[llvm] 317d70e - [SLP][X86] Add fmuladd test coverage
Simon Pilgrim via llvm-commits
llvm-commits at lists.llvm.org
Mon Aug 2 12:59:28 PDT 2021
Author: Simon Pilgrim
Date: 2021-08-02T20:59:12+01:00
New Revision: 317d70ea91841dfe503fddfe94b4a03b829917c6
URL: https://github.com/llvm/llvm-project/commit/317d70ea91841dfe503fddfe94b4a03b829917c6
DIFF: https://github.com/llvm/llvm-project/commit/317d70ea91841dfe503fddfe94b4a03b829917c6.diff
LOG: [SLP][X86] Add fmuladd test coverage
Added:
llvm/test/Transforms/SLPVectorizer/X86/fmuladd.ll
Modified:
Removed:
################################################################################
diff --git a/llvm/test/Transforms/SLPVectorizer/X86/fmuladd.ll b/llvm/test/Transforms/SLPVectorizer/X86/fmuladd.ll
new file mode 100644
index 0000000000000..f7b8087134f38
--- /dev/null
+++ b/llvm/test/Transforms/SLPVectorizer/X86/fmuladd.ll
@@ -0,0 +1,407 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -mtriple=x86_64-unknown -basic-aa -slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,SSE
+; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=corei7-avx -basic-aa -slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,AVX
+; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=bdver1 -basic-aa -slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,AVX,AVX256
+; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=core-avx2 -basic-aa -slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,AVX,AVX256
+; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=skylake-avx512 -mattr=-prefer-256-bit -basic-aa -slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,AVX,AVX512
+; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=skylake-avx512 -mattr=+prefer-256-bit -basic-aa -slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,AVX,AVX256
+
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+
+ at srcA64 = common global [8 x double] zeroinitializer, align 64
+ at srcB64 = common global [8 x double] zeroinitializer, align 64
+ at srcC64 = common global [8 x double] zeroinitializer, align 64
+ at srcA32 = common global [16 x float] zeroinitializer, align 64
+ at srcB32 = common global [16 x float] zeroinitializer, align 64
+ at srcC32 = common global [16 x float] zeroinitializer, align 64
+ at dst64 = common global [8 x double] zeroinitializer, align 64
+ at dst32 = common global [16 x float] zeroinitializer, align 64
+
+declare float @llvm.fmuladd.f32(float, float, float)
+declare double @llvm.fmuladd.f64(double, double, double)
+
+;
+; fmuladd
+;
+
+define void @fmuladd_2f64() #0 {
+; CHECK-LABEL: @fmuladd_2f64(
+; CHECK-NEXT: [[TMP1:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @srcA64 to <2 x double>*), align 8
+; CHECK-NEXT: [[TMP2:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @srcB64 to <2 x double>*), align 8
+; CHECK-NEXT: [[TMP3:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @srcC64 to <2 x double>*), align 8
+; CHECK-NEXT: [[TMP4:%.*]] = call <2 x double> @llvm.fmuladd.v2f64(<2 x double> [[TMP1]], <2 x double> [[TMP2]], <2 x double> [[TMP3]])
+; CHECK-NEXT: store <2 x double> [[TMP4]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 8
+; CHECK-NEXT: ret void
+;
+ %a0 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 0), align 8
+ %a1 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 1), align 8
+ %b0 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 0), align 8
+ %b1 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 1), align 8
+ %c0 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcC64, i32 0, i64 0), align 8
+ %c1 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcC64, i32 0, i64 1), align 8
+ %fmuladd0 = call double @llvm.fmuladd.f64(double %a0, double %b0, double %c0)
+ %fmuladd1 = call double @llvm.fmuladd.f64(double %a1, double %b1, double %c1)
+ store double %fmuladd0, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
+ store double %fmuladd1, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
+ ret void
+}
+
+define void @fmuladd_4f64() #0 {
+; SSE-LABEL: @fmuladd_4f64(
+; SSE-NEXT: [[TMP1:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @srcA64 to <2 x double>*), align 8
