RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-23 11:13:28 +01:00
Code Issues Projects Releases Wiki Activity
179c563da5
llvm-mirror/test/CodeGen/PowerPC/scalar_vector_test_1.ll

275 lines
8.3 KiB
LLVM
Raw Normal View History

[PowerPC] Improve codegen for vector loads using scalar_to_vector This patch aims to improve the codegen for vector loads involving the scalar_to_vector (load X) sequence. Initially, ld->mv instructions were used for scalar_to_vector (load X), so this patch allows scalar_to_vector (load X) to utilize: LXSD and LXSDX for i64 and f64 LXSIWAX for i32 (sign extension to i64) LXSIWZX for i32 and f64 Committing on behalf of Amy Kwan. Differential Revision: https://reviews.llvm.org/D48950 llvm-svn: 339260
2018-08-08 17:20:43 +02:00
; RUN: llc -mcpu=pwr9 -verify-machineinstrs -ppc-vsr-nums-as-vr -ppc-asm-full-reg-names \
; RUN: -mtriple=powerpc64le-unknown-linux-gnu < %s | FileCheck %s --check-prefix=P9LE
; RUN: llc -mcpu=pwr9 -verify-machineinstrs -ppc-vsr-nums-as-vr -ppc-asm-full-reg-names \
; RUN: -mtriple=powerpc64-unknown-linux-gnu < %s | FileCheck %s --check-prefix=P9BE
; RUN: llc -mcpu=pwr8 -verify-machineinstrs -ppc-vsr-nums-as-vr -ppc-asm-full-reg-names \
; RUN: -mtriple=powerpc64le-unknown-linux-gnu < %s | FileCheck %s --check-prefix=P8LE
; RUN: llc -mcpu=pwr8 -verify-machineinstrs -ppc-vsr-nums-as-vr -ppc-asm-full-reg-names \
; RUN: -mtriple=powerpc64-unknown-linux-gnu < %s | FileCheck %s --check-prefix=P8BE
Test commit llvm-svn: 349633
2018-12-19 16:21:07 +01:00
[PowerPC] Improve codegen for vector loads using scalar_to_vector This patch aims to improve the codegen for vector loads involving the scalar_to_vector (load X) sequence. Initially, ld->mv instructions were used for scalar_to_vector (load X), so this patch allows scalar_to_vector (load X) to utilize: LXSD and LXSDX for i64 and f64 LXSIWAX for i32 (sign extension to i64) LXSIWZX for i32 and f64 Committing on behalf of Amy Kwan. Differential Revision: https://reviews.llvm.org/D48950 llvm-svn: 339260
2018-08-08 17:20:43 +02:00
; Function Attrs: norecurse nounwind readonly
define <2 x i64> @s2v_test1(i64* nocapture readonly %int64, <2 x i64> %vec) {
; P9LE-LABEL: s2v_test1:
; P9LE: # %bb.0: # %entry
; P9LE-NEXT: lfd f0, 0(r3)
[PowerPC] Canonicalize shuffles to match more single-instruction masks on LE We currently miss a number of opportunities to emit single-instruction VMRG[LH][BHW] instructions for shuffles on little endian subtargets. Although this in itself is not a huge performance opportunity since loading the permute vector for a VPERM can always be pulled out of loops, producing such merge instructions is useful to downstream optimizations. Since VPERM is essentially opaque to all subsequent optimizations, we want to avoid it as much as possible. Other permute instructions have semantics that can be reasoned about much more easily in later optimizations. This patch does the following: - Canonicalize shuffles so that the first element comes from the first vector (since that's what most of the mask matching functions want) - Switch the elements that come from splat vectors so that they match the corresponding elements from the other vector (to allow for merges) - Adds debugging messages for when a shuffle is matched to a VPERM so that anyone interested in improving this further can get the info for their code Differential revision: https://reviews.llvm.org/D77448
2020-06-19 04:53:50 +02:00
; P9LE-NEXT: xxmrghd v2, v2, vs0
[PowerPC] Improve codegen for vector loads using scalar_to_vector This patch aims to improve the codegen for vector loads involving the scalar_to_vector (load X) sequence. Initially, ld->mv instructions were used for scalar_to_vector (load X), so this patch allows scalar_to_vector (load X) to utilize: LXSD and LXSDX for i64 and f64 LXSIWAX for i32 (sign extension to i64) LXSIWZX for i32 and f64 Committing on behalf of Amy Kwan. Differential Revision: https://reviews.llvm.org/D48950 llvm-svn: 339260
2018-08-08 17:20:43 +02:00
; P9LE-NEXT: blr
; P9BE-LABEL: s2v_test1:
; P9BE: # %bb.0: # %entry
; P9BE-NEXT: lfd f0, 0(r3)
; P9BE-NEXT: xxpermdi v2, vs0, v2, 1
; P9BE-NEXT: blr
entry:
%0 = load i64, i64* %int64, align 8
%vecins = insertelement <2 x i64> %vec, i64 %0, i32 0
ret <2 x i64> %vecins
}
; Function Attrs: norecurse nounwind readonly
define <2 x i64> @s2v_test2(i64* nocapture readonly %int64, <2 x i64> %vec) {
; P9LE-LABEL: s2v_test2:
; P9LE: # %bb.0: # %entry
; P9LE-NEXT: lfd f0, 8(r3)
