, float) nounwind readonly alway
declare <2 x double> @llvm.x86.sse2.sqrt.sd(<2 x double>) nounwind readnone
define internal double @__sqrt_uniform_double(double) nounwind alwaysinline {
define double @__sqrt_uniform_double(double) nounwind alwaysinline {
sse_unary_scalar(ret, 2, double, @llvm.x86.sse2.sqrt.sd, %0)
ret double %ret
}
@@ -209,13 +209,13 @@ define internal double @__sqrt_uniform_double(double) nounwind alwaysinline {
declare <2 x double> @llvm.x86.sse2.max.sd(<2 x double>, <2 x double>) nounwind readnone
declare <2 x double> @llvm.x86.sse2.min.sd(<2 x double>, <2 x double>) nounwind readnone
define internal double @__min_uniform_double(double, double) nounwind readnone {
define double @__min_uniform_double(double, double) nounwind readnone {
sse_binary_scalar(ret, 2, double, @llvm.x86.sse2.min.sd, %0, %1)
ret double %ret
}
define internal double @__max_uniform_double(double, double) nounwind readnone {
define double @__max_uniform_double(double, double) nounwind readnone {
sse_binary_scalar(ret, 2, double, @llvm.x86.sse2.max.sd, %0, %1)
ret double %ret
}
@@ -226,12 +226,12 @@ define internal double @__max_uniform_double(double, double) nounwind readnone {
declare <4 x i32> @llvm.x86.sse41.pminsd(<4 x i32>, <4 x i32>) nounwind readnone
declare <4 x i32> @llvm.x86.sse41.pmaxsd(<4 x i32>, <4 x i32>) nounwind readnone
define internal i32 @__min_uniform_int32(i32, i32) nounwind readonly alwaysinline {
define i32 @__min_uniform_int32(i32, i32) nounwind readonly alwaysinline {
sse_binary_scalar(ret, 4, i32, @llvm.x86.sse41.pminsd, %0, %1)
ret i32 %ret
}
define internal i32 @__max_uniform_int32(i32, i32) nounwind readonly alwaysinline {
define i32 @__max_uniform_int32(i32, i32) nounwind readonly alwaysinline {
sse_binary_scalar(ret, 4, i32, @llvm.x86.sse41.pmaxsd, %0, %1)
ret i32 %ret
}
@@ -242,12 +242,12 @@ define internal i32 @__max_uniform_int32(i32, i32) nounwind readonly alwaysinlin
declare <4 x i32> @llvm.x86.sse41.pminud(<4 x i32>, <4 x i32>) nounwind readnone
declare <4 x i32> @llvm.x86.sse41.pmaxud(<4 x i32>, <4 x i32>) nounwind readnone
define internal i32 @__min_uniform_uint32(i32, i32) nounwind readonly alwaysinline {
define i32 @__min_uniform_uint32(i32, i32) nounwind readonly alwaysinline {
sse_binary_scalar(ret, 4, i32, @llvm.x86.sse41.pminud, %0, %1)
ret i32 %ret
}
define internal i32 @__max_uniform_uint32(i32, i32) nounwind readonly alwaysinline {
define i32 @__max_uniform_uint32(i32, i32) nounwind readonly alwaysinline {
sse_binary_scalar(ret, 4, i32, @llvm.x86.sse41.pmaxud, %0, %1)
ret i32 %ret
}
@@ -258,14 +258,14 @@ define internal i32 @__max_uniform_uint32(i32, i32) nounwind readonly alwaysinli
declare i32 @llvm.ctpop.i32(i32) nounwind readnone
define internal i32 @__popcnt_int32(i32) nounwind readonly alwaysinline {
define i32 @__popcnt_int32(i32) nounwind readonly alwaysinline {
%call = call i32 @llvm.ctpop.i32(i32 %0)
ret i32 %call
}
declare i64 @llvm.ctpop.i64(i64) nounwind readnone
define internal i64 @__popcnt_int64(i64) nounwind readonly alwaysinline {
define i64 @__popcnt_int64(i64) nounwind readonly alwaysinline {
%call = call i64 @llvm.ctpop.i64(i64 %0)
ret i64 %call
}