+; SSE-NEXT: [[TMP2:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 2) to <2 x double>*), align 8
+; SSE-NEXT: [[TMP3:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @srcB64 to <2 x double>*), align 8
+; SSE-NEXT: [[TMP4:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 2) to <2 x double>*), align 8
+; SSE-NEXT: [[TMP5:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @srcC64 to <2 x double>*), align 8
+; SSE-NEXT: [[TMP6:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @srcC64, i32 0, i64 2) to <2 x double>*), align 8
+; SSE-NEXT: [[TMP7:%.*]] = call <2 x double> @llvm.fmuladd.v2f64(<2 x double> [[TMP1]], <2 x double> [[TMP3]], <2 x double> [[TMP5]])
+; SSE-NEXT: [[TMP8:%.*]] = call <2 x double> @llvm.fmuladd.v2f64(<2 x double> [[TMP2]], <2 x double> [[TMP4]], <2 x double> [[TMP6]])
+; SSE-NEXT: store <2 x double> [[TMP7]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 8
+; SSE-NEXT: store <2 x double> [[TMP8]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2) to <2 x double>*), align 8
+; SSE-NEXT: ret void
+;
+; AVX-LABEL: @fmuladd_4f64(
+; AVX-NEXT: [[TMP1:%.*]] = load <4 x double>, <4 x double>* bitcast ([8 x double]* @srcA64 to <4 x double>*), align 8
+; AVX-NEXT: [[TMP2:%.*]] = load <4 x double>, <4 x double>* bitcast ([8 x double]* @srcB64 to <4 x double>*), align 8
+; AVX-NEXT: [[TMP3:%.*]] = load <4 x double>, <4 x double>* bitcast ([8 x double]* @srcC64 to <4 x double>*), align 8
+; AVX-NEXT: [[TMP4:%.*]] = call <4 x double> @llvm.fmuladd.v4f64(<4 x double> [[TMP1]], <4 x double> [[TMP2]], <4 x double> [[TMP3]])
+; AVX-NEXT: store <4 x double> [[TMP4]], <4 x double>* bitcast ([8 x double]* @dst64 to <4 x double>*), align 8
+; AVX-NEXT: ret void
+;
+ %a0 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 0), align 8
+ %a1 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 1), align 8
+ %a2 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 2), align 8
+ %a3 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 3), align 8
+ %b0 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 0), align 8
+ %b1 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 1), align 8
+ %b2 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 2), align 8
+ %b3 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 3), align 8
+ %c0 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcC64, i32 0, i64 0), align 8
+ %c1 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcC64, i32 0, i64 1), align 8
+ %c2 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcC64, i32 0, i64 2), align 8
+ %c3 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcC64, i32 0, i64 3), align 8
+ %fmuladd0 = call double @llvm.fmuladd.f64(double %a0, double %b0, double %c0)
+ %fmuladd1 = call double @llvm.fmuladd.f64(double %a1, double %b1, double %c1)
+ %fmuladd2 = call double @llvm.fmuladd.f64(double %a2, double %b2, double %c2)
+ %fmuladd3 = call double @llvm.fmuladd.f64(double %a3, double %b3, double %c3)
+ store double %fmuladd0, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
+ store double %fmuladd1, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
+ store double %fmuladd2, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 8
+ store double %fmuladd3, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
+ ret void
+}
+
+define void @fmuladd_8f64() #0 {
+; SSE-LABEL: @fmuladd_8f64(
+; SSE-NEXT: [[TMP1:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @srcA64 to <2 x double>*), align 4
+; SSE-NEXT: [[TMP2:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 2) to <2 x double>*), align 4
+; SSE-NEXT: [[TMP3:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 4) to <2 x double>*), align 4
+; SSE-NEXT: [[TMP4:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 6) to <2 x double>*), align 4
+; SSE-NEXT: [[TMP5:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @srcB64 to <2 x double>*), align 4
+; SSE-NEXT: [[TMP6:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 2) to <2 x double>*), align 4
+; SSE-NEXT: [[TMP7:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 4) to <2 x double>*), align 4