[PowerPC] Canonicalize shuffles to match more single-instruction masks on LE We currently miss a number of opportunities to emit single-instruction VMRG[LH][BHW] instructions for shuffles on little endian subtargets. Although this in itself is not a huge performance opportunity since loading the permute vector for a VPERM can always be pulled out of loops, producing such merge instructions is useful to downstream optimizations. Since VPERM is essentially opaque to all subsequent optimizations, we want to avoid it as much as possible. Other permute instructions have semantics that can be reasoned about much more easily in later optimizations. This patch does the following: - Canonicalize shuffles so that the first element comes from the first vector (since that's what most of the mask matching functions want) - Switch the elements that come from splat vectors so that they match the corresponding elements from the other vector (to allow for merges) - Adds debugging messages for when a shuffle is matched to a VPERM so that anyone interested in improving this further can get the info for their code Differential revision: https://reviews.llvm.org/D77448
2020-06-19 04:53:50 +02:00
; P9LE-NEXT: xxmrghd v2, v2, vs0
[PowerPC] Improve codegen for vector loads using scalar_to_vector This patch aims to improve the codegen for vector loads involving the scalar_to_vector (load X) sequence. Initially, ld->mv instructions were used for scalar_to_vector (load X), so this patch allows scalar_to_vector (load X) to utilize: LXSD and LXSDX for i64 and f64 LXSIWAX for i32 (sign extension to i64) LXSIWZX for i32 and f64 Committing on behalf of Amy Kwan. Differential Revision: https://reviews.llvm.org/D48950 llvm-svn: 339260
2018-08-08 17:20:43 +02:00
; P9LE-NEXT: blr
; P9BE-LABEL: s2v_test2:
; P9BE: # %bb.0: # %entry
; P9BE-NEXT: lfd f0, 8(r3)
; P9BE-NEXT: xxpermdi v2, vs0, v2, 1
; P9BE-NEXT: blr
entry:
%arrayidx = getelementptr inbounds i64, i64* %int64, i64 1
%0 = load i64, i64* %arrayidx, align 8
%vecins = insertelement <2 x i64> %vec, i64 %0, i32 0
ret <2 x i64> %vecins
}
; Function Attrs: norecurse nounwind readonly
define <2 x i64> @s2v_test3(i64* nocapture readonly %int64, <2 x i64> %vec, i32 signext %Idx) {
; P9LE-LABEL: s2v_test3:
; P9LE: # %bb.0: # %entry
; P9LE-NEXT: sldi r4, r7, 3
; P9LE-NEXT: lfdx f0, r3, r4
[PowerPC] Canonicalize shuffles to match more single-instruction masks on LE We currently miss a number of opportunities to emit single-instruction VMRG[LH][BHW] instructions for shuffles on little endian subtargets. Although this in itself is not a huge performance opportunity since loading the permute vector for a VPERM can always be pulled out of loops, producing such merge instructions is useful to downstream optimizations. Since VPERM is essentially opaque to all subsequent optimizations, we want to avoid it as much as possible. Other permute instructions have semantics that can be reasoned about much more easily in later optimizations. This patch does the following: - Canonicalize shuffles so that the first element comes from the first vector (since that's what most of the mask matching functions want) - Switch the elements that come from splat vectors so that they match the corresponding elements from the other vector (to allow for merges) - Adds debugging messages for when a shuffle is matched to a VPERM so that anyone interested in improving this further can get the info for their code Differential revision: https://reviews.llvm.org/D77448
2020-06-19 04:53:50 +02:00
; P9LE-NEXT: xxmrghd v2, v2, vs0
[PowerPC] Improve codegen for vector loads using scalar_to_vector This patch aims to improve the codegen for vector loads involving the scalar_to_vector (load X) sequence. Initially, ld->mv instructions were used for scalar_to_vector (load X), so this patch allows scalar_to_vector (load X) to utilize: LXSD and LXSDX for i64 and f64 LXSIWAX for i32 (sign extension to i64) LXSIWZX for i32 and f64 Committing on behalf of Amy Kwan. Differential Revision: https://reviews.llvm.org/D48950 llvm-svn: 339260
2018-08-08 17:20:43 +02:00
; P9LE-NEXT: blr
; P9BE-LABEL: s2v_test3
; P9BE: # %bb.0: # %entry
; P9BE-NEXT: sldi r4, r7, 3
; P9BE-NEXT: lfdx f0, r3, r4
; P9BE-NEXT: xxpermdi v2, vs0, v2, 1
; P9BE-NEXT: blr
entry:
%idxprom = sext i32 %Idx to i64
%arrayidx = getelementptr inbounds i64, i64* %int64, i64 %idxprom
%0 = load i64, i64* %arrayidx, align 8
%vecins = insertelement <2 x i64> %vec, i64 %0, i32 0
ret <2 x i64> %vecins
}
; Function Attrs: norecurse nounwind readonly
define <2 x i64> @s2v_test4(i64* nocapture readonly %int64, <2 x i64> %vec) {
; P9LE-LABEL: s2v_test4:
; P9LE: # %bb.0: # %entry
; P9LE-NEXT: lfd f0, 8(r3)
