AMDGPU: Expand testing of AMDGPUCodeGenPrepare fdiv handling

[arsenm]

Summary

• Switch to generated checks
• Use a different run line per denormal mode to reduce test duplication
• Add test coverage for rsqrt cases
• Add test coverage for repeated arcp denominator
• Fix the optnone test

[arsenm]

More vector tests

[arsenm]

More tests with flags

[arsenm]

Test known input classes for inf/nan/sub/zero

[arsenm]

Test some assumes

[arsenm]

More negated cases

[arsenm]

ef4a2b6096fe11a1f6562cea62b65c6e3b4d2cde

llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-fdiv.ll

-; RUN: opt -S -mtriple=amdgcn-- -amdgpu-codegenprepare %s | FileCheck %s
-; RUN: opt -S -amdgpu-codegenprepare %s | FileCheck -check-prefix=NOOP %s
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 3
+; RUN: opt -S -mtriple=amdgcn-- -passes=amdgpu-codegenprepare -denormal-fp-math-f32=ieee %s | FileCheck -check-prefixes=CHECK,IEEE %s
+; RUN: opt -S -mtriple=amdgcn-- -passes=amdgpu-codegenprepare -denormal-fp-math-f32=dynamic %s | FileCheck -check-prefixes=CHECK,IEEE %s
+; RUN: opt -S -mtriple=amdgcn-- -passes=amdgpu-codegenprepare -denormal-fp-math-f32=preserve-sign %s | FileCheck -check-prefixes=CHECK,DAZ %s
+
 ; Make sure this doesn't crash with no triple
+; TODO: Delete when old PM deleted
+; RUN: opt -amdgpu-codegenprepare -disable-output %s
+
 
-; NOOP-LABEL: @noop_fdiv_fpmath(
-; NOOP: %md.25ulp = fdiv float %a, %b, !fpmath !0
-define amdgpu_kernel void @noop_fdiv_fpmath(ptr addrspace(1) %out, float %a, float %b) {
+define amdgpu_kernel void @noop_fdiv_fpmath(ptr addrspace(1) %out, float %a, float %b) #0 {
+; CHECK-LABEL: define amdgpu_kernel void @noop_fdiv_fpmath
+; CHECK-SAME: (ptr addrspace(1) [[OUT:%.*]], float [[A:%.*]], float [[B:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT:    [[MD_25ULP:%.*]] = fdiv float [[A]], [[B]], !fpmath [[META0:![0-9]+]]
+; CHECK-NEXT:    store volatile float [[MD_25ULP]], ptr addrspace(1) [[OUT]], align 4
+; CHECK-NEXT:    ret void
+;
   %md.25ulp = fdiv float %a, %b, !fpmath !0
-  store volatile float %md.25ulp, ptr addrspace(1) %out
+  store volatile float %md.25ulp, ptr addrspace(1) %out, align 4
   ret void
 }
 