+; SSE-NEXT: [[TMP8:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 6) to <2 x double>*), align 4
+; SSE-NEXT: [[TMP9:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @srcC64 to <2 x double>*), align 4
+; SSE-NEXT: [[TMP10:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @srcC64, i32 0, i64 2) to <2 x double>*), align 4
+; SSE-NEXT: [[TMP11:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @srcC64, i32 0, i64 4) to <2 x double>*), align 4
+; SSE-NEXT: [[TMP12:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @srcC64, i32 0, i64 6) to <2 x double>*), align 4
+; SSE-NEXT: [[TMP13:%.*]] = call <2 x double> @llvm.fmuladd.v2f64(<2 x double> [[TMP1]], <2 x double> [[TMP5]], <2 x double> [[TMP9]])
+; SSE-NEXT: [[TMP14:%.*]] = call <2 x double> @llvm.fmuladd.v2f64(<2 x double> [[TMP2]], <2 x double> [[TMP6]], <2 x double> [[TMP10]])
+; SSE-NEXT: [[TMP15:%.*]] = call <2 x double> @llvm.fmuladd.v2f64(<2 x double> [[TMP3]], <2 x double> [[TMP7]], <2 x double> [[TMP11]])
+; SSE-NEXT: [[TMP16:%.*]] = call <2 x double> @llvm.fmuladd.v2f64(<2 x double> [[TMP4]], <2 x double> [[TMP8]], <2 x double> [[TMP12]])
+; SSE-NEXT: store <2 x double> [[TMP13]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 4
+; SSE-NEXT: store <2 x double> [[TMP14]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2) to <2 x double>*), align 4
+; SSE-NEXT: store <2 x double> [[TMP15]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4) to <2 x double>*), align 4
+; SSE-NEXT: store <2 x double> [[TMP16]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 6) to <2 x double>*), align 4
+; SSE-NEXT: ret void
+;
+; AVX256-LABEL: @fmuladd_8f64(
+; AVX256-NEXT: [[TMP1:%.*]] = load <4 x double>, <4 x double>* bitcast ([8 x double]* @srcA64 to <4 x double>*), align 4
+; AVX256-NEXT: [[TMP2:%.*]] = load <4 x double>, <4 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 4) to <4 x double>*), align 4
+; AVX256-NEXT: [[TMP3:%.*]] = load <4 x double>, <4 x double>* bitcast ([8 x double]* @srcB64 to <4 x double>*), align 4
+; AVX256-NEXT: [[TMP4:%.*]] = load <4 x double>, <4 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 4) to <4 x double>*), align 4
+; AVX256-NEXT: [[TMP5:%.*]] = load <4 x double>, <4 x double>* bitcast ([8 x double]* @srcC64 to <4 x double>*), align 4
+; AVX256-NEXT: [[TMP6:%.*]] = load <4 x double>, <4 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @srcC64, i32 0, i64 4) to <4 x double>*), align 4
+; AVX256-NEXT: [[TMP7:%.*]] = call <4 x double> @llvm.fmuladd.v4f64(<4 x double> [[TMP1]], <4 x double> [[TMP3]], <4 x double> [[TMP5]])
+; AVX256-NEXT: [[TMP8:%.*]] = call <4 x double> @llvm.fmuladd.v4f64(<4 x double> [[TMP2]], <4 x double> [[TMP4]], <4 x double> [[TMP6]])
+; AVX256-NEXT: store <4 x double> [[TMP7]], <4 x double>* bitcast ([8 x double]* @dst64 to <4 x double>*), align 4
+; AVX256-NEXT: store <4 x double> [[TMP8]], <4 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4) to <4 x double>*), align 4
+; AVX256-NEXT: ret void
+;
+; AVX512-LABEL: @fmuladd_8f64(
+; AVX512-NEXT: [[TMP1:%.*]] = load <8 x double>, <8 x double>* bitcast ([8 x double]* @srcA64 to <8 x double>*), align 4
+; AVX512-NEXT: [[TMP2:%.*]] = load <8 x double>, <8 x double>* bitcast ([8 x double]* @srcB64 to <8 x double>*), align 4
+; AVX512-NEXT: [[TMP3:%.*]] = load <8 x double>, <8 x double>* bitcast ([8 x double]* @srcC64 to <8 x double>*), align 4
+; AVX512-NEXT: [[TMP4:%.*]] = call <8 x double> @llvm.fmuladd.v8f64(<8 x double> [[TMP1]], <8 x double> [[TMP2]], <8 x double> [[TMP3]])
+; AVX512-NEXT: store <8 x double> [[TMP4]], <8 x double>* bitcast ([8 x double]* @dst64 to <8 x double>*), align 4
+; AVX512-NEXT: ret void
+;
+ %a0 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 0), align 4
+ %a1 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 1), align 4
+ %a2 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 2), align 4
+ %a3 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 3), align 4
+ %a4 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 4), align 4
+ %a5 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 5), align 4