[PowerPC] Canonicalize shuffles to match more single-instruction masks on LE We currently miss a number of opportunities to emit single-instruction VMRG[LH][BHW] instructions for shuffles on little endian subtargets. Although this in itself is not a huge performance opportunity since loading the permute vector for a VPERM can always be pulled out of loops, producing such merge instructions is useful to downstream optimizations. Since VPERM is essentially opaque to all subsequent optimizations, we want to avoid it as much as possible. Other permute instructions have semantics that can be reasoned about much more easily in later optimizations. This patch does the following: - Canonicalize shuffles so that the first element comes from the first vector (since that's what most of the mask matching functions want) - Switch the elements that come from splat vectors so that they match the corresponding elements from the other vector (to allow for merges) - Adds debugging messages for when a shuffle is matched to a VPERM so that anyone interested in improving this further can get the info for their code Differential revision: https://reviews.llvm.org/D77448
2020-06-19 04:53:50 +02:00
; P9LE-NEXT: xxmrghd v2, v2, vs0
[PowerPC] Improve codegen for vector loads using scalar_to_vector This patch aims to improve the codegen for vector loads involving the scalar_to_vector (load X) sequence. Initially, ld->mv instructions were used for scalar_to_vector (load X), so this patch allows scalar_to_vector (load X) to utilize: LXSD and LXSDX for i64 and f64 LXSIWAX for i32 (sign extension to i64) LXSIWZX for i32 and f64 Committing on behalf of Amy Kwan. Differential Revision: https://reviews.llvm.org/D48950 llvm-svn: 339260
2018-08-08 17:20:43 +02:00
; P9LE-NEXT: blr
; P9BE-LABEL: s2v_test4:
; P9BE: # %bb.0: # %entry
; P9BE-NEXT: lfd f0, 8(r3)
; P9BE-NEXT: xxpermdi v2, vs0, v2, 1
; P9BE-NEXT: blr
entry:
%arrayidx = getelementptr inbounds i64, i64* %int64, i64 1
%0 = load i64, i64* %arrayidx, align 8
%vecins = insertelement <2 x i64> %vec, i64 %0, i32 0
ret <2 x i64> %vecins
}
; Function Attrs: norecurse nounwind readonly
define <2 x i64> @s2v_test5(<2 x i64> %vec, i64* nocapture readonly %ptr1) {
; P9LE-LABEL: s2v_test5:
; P9LE: # %bb.0: # %entry
; P9LE-NEXT: lfd f0, 0(r5)
[PowerPC] Canonicalize shuffles to match more single-instruction masks on LE We currently miss a number of opportunities to emit single-instruction VMRG[LH][BHW] instructions for shuffles on little endian subtargets. Although this in itself is not a huge performance opportunity since loading the permute vector for a VPERM can always be pulled out of loops, producing such merge instructions is useful to downstream optimizations. Since VPERM is essentially opaque to all subsequent optimizations, we want to avoid it as much as possible. Other permute instructions have semantics that can be reasoned about much more easily in later optimizations. This patch does the following: - Canonicalize shuffles so that the first element comes from the first vector (since that's what most of the mask matching functions want) - Switch the elements that come from splat vectors so that they match the corresponding elements from the other vector (to allow for merges) - Adds debugging messages for when a shuffle is matched to a VPERM so that anyone interested in improving this further can get the info for their code Differential revision: https://reviews.llvm.org/D77448
2020-06-19 04:53:50 +02:00
; P9LE-NEXT: xxmrghd v2, v2, vs0
[PowerPC] Improve codegen for vector loads using scalar_to_vector This patch aims to improve the codegen for vector loads involving the scalar_to_vector (load X) sequence. Initially, ld->mv instructions were used for scalar_to_vector (load X), so this patch allows scalar_to_vector (load X) to utilize: LXSD and LXSDX for i64 and f64 LXSIWAX for i32 (sign extension to i64) LXSIWZX for i32 and f64 Committing on behalf of Amy Kwan. Differential Revision: https://reviews.llvm.org/D48950 llvm-svn: 339260
2018-08-08 17:20:43 +02:00
; P9LE-NEXT: blr
; P9BE-LABEL: s2v_test5:
; P9BE: # %bb.0: # %entry
; P9BE-NEXT: lfd f0, 0(r5)
; P9BE-NEXT: xxpermdi v2, vs0, v2, 1
; P9BE-NEXT: blr
entry:
%0 = load i64, i64* %ptr1, align 8
%vecins = insertelement <2 x i64> %vec, i64 %0, i32 0
ret <2 x i64> %vecins
}
; Function Attrs: norecurse nounwind readonly
define <2 x double> @s2v_test_f1(double* nocapture readonly %f64, <2 x double> %vec) {
; P9LE-LABEL: s2v_test_f1:
; P9LE: # %bb.0: # %entry
; P9LE-NEXT: lfd f0, 0(r3)
[PowerPC] Canonicalize shuffles to match more single-instruction masks on LE We currently miss a number of opportunities to emit single-instruction VMRG[LH][BHW] instructions for shuffles on little endian subtargets. Although this in itself is not a huge performance opportunity since loading the permute vector for a VPERM can always be pulled out of loops, producing such merge instructions is useful to downstream optimizations. Since VPERM is essentially opaque to all subsequent optimizations, we want to avoid it as much as possible. Other permute instructions have semantics that can be reasoned about much more easily in later optimizations. This patch does the following: - Canonicalize shuffles so that the first element comes from the first vector (since that's what most of the mask matching functions want) - Switch the elements that come from splat vectors so that they match the corresponding elements from the other vector (to allow for merges) - Adds debugging messages for when a shuffle is matched to a VPERM so that anyone interested in improving this further can get the info for their code Differential revision: https://reviews.llvm.org/D77448