-; CHECK-LABEL: @fdiv_fpmath(
-; CHECK: %no.md = fdiv float %a, %b{{$}}
-; CHECK: %md.half.ulp = fdiv float %a, %b
-; CHECK: %md.1ulp = fdiv float %a, %b
-; CHECK: %md.25ulp = call float @llvm.amdgcn.fdiv.fast(float %a, float %b)
-; CHECK: %md.3ulp = call float @llvm.amdgcn.fdiv.fast(float %a, float %b)
-; CHECK: %[[FAST_RCP:[0-9]+]] = call fast float @llvm.amdgcn.rcp.f32(float %b)
-; CHECK: %fast.md.25ulp = fmul fast float %a, %[[FAST_RCP]]
-; CHECK: %[[AFN_RCP:[0-9]+]] = call afn float @llvm.amdgcn.rcp.f32(float %b)
-; CHECK: afn.md.25ulp = fmul afn float %a, %[[AFN_RCP]]
-define amdgpu_kernel void @fdiv_fpmath(ptr addrspace(1) %out, float %a, float %b) #1 {
+define amdgpu_kernel void @fdiv_fpmath_f32(ptr addrspace(1) %out, float %a, float %b) {
+; IEEE-LABEL: define amdgpu_kernel void @fdiv_fpmath_f32
+; IEEE-SAME: (ptr addrspace(1) [[OUT:%.*]], float [[A:%.*]], float [[B:%.*]]) #[[ATTR1:[0-9]+]] {
+; IEEE-NEXT:    [[NO_MD:%.*]] = fdiv float [[A]], [[B]]
+; IEEE-NEXT:    store volatile float [[NO_MD]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    [[MD_HALF_ULP:%.*]] = fdiv float [[A]], [[B]], !fpmath [[META1:![0-9]+]]
+; IEEE-NEXT:    store volatile float [[MD_HALF_ULP]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    [[MD_1ULP:%.*]] = fdiv float [[A]], [[B]], !fpmath [[META2:![0-9]+]]
+; IEEE-NEXT:    store volatile float [[MD_1ULP]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    [[MD_25ULP:%.*]] = fdiv float [[A]], [[B]], !fpmath [[META0]]
+; IEEE-NEXT:    store volatile float [[MD_25ULP]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    [[MD_3ULP:%.*]] = fdiv float [[A]], [[B]], !fpmath [[META3:![0-9]+]]
+; IEEE-NEXT:    store volatile float [[MD_3ULP]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    [[TMP1:%.*]] = call fast float @llvm.amdgcn.rcp.f32(float [[B]])
+; IEEE-NEXT:    [[FAST_MD_25ULP:%.*]] = fmul fast float [[A]], [[TMP1]]
+; IEEE-NEXT:    store volatile float [[FAST_MD_25ULP]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    [[TMP2:%.*]] = call afn float @llvm.amdgcn.rcp.f32(float [[B]])
+; IEEE-NEXT:    [[AFN_MD_25ULP:%.*]] = fmul afn float [[A]], [[TMP2]]
+; IEEE-NEXT:    store volatile float [[AFN_MD_25ULP]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    ret void
+;
+; DAZ-LABEL: define amdgpu_kernel void @fdiv_fpmath_f32
+; DAZ-SAME: (ptr addrspace(1) [[OUT:%.*]], float [[A:%.*]], float [[B:%.*]]) #[[ATTR1:[0-9]+]] {
+; DAZ-NEXT:    [[NO_MD:%.*]] = fdiv float [[A]], [[B]]
+; DAZ-NEXT:    store volatile float [[NO_MD]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    [[MD_HALF_ULP:%.*]] = fdiv float [[A]], [[B]], !fpmath [[META1:![0-9]+]]
+; DAZ-NEXT:    store volatile float [[MD_HALF_ULP]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    [[MD_1ULP:%.*]] = fdiv float [[A]], [[B]], !fpmath [[META2:![0-9]+]]
+; DAZ-NEXT:    store volatile float [[MD_1ULP]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    [[MD_25ULP:%.*]] = call float @llvm.amdgcn.fdiv.fast(float [[A]], float [[B]])
+; DAZ-NEXT:    store volatile float [[MD_25ULP]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    [[MD_3ULP:%.*]] = call float @llvm.amdgcn.fdiv.fast(float [[A]], float [[B]])
+; DAZ-NEXT:    store volatile float [[MD_3ULP]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    [[TMP1:%.*]] = call fast float @llvm.amdgcn.rcp.f32(float [[B]])
+; DAZ-NEXT:    [[FAST_MD_25ULP:%.*]] = fmul fast float [[A]], [[TMP1]]
+; DAZ-NEXT:    store volatile float [[FAST_MD_25ULP]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    [[TMP2:%.*]] = call afn float @llvm.amdgcn.rcp.f32(float [[B]])
+; DAZ-NEXT:    [[AFN_MD_25ULP:%.*]] = fmul afn float [[A]], [[TMP2]]
+; DAZ-NEXT:    store volatile float [[AFN_MD_25ULP]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    ret void
+;
   %no.md = fdiv float %a, %b
-  store volatile float %no.md, ptr addrspace(1) %out
-
+  store volatile float %no.md, ptr addrspace(1) %out, align 4
   %md.half.ulp = fdiv float %a, %b, !fpmath !1
-  store volatile float %md.half.ulp, ptr addrspace(1) %out
-
+  store volatile float %md.half.ulp, ptr addrspace(1) %out, align 4
   %md.1ulp = fdiv float %a, %b, !fpmath !2
-  store volatile float %md.1ulp, ptr addrspace(1) %out
-
+  store volatile float %md.1ulp, ptr addrspace(1) %out, align 4
   %md.25ulp = fdiv float %a, %b, !fpmath !0
-  store volatile float %md.25ulp, ptr addrspace(1) %out
-
+  store volatile float %md.25ulp, ptr addrspace(1) %out, align 4
   %md.3ulp = fdiv float %a, %b, !fpmath !3
-  store volatile float %md.3ulp, ptr addrspace(1) %out
-
+  store volatile float %md.3ulp, ptr addrspace(1) %out, align 4
   %fast.md.25ulp = fdiv fast float %a, %b, !fpmath !0
-  store volatile float %fast.md.25ulp, ptr addrspace(1) %out
-
+  store volatile float %fast.md.25ulp, ptr addrspace(1) %out, align 4
   %afn.md.25ulp = fdiv afn float %a, %b, !fpmath !0
-  store volatile float %afn.md.25ulp, ptr addrspace(1) %out
+  store volatile float %afn.md.25ulp, ptr addrspace(1) %out, align 4
+  ret void
+}
 