+ %a6 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 6), align 4
+ %a7 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 7), align 4
+ %b0 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 0), align 4
+ %b1 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 1), align 4
+ %b2 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 2), align 4
+ %b3 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 3), align 4
+ %b4 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 4), align 4
+ %b5 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 5), align 4
+ %b6 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 6), align 4
+ %b7 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 7), align 4
+ %c0 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcC64, i32 0, i64 0), align 4
+ %c1 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcC64, i32 0, i64 1), align 4
+ %c2 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcC64, i32 0, i64 2), align 4
+ %c3 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcC64, i32 0, i64 3), align 4
+ %c4 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcC64, i32 0, i64 4), align 4
+ %c5 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcC64, i32 0, i64 5), align 4
+ %c6 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcC64, i32 0, i64 6), align 4
+ %c7 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcC64, i32 0, i64 7), align 4
+ %fmuladd0 = call double @llvm.fmuladd.f64(double %a0, double %b0, double %c0)
+ %fmuladd1 = call double @llvm.fmuladd.f64(double %a1, double %b1, double %c1)
+ %fmuladd2 = call double @llvm.fmuladd.f64(double %a2, double %b2, double %c2)
+ %fmuladd3 = call double @llvm.fmuladd.f64(double %a3, double %b3, double %c3)
+ %fmuladd4 = call double @llvm.fmuladd.f64(double %a4, double %b4, double %c4)
+ %fmuladd5 = call double @llvm.fmuladd.f64(double %a5, double %b5, double %c5)
+ %fmuladd6 = call double @llvm.fmuladd.f64(double %a6, double %b6, double %c6)
+ %fmuladd7 = call double @llvm.fmuladd.f64(double %a7, double %b7, double %c7)
+ store double %fmuladd0, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 4
+ store double %fmuladd1, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 4
+ store double %fmuladd2, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 4
+ store double %fmuladd3, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 4
+ store double %fmuladd4, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4), align 4
+ store double %fmuladd5, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 5), align 4
+ store double %fmuladd6, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 6), align 4
+ store double %fmuladd7, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 7), align 4
+ ret void
+}
+
+define void @fmuladd_4f32() #0 {
+; CHECK-LABEL: @fmuladd_4f32(
+; CHECK-NEXT: [[TMP1:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @srcA32 to <4 x float>*), align 4
+; CHECK-NEXT: [[TMP2:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @srcB32 to <4 x float>*), align 4
+; CHECK-NEXT: [[TMP3:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @srcC32 to <4 x float>*), align 4
+; CHECK-NEXT: [[TMP4:%.*]] = call <4 x float> @llvm.fmuladd.v4f32(<4 x float> [[TMP1]], <4 x float> [[TMP2]], <4 x float> [[TMP3]])
+; CHECK-NEXT: store <4 x float> [[TMP4]], <4 x float>* bitcast ([16 x float]* @dst32 to <4 x float>*), align 4
+; CHECK-NEXT: ret void
+;
+ %a0 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 0), align 4
+ %a1 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 1), align 4
+ %a2 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 2), align 4
+ %a3 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 3), align 4
+ %b0 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 0), align 4
+ %b1 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 1), align 4
+ %b2 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 2), align 4
+ %b3 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 3), align 4
+ %c0 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 0), align 4