2020-06-19 04:53:50 +02:00
; P9LE-NEXT: xxmrghd v2, v2, vs0
[PowerPC] Improve codegen for vector loads using scalar_to_vector This patch aims to improve the codegen for vector loads involving the scalar_to_vector (load X) sequence. Initially, ld->mv instructions were used for scalar_to_vector (load X), so this patch allows scalar_to_vector (load X) to utilize: LXSD and LXSDX for i64 and f64 LXSIWAX for i32 (sign extension to i64) LXSIWZX for i32 and f64 Committing on behalf of Amy Kwan. Differential Revision: https://reviews.llvm.org/D48950 llvm-svn: 339260
2018-08-08 17:20:43 +02:00
; P9LE-NEXT: blr
; P9BE-LABEL: s2v_test_f1:
; P9BE: # %bb.0: # %entry
; P9BE-NEXT: lfd f0, 0(r3)
; P9BE-NEXT: xxpermdi v2, vs0, v2, 1
; P9BE-NEXT: blr
; P8LE-LABEL: s2v_test_f1:
; P8LE: # %bb.0: # %entry
; P8LE-NEXT: lfdx f0, 0, r3
[PowerPC] Canonicalize shuffles to match more single-instruction masks on LE We currently miss a number of opportunities to emit single-instruction VMRG[LH][BHW] instructions for shuffles on little endian subtargets. Although this in itself is not a huge performance opportunity since loading the permute vector for a VPERM can always be pulled out of loops, producing such merge instructions is useful to downstream optimizations. Since VPERM is essentially opaque to all subsequent optimizations, we want to avoid it as much as possible. Other permute instructions have semantics that can be reasoned about much more easily in later optimizations. This patch does the following: - Canonicalize shuffles so that the first element comes from the first vector (since that's what most of the mask matching functions want) - Switch the elements that come from splat vectors so that they match the corresponding elements from the other vector (to allow for merges) - Adds debugging messages for when a shuffle is matched to a VPERM so that anyone interested in improving this further can get the info for their code Differential revision: https://reviews.llvm.org/D77448
2020-06-19 04:53:50 +02:00
; P8LE-NEXT: xxmrghd v2, v2, vs0
[PowerPC] Improve codegen for vector loads using scalar_to_vector This patch aims to improve the codegen for vector loads involving the scalar_to_vector (load X) sequence. Initially, ld->mv instructions were used for scalar_to_vector (load X), so this patch allows scalar_to_vector (load X) to utilize: LXSD and LXSDX for i64 and f64 LXSIWAX for i32 (sign extension to i64) LXSIWZX for i32 and f64 Committing on behalf of Amy Kwan. Differential Revision: https://reviews.llvm.org/D48950 llvm-svn: 339260
2018-08-08 17:20:43 +02:00
; P8LE-NEXT: blr
; P8BE-LABEL: s2v_test_f1:
; P8BE: # %bb.0: # %entry
; P8BE-NEXT: lfdx f0, 0, r3
; P8BE-NEXT: xxpermdi v2, vs0, v2, 1
; P8BE-NEXT: blr
entry:
%0 = load double, double* %f64, align 8
%vecins = insertelement <2 x double> %vec, double %0, i32 0
ret <2 x double> %vecins
}
; Function Attrs: norecurse nounwind readonly
define <2 x double> @s2v_test_f2(double* nocapture readonly %f64, <2 x double> %vec) {
; P9LE-LABEL: s2v_test_f2:
; P9LE: # %bb.0: # %entry
; P9LE-NEXT: lfd f0, 8(r3)
[PowerPC] Canonicalize shuffles to match more single-instruction masks on LE We currently miss a number of opportunities to emit single-instruction VMRG[LH][BHW] instructions for shuffles on little endian subtargets. Although this in itself is not a huge performance opportunity since loading the permute vector for a VPERM can always be pulled out of loops, producing such merge instructions is useful to downstream optimizations. Since VPERM is essentially opaque to all subsequent optimizations, we want to avoid it as much as possible. Other permute instructions have semantics that can be reasoned about much more easily in later optimizations. This patch does the following: - Canonicalize shuffles so that the first element comes from the first vector (since that's what most of the mask matching functions want) - Switch the elements that come from splat vectors so that they match the corresponding elements from the other vector (to allow for merges) - Adds debugging messages for when a shuffle is matched to a VPERM so that anyone interested in improving this further can get the info for their code Differential revision: https://reviews.llvm.org/D77448
2020-06-19 04:53:50 +02:00
; P9LE-NEXT: xxmrghd v2, v2, vs0