+define amdgpu_kernel void @rcp_fdiv_f32_fpmath(ptr addrspace(1) %out, float %x) {
+; IEEE-LABEL: define amdgpu_kernel void @rcp_fdiv_f32_fpmath
+; IEEE-SAME: (ptr addrspace(1) [[OUT:%.*]], float [[X:%.*]]) #[[ATTR1]] {
+; IEEE-NEXT:    [[NO_MD:%.*]] = fdiv float 1.000000e+00, [[X]]
+; IEEE-NEXT:    store volatile float [[NO_MD]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    [[MD_1ULP:%.*]] = fdiv float 1.000000e+00, [[X]], !fpmath [[META2]]
+; IEEE-NEXT:    store volatile float [[MD_1ULP]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    [[MD_25ULP:%.*]] = call float @llvm.amdgcn.fdiv.fast(float 1.000000e+00, float [[X]])
+; IEEE-NEXT:    store volatile float [[MD_25ULP]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    [[MD_HALF_ULP:%.*]] = fdiv float 1.000000e+00, [[X]], !fpmath [[META1]]
+; IEEE-NEXT:    store volatile float [[MD_HALF_ULP]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    [[AFN_NO_MD:%.*]] = call afn float @llvm.amdgcn.rcp.f32(float [[X]])
+; IEEE-NEXT:    store volatile float [[AFN_NO_MD]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    [[AFN_25ULP:%.*]] = call afn float @llvm.amdgcn.rcp.f32(float [[X]])
+; IEEE-NEXT:    store volatile float [[AFN_25ULP]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    [[FAST_NO_MD:%.*]] = call fast float @llvm.amdgcn.rcp.f32(float [[X]])
+; IEEE-NEXT:    store volatile float [[FAST_NO_MD]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    [[FAST_25ULP:%.*]] = call fast float @llvm.amdgcn.rcp.f32(float [[X]])
+; IEEE-NEXT:    store volatile float [[FAST_25ULP]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    ret void
+;
+; DAZ-LABEL: define amdgpu_kernel void @rcp_fdiv_f32_fpmath
+; DAZ-SAME: (ptr addrspace(1) [[OUT:%.*]], float [[X:%.*]]) #[[ATTR1]] {
+; DAZ-NEXT:    [[NO_MD:%.*]] = fdiv float 1.000000e+00, [[X]]
+; DAZ-NEXT:    store volatile float [[NO_MD]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    [[MD_1ULP:%.*]] = call float @llvm.amdgcn.rcp.f32(float [[X]])
+; DAZ-NEXT:    store volatile float [[MD_1ULP]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    [[MD_25ULP:%.*]] = call float @llvm.amdgcn.rcp.f32(float [[X]])
+; DAZ-NEXT:    store volatile float [[MD_25ULP]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    [[MD_HALF_ULP:%.*]] = fdiv float 1.000000e+00, [[X]], !fpmath [[META1]]
+; DAZ-NEXT:    store volatile float [[MD_HALF_ULP]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    [[AFN_NO_MD:%.*]] = call afn float @llvm.amdgcn.rcp.f32(float [[X]])
+; DAZ-NEXT:    store volatile float [[AFN_NO_MD]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    [[AFN_25ULP:%.*]] = call afn float @llvm.amdgcn.rcp.f32(float [[X]])
+; DAZ-NEXT:    store volatile float [[AFN_25ULP]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    [[FAST_NO_MD:%.*]] = call fast float @llvm.amdgcn.rcp.f32(float [[X]])
+; DAZ-NEXT:    store volatile float [[FAST_NO_MD]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    [[FAST_25ULP:%.*]] = call fast float @llvm.amdgcn.rcp.f32(float [[X]])
+; DAZ-NEXT:    store volatile float [[FAST_25ULP]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    ret void
+;
+  %no.md = fdiv float 1.000000e+00, %x
+  store volatile float %no.md, ptr addrspace(1) %out, align 4
+  %md.1ulp = fdiv float 1.000000e+00, %x, !fpmath !2
+  store volatile float %md.1ulp, ptr addrspace(1) %out, align 4
+  %md.25ulp = fdiv float 1.000000e+00, %x, !fpmath !0
+  store volatile float %md.25ulp, ptr addrspace(1) %out, align 4
+  %md.half.ulp = fdiv float 1.000000e+00, %x, !fpmath !1
+  store volatile float %md.half.ulp, ptr addrspace(1) %out, align 4
+  %afn.no.md = fdiv afn float 1.000000e+00, %x
+  store volatile float %afn.no.md, ptr addrspace(1) %out, align 4
+  %afn.25ulp = fdiv afn float 1.000000e+00, %x, !fpmath !0
+  store volatile float %afn.25ulp, ptr addrspace(1) %out, align 4
+  %fast.no.md = fdiv fast float 1.000000e+00, %x
+  store volatile float %fast.no.md, ptr addrspace(1) %out, align 4
+  %fast.25ulp = fdiv fast float 1.000000e+00, %x, !fpmath !0
+  store volatile float %fast.25ulp, ptr addrspace(1) %out, align 4
   ret void
 }
 