+ %c1 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 1), align 4
+ %c2 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 2), align 4
+ %c3 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 3), align 4
+ %fmuladd0 = call float @llvm.fmuladd.f32(float %a0, float %b0, float %c0)
+ %fmuladd1 = call float @llvm.fmuladd.f32(float %a1, float %b1, float %c1)
+ %fmuladd2 = call float @llvm.fmuladd.f32(float %a2, float %b2, float %c2)
+ %fmuladd3 = call float @llvm.fmuladd.f32(float %a3, float %b3, float %c3)
+ store float %fmuladd0, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0), align 4
+ store float %fmuladd1, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1), align 4
+ store float %fmuladd2, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2), align 4
+ store float %fmuladd3, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3), align 4
+ ret void
+}
+
+define void @fmuladd_8f32() #0 {
+; SSE-LABEL: @fmuladd_8f32(
+; SSE-NEXT: [[TMP1:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @srcA32 to <4 x float>*), align 4
+; SSE-NEXT: [[TMP2:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 4) to <4 x float>*), align 4
+; SSE-NEXT: [[TMP3:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @srcB32 to <4 x float>*), align 4
+; SSE-NEXT: [[TMP4:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 4) to <4 x float>*), align 4
+; SSE-NEXT: [[TMP5:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @srcC32 to <4 x float>*), align 4
+; SSE-NEXT: [[TMP6:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 4) to <4 x float>*), align 4
+; SSE-NEXT: [[TMP7:%.*]] = call <4 x float> @llvm.fmuladd.v4f32(<4 x float> [[TMP1]], <4 x float> [[TMP3]], <4 x float> [[TMP5]])
+; SSE-NEXT: [[TMP8:%.*]] = call <4 x float> @llvm.fmuladd.v4f32(<4 x float> [[TMP2]], <4 x float> [[TMP4]], <4 x float> [[TMP6]])
+; SSE-NEXT: store <4 x float> [[TMP7]], <4 x float>* bitcast ([16 x float]* @dst32 to <4 x float>*), align 4
+; SSE-NEXT: store <4 x float> [[TMP8]], <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4) to <4 x float>*), align 4
+; SSE-NEXT: ret void
+;
+; AVX-LABEL: @fmuladd_8f32(
+; AVX-NEXT: [[TMP1:%.*]] = load <8 x float>, <8 x float>* bitcast ([16 x float]* @srcA32 to <8 x float>*), align 4
+; AVX-NEXT: [[TMP2:%.*]] = load <8 x float>, <8 x float>* bitcast ([16 x float]* @srcB32 to <8 x float>*), align 4
+; AVX-NEXT: [[TMP3:%.*]] = load <8 x float>, <8 x float>* bitcast ([16 x float]* @srcC32 to <8 x float>*), align 4
+; AVX-NEXT: [[TMP4:%.*]] = call <8 x float> @llvm.fmuladd.v8f32(<8 x float> [[TMP1]], <8 x float> [[TMP2]], <8 x float> [[TMP3]])
+; AVX-NEXT: store <8 x float> [[TMP4]], <8 x float>* bitcast ([16 x float]* @dst32 to <8 x float>*), align 4
+; AVX-NEXT: ret void
+;
+ %a0 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 0), align 4
+ %a1 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 1), align 4
+ %a2 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 2), align 4
+ %a3 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 3), align 4
+ %a4 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 4), align 4
+ %a5 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 5), align 4
+ %a6 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 6), align 4
+ %a7 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 7), align 4
+ %b0 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 0), align 4
+ %b1 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 1), align 4
+ %b2 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 2), align 4
+ %b3 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 3), align 4
+ %b4 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 4), align 4
+ %b5 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 5), align 4
+ %b6 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 6), align 4
+ %b7 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 7), align 4
+ %c0 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 0), align 4
+ %c1 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 1), align 4
+ %c2 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 2), align 4
+ %c3 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 3), align 4