[PowerPC] Improve codegen for vector loads using scalar_to_vector This patch aims to improve the codegen for vector loads involving the scalar_to_vector (load X) sequence. Initially, ld->mv instructions were used for scalar_to_vector (load X), so this patch allows scalar_to_vector (load X) to utilize: LXSD and LXSDX for i64 and f64 LXSIWAX for i32 (sign extension to i64) LXSIWZX for i32 and f64 Committing on behalf of Amy Kwan. Differential Revision: https://reviews.llvm.org/D48950 llvm-svn: 339260
2018-08-08 17:20:43 +02:00
; P9LE-NEXT: blr
; P9BE-LABEL: s2v_test_f2:
; P9BE: # %bb.0: # %entry
; P9BE-NEXT: lfd f0, 8(r3)
; P9BE-NEXT: xxpermdi v2, vs0, v2, 1
; P9BE-NEXT: blr
; P8LE-LABEL: s2v_test_f2:
; P8LE: # %bb.0: # %entry
[PowerPC] Add a peephole post RA to transform the inst that fed by add If the arch is P8, we will select XFLOAD to load the floating point, and then, expand it to vsx and non-vsx X-form instruction post RA. This patch is trying to convert the X-form to D-form if it meets the requirement that one operand of the x-form inst is the special Zero register, and another operand fed by add inst. i.e. y = add imm, reg LFDX. 0, y --> LFD imm(reg) Reviewers: Nemanjai Differential Revision: https://reviews.llvm.org/D49007 llvm-svn: 340149
2018-08-20 04:52:55 +02:00
; P8LE-NEXT: lfd f0, 8(r3)
[PowerPC] Canonicalize shuffles to match more single-instruction masks on LE We currently miss a number of opportunities to emit single-instruction VMRG[LH][BHW] instructions for shuffles on little endian subtargets. Although this in itself is not a huge performance opportunity since loading the permute vector for a VPERM can always be pulled out of loops, producing such merge instructions is useful to downstream optimizations. Since VPERM is essentially opaque to all subsequent optimizations, we want to avoid it as much as possible. Other permute instructions have semantics that can be reasoned about much more easily in later optimizations. This patch does the following: - Canonicalize shuffles so that the first element comes from the first vector (since that's what most of the mask matching functions want) - Switch the elements that come from splat vectors so that they match the corresponding elements from the other vector (to allow for merges) - Adds debugging messages for when a shuffle is matched to a VPERM so that anyone interested in improving this further can get the info for their code Differential revision: https://reviews.llvm.org/D77448
2020-06-19 04:53:50 +02:00
; P8LE-NEXT: xxmrghd v2, v2, vs0
[PowerPC] Improve codegen for vector loads using scalar_to_vector This patch aims to improve the codegen for vector loads involving the scalar_to_vector (load X) sequence. Initially, ld->mv instructions were used for scalar_to_vector (load X), so this patch allows scalar_to_vector (load X) to utilize: LXSD and LXSDX for i64 and f64 LXSIWAX for i32 (sign extension to i64) LXSIWZX for i32 and f64 Committing on behalf of Amy Kwan. Differential Revision: https://reviews.llvm.org/D48950 llvm-svn: 339260
2018-08-08 17:20:43 +02:00
; P8LE-NEXT: blr
; P8BE-LABEL: s2v_test_f2:
; P8BE: # %bb.0: # %entry
[PowerPC] Add a peephole post RA to transform the inst that fed by add If the arch is P8, we will select XFLOAD to load the floating point, and then, expand it to vsx and non-vsx X-form instruction post RA. This patch is trying to convert the X-form to D-form if it meets the requirement that one operand of the x-form inst is the special Zero register, and another operand fed by add inst. i.e. y = add imm, reg LFDX. 0, y --> LFD imm(reg) Reviewers: Nemanjai Differential Revision: https://reviews.llvm.org/D49007 llvm-svn: 340149
2018-08-20 04:52:55 +02:00
; P8BE-NEXT: lfd f0, 8(r3)
[PowerPC] Improve codegen for vector loads using scalar_to_vector This patch aims to improve the codegen for vector loads involving the scalar_to_vector (load X) sequence. Initially, ld->mv instructions were used for scalar_to_vector (load X), so this patch allows scalar_to_vector (load X) to utilize: LXSD and LXSDX for i64 and f64 LXSIWAX for i32 (sign extension to i64) LXSIWZX for i32 and f64 Committing on behalf of Amy Kwan. Differential Revision: https://reviews.llvm.org/D48950 llvm-svn: 339260
2018-08-08 17:20:43 +02:00
; P8BE-NEXT: xxpermdi v2, vs0, v2, 1
; P8BE-NEXT: blr
entry:
%arrayidx = getelementptr inbounds double, double* %f64, i64 1
%0 = load double, double* %arrayidx, align 8
%vecins = insertelement <2 x double> %vec, double %0, i32 0
ret <2 x double> %vecins
}
; Function Attrs: norecurse nounwind readonly
define <2 x double> @s2v_test_f3(double* nocapture readonly %f64, <2 x double> %vec, i32 signext %Idx) {
; P9LE-LABEL: s2v_test_f3:
; P9LE: # %bb.0: # %entry
; P9LE-NEXT: sldi r4, r7, 3
; P9LE-NEXT: lfdx f0, r3, r4
[PowerPC] Canonicalize shuffles to match more single-instruction masks on LE We currently miss a number of opportunities to emit single-instruction VMRG[LH][BHW] instructions for shuffles on little endian subtargets. Although this in itself is not a huge performance opportunity since loading the permute vector for a VPERM can always be pulled out of loops, producing such merge instructions is useful to downstream optimizations. Since VPERM is essentially opaque to all subsequent optimizations, we want to avoid it as much as possible. Other permute instructions have semantics that can be reasoned about much more easily in later optimizations. This patch does the following: - Canonicalize shuffles so that the first element comes from the first vector (since that's what most of the mask matching functions want) - Switch the elements that come from splat vectors so that they match the corresponding elements from the other vector (to allow for merges) - Adds debugging messages for when a shuffle is matched to a VPERM so that anyone interested in improving this further can get the info for their code Differential revision: https://reviews.llvm.org/D77448
2020-06-19 04:53:50 +02:00
; P9LE-NEXT: xxmrghd v2, v2, vs0
[PowerPC] Improve codegen for vector loads using scalar_to_vector This patch aims to improve the codegen for vector loads involving the scalar_to_vector (load X) sequence. Initially, ld->mv instructions were used for scalar_to_vector (load X), so this patch allows scalar_to_vector (load X) to utilize: LXSD and LXSDX for i64 and f64 LXSIWAX for i32 (sign extension to i64) LXSIWZX for i32 and f64 Committing on behalf of Amy Kwan. Differential Revision: https://reviews.llvm.org/D48950 llvm-svn: 339260
2018-08-08 17:20:43 +02:00
; P9LE-NEXT: blr
; P9BE-LABEL: s2v_test_f3:
; P9BE: # %bb.0: # %entry
; P9BE-NEXT: sldi r4, r7, 3
; P9BE-NEXT: lfdx f0, r3, r4
; P9BE-NEXT: xxpermdi v2, vs0, v2, 1
; P9BE-NEXT: blr
; P8LE-LABEL: s2v_test_f3:
; P8LE: # %bb.0: # %entry
; P8LE-NEXT: sldi r4, r7, 3
; P8LE-NEXT: lfdx f0, r3, r4
[PowerPC] Canonicalize shuffles to match more single-instruction masks on LE We currently miss a number of opportunities to emit single-instruction VMRG[LH][BHW] instructions for shuffles on little endian subtargets. Although this in itself is not a huge performance opportunity since loading the permute vector for a VPERM can always be pulled out of loops, producing such merge instructions is useful to downstream optimizations. Since VPERM is essentially opaque to all subsequent optimizations, we want to avoid it as much as possible. Other permute instructions have semantics that can be reasoned about much more easily in later optimizations. This patch does the following: - Canonicalize shuffles so that the first element comes from the first vector (since that's what most of the mask matching functions want) - Switch the elements that come from splat vectors so that they match the corresponding elements from the other vector (to allow for merges) - Adds debugging messages for when a shuffle is matched to a VPERM so that anyone interested in improving this further can get the info for their code Differential revision: https://reviews.llvm.org/D77448
2020-06-19 04:53:50 +02:00
; P8LE-NEXT: xxmrghd v2, v2, vs0
[PowerPC] Improve codegen for vector loads using scalar_to_vector This patch aims to improve the codegen for vector loads involving the scalar_to_vector (load X) sequence. Initially, ld->mv instructions were used for scalar_to_vector (load X), so this patch allows scalar_to_vector (load X) to utilize: LXSD and LXSDX for i64 and f64 LXSIWAX for i32 (sign extension to i64) LXSIWZX for i32 and f64 Committing on behalf of Amy Kwan. Differential Revision: https://reviews.llvm.org/D48950 llvm-svn: 339260
2018-08-08 17:20:43 +02:00
; P8LE-NEXT: blr
; P8BE-LABEL: s2v_test_f3:
; P8BE: # %bb.0: # %entry
; P8BE-NEXT: sldi r4, r7, 3
; P8BE-NEXT: lfdx f0, r3, r4
; P8BE-NEXT: xxpermdi v2, vs0, v2, 1
; P8BE-NEXT: blr
entry:
%idxprom = sext i32 %Idx to i64
%arrayidx = getelementptr inbounds double, double* %f64, i64 %idxprom
%0 = load double, double* %arrayidx, align 8
%vecins = insertelement <2 x double> %vec, double %0, i32 0
ret <2 x double> %vecins
}
; Function Attrs: norecurse nounwind readonly
define <2 x double> @s2v_test_f4(double* nocapture readonly %f64, <2 x double> %vec) {
; P9LE-LABEL: s2v_test_f4:
; P9LE: # %bb.0: # %entry
; P9LE-NEXT: lfd f0, 8(r3)