+ %c4 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 4), align 4
+ %c5 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 5), align 4
+ %c6 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 6), align 4
+ %c7 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 7), align 4
+ %fmuladd0 = call float @llvm.fmuladd.f32(float %a0, float %b0, float %c0)
+ %fmuladd1 = call float @llvm.fmuladd.f32(float %a1, float %b1, float %c1)
+ %fmuladd2 = call float @llvm.fmuladd.f32(float %a2, float %b2, float %c2)
+ %fmuladd3 = call float @llvm.fmuladd.f32(float %a3, float %b3, float %c3)
+ %fmuladd4 = call float @llvm.fmuladd.f32(float %a4, float %b4, float %c4)
+ %fmuladd5 = call float @llvm.fmuladd.f32(float %a5, float %b5, float %c5)
+ %fmuladd6 = call float @llvm.fmuladd.f32(float %a6, float %b6, float %c6)
+ %fmuladd7 = call float @llvm.fmuladd.f32(float %a7, float %b7, float %c7)
+ store float %fmuladd0, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0), align 4
+ store float %fmuladd1, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1), align 4
+ store float %fmuladd2, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2), align 4
+ store float %fmuladd3, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3), align 4
+ store float %fmuladd4, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4), align 4
+ store float %fmuladd5, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 5), align 4
+ store float %fmuladd6, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 6), align 4
+ store float %fmuladd7, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 7), align 4
+ ret void
+}
+
+define void @fmuladd_16f32() #0 {
+; SSE-LABEL: @fmuladd_16f32(
+; SSE-NEXT: [[TMP1:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @srcA32 to <4 x float>*), align 4
+; SSE-NEXT: [[TMP2:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 4) to <4 x float>*), align 4
+; SSE-NEXT: [[TMP3:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 8) to <4 x float>*), align 4
+; SSE-NEXT: [[TMP4:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 12) to <4 x float>*), align 4
+; SSE-NEXT: [[TMP5:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @srcB32 to <4 x float>*), align 4
+; SSE-NEXT: [[TMP6:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 4) to <4 x float>*), align 4
+; SSE-NEXT: [[TMP7:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 8) to <4 x float>*), align 4
+; SSE-NEXT: [[TMP8:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 12) to <4 x float>*), align 4
+; SSE-NEXT: [[TMP9:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @srcC32 to <4 x float>*), align 4
+; SSE-NEXT: [[TMP10:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 4) to <4 x float>*), align 4
+; SSE-NEXT: [[TMP11:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 8) to <4 x float>*), align 4
+; SSE-NEXT: [[TMP12:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 12) to <4 x float>*), align 4
+; SSE-NEXT: [[TMP13:%.*]] = call <4 x float> @llvm.fmuladd.v4f32(<4 x float> [[TMP1]], <4 x float> [[TMP5]], <4 x float> [[TMP9]])
+; SSE-NEXT: [[TMP14:%.*]] = call <4 x float> @llvm.fmuladd.v4f32(<4 x float> [[TMP2]], <4 x float> [[TMP6]], <4 x float> [[TMP10]])
+; SSE-NEXT: [[TMP15:%.*]] = call <4 x float> @llvm.fmuladd.v4f32(<4 x float> [[TMP3]], <4 x float> [[TMP7]], <4 x float> [[TMP11]])
+; SSE-NEXT: [[TMP16:%.*]] = call <4 x float> @llvm.fmuladd.v4f32(<4 x float> [[TMP4]], <4 x float> [[TMP8]], <4 x float> [[TMP12]])
+; SSE-NEXT: store <4 x float> [[TMP13]], <4 x float>* bitcast ([16 x float]* @dst32 to <4 x float>*), align 4
+; SSE-NEXT: store <4 x float> [[TMP14]], <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4) to <4 x float>*), align 4
+; SSE-NEXT: store <4 x float> [[TMP15]], <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 8) to <4 x float>*), align 4
+; SSE-NEXT: store <4 x float> [[TMP16]], <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 12) to <4 x float>*), align 4
+; SSE-NEXT: ret void
+;
+; AVX256-LABEL: @fmuladd_16f32(
+; AVX256-NEXT: [[TMP1:%.*]] = load <8 x float>, <8 x float>* bitcast ([16 x float]* @srcA32 to <8 x float>*), align 4