[PowerPC] Canonicalize shuffles to match more single-instruction masks on LE We currently miss a number of opportunities to emit single-instruction VMRG[LH][BHW] instructions for shuffles on little endian subtargets. Although this in itself is not a huge performance opportunity since loading the permute vector for a VPERM can always be pulled out of loops, producing such merge instructions is useful to downstream optimizations. Since VPERM is essentially opaque to all subsequent optimizations, we want to avoid it as much as possible. Other permute instructions have semantics that can be reasoned about much more easily in later optimizations. This patch does the following: - Canonicalize shuffles so that the first element comes from the first vector (since that's what most of the mask matching functions want) - Switch the elements that come from splat vectors so that they match the corresponding elements from the other vector (to allow for merges) - Adds debugging messages for when a shuffle is matched to a VPERM so that anyone interested in improving this further can get the info for their code Differential revision: https://reviews.llvm.org/D77448
2020-06-19 04:53:50 +02:00
; P9LE-NEXT: xxmrghd v2, v2, vs0
[PowerPC] Improve codegen for vector loads using scalar_to_vector This patch aims to improve the codegen for vector loads involving the scalar_to_vector (load X) sequence. Initially, ld->mv instructions were used for scalar_to_vector (load X), so this patch allows scalar_to_vector (load X) to utilize: LXSD and LXSDX for i64 and f64 LXSIWAX for i32 (sign extension to i64) LXSIWZX for i32 and f64 Committing on behalf of Amy Kwan. Differential Revision: https://reviews.llvm.org/D48950 llvm-svn: 339260
2018-08-08 17:20:43 +02:00
; P9LE-NEXT: blr
; P9BE-LABEL: s2v_test_f4:
; P9BE: # %bb.0: # %entry
; P9BE-NEXT: lfd f0, 8(r3)
; P9BE-NEXT: xxpermdi v2, vs0, v2, 1
; P9BE-NEXT: blr
; P8LE-LABEL: s2v_test_f4:
; P8LE: # %bb.0: # %entry
[PowerPC] Add a peephole post RA to transform the inst that fed by add If the arch is P8, we will select XFLOAD to load the floating point, and then, expand it to vsx and non-vsx X-form instruction post RA. This patch is trying to convert the X-form to D-form if it meets the requirement that one operand of the x-form inst is the special Zero register, and another operand fed by add inst. i.e. y = add imm, reg LFDX. 0, y --> LFD imm(reg) Reviewers: Nemanjai Differential Revision: https://reviews.llvm.org/D49007 llvm-svn: 340149
2018-08-20 04:52:55 +02:00
; P8LE-NEXT: lfd f0, 8(r3)
[PowerPC] Canonicalize shuffles to match more single-instruction masks on LE We currently miss a number of opportunities to emit single-instruction VMRG[LH][BHW] instructions for shuffles on little endian subtargets. Although this in itself is not a huge performance opportunity since loading the permute vector for a VPERM can always be pulled out of loops, producing such merge instructions is useful to downstream optimizations. Since VPERM is essentially opaque to all subsequent optimizations, we want to avoid it as much as possible. Other permute instructions have semantics that can be reasoned about much more easily in later optimizations. This patch does the following: - Canonicalize shuffles so that the first element comes from the first vector (since that's what most of the mask matching functions want) - Switch the elements that come from splat vectors so that they match the corresponding elements from the other vector (to allow for merges) - Adds debugging messages for when a shuffle is matched to a VPERM so that anyone interested in improving this further can get the info for their code Differential revision: https://reviews.llvm.org/D77448
2020-06-19 04:53:50 +02:00
; P8LE-NEXT: xxmrghd v2, v2, vs0
[PowerPC] Improve codegen for vector loads using scalar_to_vector This patch aims to improve the codegen for vector loads involving the scalar_to_vector (load X) sequence. Initially, ld->mv instructions were used for scalar_to_vector (load X), so this patch allows scalar_to_vector (load X) to utilize: LXSD and LXSDX for i64 and f64 LXSIWAX for i32 (sign extension to i64) LXSIWZX for i32 and f64 Committing on behalf of Amy Kwan. Differential Revision: https://reviews.llvm.org/D48950 llvm-svn: 339260
2018-08-08 17:20:43 +02:00
; P8LE-NEXT: blr
; P8BE-LABEL: s2v_test_f4:
; P8BE: # %bb.0: # %entry
[PowerPC] Add a peephole post RA to transform the inst that fed by add If the arch is P8, we will select XFLOAD to load the floating point, and then, expand it to vsx and non-vsx X-form instruction post RA. This patch is trying to convert the X-form to D-form if it meets the requirement that one operand of the x-form inst is the special Zero register, and another operand fed by add inst. i.e. y = add imm, reg LFDX. 0, y --> LFD imm(reg) Reviewers: Nemanjai Differential Revision: https://reviews.llvm.org/D49007 llvm-svn: 340149
2018-08-20 04:52:55 +02:00
; P8BE-NEXT: lfd f0, 8(r3)
[PowerPC] Improve codegen for vector loads using scalar_to_vector This patch aims to improve the codegen for vector loads involving the scalar_to_vector (load X) sequence. Initially, ld->mv instructions were used for scalar_to_vector (load X), so this patch allows scalar_to_vector (load X) to utilize: LXSD and LXSDX for i64 and f64 LXSIWAX for i32 (sign extension to i64) LXSIWZX for i32 and f64 Committing on behalf of Amy Kwan. Differential Revision: https://reviews.llvm.org/D48950 llvm-svn: 339260