+; AVX256-NEXT: [[TMP2:%.*]] = load <8 x float>, <8 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 8) to <8 x float>*), align 4
+; AVX256-NEXT: [[TMP3:%.*]] = load <8 x float>, <8 x float>* bitcast ([16 x float]* @srcB32 to <8 x float>*), align 4
+; AVX256-NEXT: [[TMP4:%.*]] = load <8 x float>, <8 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 8) to <8 x float>*), align 4
+; AVX256-NEXT: [[TMP5:%.*]] = load <8 x float>, <8 x float>* bitcast ([16 x float]* @srcC32 to <8 x float>*), align 4
+; AVX256-NEXT: [[TMP6:%.*]] = load <8 x float>, <8 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 8) to <8 x float>*), align 4
+; AVX256-NEXT: [[TMP7:%.*]] = call <8 x float> @llvm.fmuladd.v8f32(<8 x float> [[TMP1]], <8 x float> [[TMP3]], <8 x float> [[TMP5]])
+; AVX256-NEXT: [[TMP8:%.*]] = call <8 x float> @llvm.fmuladd.v8f32(<8 x float> [[TMP2]], <8 x float> [[TMP4]], <8 x float> [[TMP6]])
+; AVX256-NEXT: store <8 x float> [[TMP7]], <8 x float>* bitcast ([16 x float]* @dst32 to <8 x float>*), align 4
+; AVX256-NEXT: store <8 x float> [[TMP8]], <8 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 8) to <8 x float>*), align 4
+; AVX256-NEXT: ret void
+;
+; AVX512-LABEL: @fmuladd_16f32(
+; AVX512-NEXT: [[TMP1:%.*]] = load <16 x float>, <16 x float>* bitcast ([16 x float]* @srcA32 to <16 x float>*), align 4
+; AVX512-NEXT: [[TMP2:%.*]] = load <16 x float>, <16 x float>* bitcast ([16 x float]* @srcB32 to <16 x float>*), align 4
+; AVX512-NEXT: [[TMP3:%.*]] = load <16 x float>, <16 x float>* bitcast ([16 x float]* @srcC32 to <16 x float>*), align 4
+; AVX512-NEXT: [[TMP4:%.*]] = call <16 x float> @llvm.fmuladd.v16f32(<16 x float> [[TMP1]], <16 x float> [[TMP2]], <16 x float> [[TMP3]])
+; AVX512-NEXT: store <16 x float> [[TMP4]], <16 x float>* bitcast ([16 x float]* @dst32 to <16 x float>*), align 4
+; AVX512-NEXT: ret void
+;
+ %a0 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 0), align 4
+ %a1 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 1), align 4
+ %a2 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 2), align 4
+ %a3 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 3), align 4
+ %a4 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 4), align 4
+ %a5 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 5), align 4
+ %a6 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 6), align 4
+ %a7 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 7), align 4
+ %a8 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 8), align 4
+ %a9 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 9), align 4
+ %a10 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 10), align 4
+ %a11 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 11), align 4
+ %a12 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 12), align 4
+ %a13 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 13), align 4
+ %a14 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 14), align 4
+ %a15 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 15), align 4
+ %b0 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 0), align 4
+ %b1 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 1), align 4
+ %b2 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 2), align 4
+ %b3 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 3), align 4
+ %b4 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 4), align 4
+ %b5 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 5), align 4
+ %b6 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 6), align 4
+ %b7 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 7), align 4
+ %b8 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 8), align 4
+ %b9 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 9), align 4
+ %b10 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 10), align 4
+ %b11 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 11), align 4
+ %b12 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 12), align 4
+ %b13 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 13), align 4
+ %b14 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 14), align 4
+ %b15 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 15), align 4
+ %c0 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 0), align 4
+ %c1 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 1), align 4
+ %c2 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 2), align 4
+ %c3 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 3), align 4
+ %c4 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 4), align 4
+ %c5 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 5), align 4
+ %c6 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 6), align 4
+ %c7 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 7), align 4
+ %c8 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 8), align 4
+ %c9 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 9), align 4
+ %c10 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 10), align 4
+ %c11 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 11), align 4
+ %c12 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 12), align 4
+ %c13 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 13), align 4
+ %c14 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 14), align 4
+ %c15 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcC32, i32 0, i64 15), align 4
+ %fmuladd0 = call float @llvm.fmuladd.f32(float %a0 , float %b0 , float %c0 )
+ %fmuladd1 = call float @llvm.fmuladd.f32(float %a1 , float %b1 , float %c1 )
+ %fmuladd2 = call float @llvm.fmuladd.f32(float %a2 , float %b2 , float %c2 )
+ %fmuladd3 = call float @llvm.fmuladd.f32(float %a3 , float %b3 , float %c3 )
+ %fmuladd4 = call float @llvm.fmuladd.f32(float %a4 , float %b4 , float %c4 )
+ %fmuladd5 = call float @llvm.fmuladd.f32(float %a5 , float %b5 , float %c5 )
+ %fmuladd6 = call float @llvm.fmuladd.f32(float %a6 , float %b6 , float %c6 )
+ %fmuladd7 = call float @llvm.fmuladd.f32(float %a7 , float %b7 , float %c7 )
+ %fmuladd8 = call float @llvm.fmuladd.f32(float %a8 , float %b8 , float %c8 )
+ %fmuladd9 = call float @llvm.fmuladd.f32(float %a9 , float %b9 , float %c9 )
+ %fmuladd10 = call float @llvm.fmuladd.f32(float %a10, float %b10, float %c10)
+ %fmuladd11 = call float @llvm.fmuladd.f32(float %a11, float %b11, float %c11)
+ %fmuladd12 = call float @llvm.fmuladd.f32(float %a12, float %b12, float %c12)
+ %fmuladd13 = call float @llvm.fmuladd.f32(float %a13, float %b13, float %c13)
+ %fmuladd14 = call float @llvm.fmuladd.f32(float %a14, float %b14, float %c14)
+ %fmuladd15 = call float @llvm.fmuladd.f32(float %a15, float %b15, float %c15)
+ store float %fmuladd0 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0), align 4
+ store float %fmuladd1 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1), align 4
+ store float %fmuladd2 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2), align 4
+ store float %fmuladd3 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3), align 4
+ store float %fmuladd4 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4), align 4
+ store float %fmuladd5 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 5), align 4
+ store float %fmuladd6 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 6), align 4
+ store float %fmuladd7 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 7), align 4
+ store float %fmuladd8 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 8), align 4
+ store float %fmuladd9 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 9), align 4
+ store float %fmuladd10, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 10), align 4
+ store float %fmuladd11, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 11), align 4
+ store float %fmuladd12, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 12), align 4
+ store float %fmuladd13, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 13), align 4
+ store float %fmuladd14, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 14), align 4
+ store float %fmuladd15, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 15), align 4
+ ret void
+}
+
+attributes #0 = { nounwind }
More information about the llvm-commits
mailing list