2018-08-08 17:20:43 +02:00
; P8BE-NEXT: xxpermdi v2, vs0, v2, 1
; P8BE-NEXT: blr
entry:
%arrayidx = getelementptr inbounds double, double* %f64, i64 1
%0 = load double, double* %arrayidx, align 8
%vecins = insertelement <2 x double> %vec, double %0, i32 0
ret <2 x double> %vecins
}
; Function Attrs: norecurse nounwind readonly
define <2 x double> @s2v_test_f5(<2 x double> %vec, double* nocapture readonly %ptr1) {
; P9LE-LABEL: s2v_test_f5:
; P9LE: # %bb.0: # %entry
; P9LE-NEXT: lfd f0, 0(r5)
[PowerPC] Canonicalize shuffles to match more single-instruction masks on LE We currently miss a number of opportunities to emit single-instruction VMRG[LH][BHW] instructions for shuffles on little endian subtargets. Although this in itself is not a huge performance opportunity since loading the permute vector for a VPERM can always be pulled out of loops, producing such merge instructions is useful to downstream optimizations. Since VPERM is essentially opaque to all subsequent optimizations, we want to avoid it as much as possible. Other permute instructions have semantics that can be reasoned about much more easily in later optimizations. This patch does the following: - Canonicalize shuffles so that the first element comes from the first vector (since that's what most of the mask matching functions want) - Switch the elements that come from splat vectors so that they match the corresponding elements from the other vector (to allow for merges) - Adds debugging messages for when a shuffle is matched to a VPERM so that anyone interested in improving this further can get the info for their code Differential revision: https://reviews.llvm.org/D77448
2020-06-19 04:53:50 +02:00
; P9LE-NEXT: xxmrghd v2, v2, vs0
[PowerPC] Improve codegen for vector loads using scalar_to_vector This patch aims to improve the codegen for vector loads involving the scalar_to_vector (load X) sequence. Initially, ld->mv instructions were used for scalar_to_vector (load X), so this patch allows scalar_to_vector (load X) to utilize: LXSD and LXSDX for i64 and f64 LXSIWAX for i32 (sign extension to i64) LXSIWZX for i32 and f64 Committing on behalf of Amy Kwan. Differential Revision: https://reviews.llvm.org/D48950 llvm-svn: 339260
2018-08-08 17:20:43 +02:00
; P9LE-NEXT: blr
; P9BE-LABEL: s2v_test_f5:
; P9BE: # %bb.0: # %entry
; P9BE-NEXT: lfd f0, 0(r5)
; P9BE-NEXT: xxpermdi v2, vs0, v2, 1
; P9BE-NEXT: blr
; P8LE-LABEL: s2v_test_f5:
; P8LE: # %bb.0: # %entry
; P8LE-NEXT: lfdx f0, 0, r5
[PowerPC] Canonicalize shuffles to match more single-instruction masks on LE We currently miss a number of opportunities to emit single-instruction VMRG[LH][BHW] instructions for shuffles on little endian subtargets. Although this in itself is not a huge performance opportunity since loading the permute vector for a VPERM can always be pulled out of loops, producing such merge instructions is useful to downstream optimizations. Since VPERM is essentially opaque to all subsequent optimizations, we want to avoid it as much as possible. Other permute instructions have semantics that can be reasoned about much more easily in later optimizations. This patch does the following: - Canonicalize shuffles so that the first element comes from the first vector (since that's what most of the mask matching functions want) - Switch the elements that come from splat vectors so that they match the corresponding elements from the other vector (to allow for merges) - Adds debugging messages for when a shuffle is matched to a VPERM so that anyone interested in improving this further can get the info for their code Differential revision: https://reviews.llvm.org/D77448
2020-06-19 04:53:50 +02:00
; P8LE-NEXT: xxmrghd v2, v2, vs0
[PowerPC] Improve codegen for vector loads using scalar_to_vector This patch aims to improve the codegen for vector loads involving the scalar_to_vector (load X) sequence. Initially, ld->mv instructions were used for scalar_to_vector (load X), so this patch allows scalar_to_vector (load X) to utilize: LXSD and LXSDX for i64 and f64 LXSIWAX for i32 (sign extension to i64) LXSIWZX for i32 and f64 Committing on behalf of Amy Kwan. Differential Revision: https://reviews.llvm.org/D48950 llvm-svn: 339260
2018-08-08 17:20:43 +02:00
; P8LE-NEXT: blr
; P8BE-LABEL: s2v_test_f5:
; P8BE: # %bb.0: # %entry
; P8BE-NEXT: lfdx f0, 0, r5
; P8BE-NEXT: xxpermdi v2, vs0, v2, 1
; P8BE-NEXT: blr
entry:
%0 = load double, double* %ptr1, align 8
%vecins = insertelement <2 x double> %vec, double %0, i32 0
ret <2 x double> %vecins
}
Reference in New Issue Copy Permalink