-; CHECK-LABEL: @rcp_fdiv_fpmath(
-; CHECK: %no.md = fdiv float 1.000000e+00, %x{{$}}
-; CHECK: %md.25ulp = call float @llvm.amdgcn.rcp.f32(float %x)
-; CHECK: %md.half.ulp = fdiv float 1.000000e+00, %x
-; CHECK: %afn.no.md = call afn float @llvm.amdgcn.rcp.f32(float %x)
-; CHECK: %afn.25ulp = call afn float @llvm.amdgcn.rcp.f32(float %x)
-; CHECK: %fast.no.md = call fast float @llvm.amdgcn.rcp.f32(float %x)
-; CHECK: %fast.25ulp = call fast float @llvm.amdgcn.rcp.f32(float %x)
-define amdgpu_kernel void @rcp_fdiv_fpmath(ptr addrspace(1) %out, float %x) #1 {
-  %no.md = fdiv float 1.0, %x
-  store volatile float %no.md, ptr addrspace(1) %out
-
-  %md.25ulp = fdiv float 1.0, %x, !fpmath !0
-  store volatile float %md.25ulp, ptr addrspace(1) %out
-
-  %md.half.ulp = fdiv float 1.0, %x, !fpmath !1
-  store volatile float %md.half.ulp, ptr addrspace(1) %out
-
-  %afn.no.md = fdiv afn float 1.0, %x
-  store volatile float %afn.no.md, ptr addrspace(1) %out
-
-  %afn.25ulp = fdiv afn float 1.0, %x, !fpmath !0
-  store volatile float %afn.25ulp, ptr addrspace(1) %out
-
-  %fast.no.md = fdiv fast float 1.0, %x
-  store volatile float %fast.no.md, ptr addrspace(1) %out
-
-  %fast.25ulp = fdiv fast float 1.0, %x, !fpmath !0
-  store volatile float %fast.25ulp, ptr addrspace(1) %out
-
-  ret void
-}
-
-; CHECK-LABEL: @fdiv_fpmath_vector(
-; CHECK: %[[NO_A0:[0-9]+]] = extractelement <2 x float> %a, i64 0
-; CHECK: %[[NO_B0:[0-9]+]] = extractelement <2 x float> %b, i64 0
-; CHECK: %[[NO_FDIV0:[0-9]+]] = fdiv float %[[NO_A0]], %[[NO_B0]]
-; CHECK: %[[NO_INS0:[0-9]+]] = insertelement <2 x float> poison, float %[[NO_FDIV0]], i64 0
-; CHECK: %[[NO_A1:[0-9]+]] = extractelement <2 x float> %a, i64 1
-; CHECK: %[[NO_B1:[0-9]+]] = extractelement <2 x float> %b, i64 1
-; CHECK: %[[NO_FDIV1:[0-9]+]] = fdiv float %[[NO_A1]], %[[NO_B1]]
-; CHECK: %no.md = insertelement <2 x float> %[[NO_INS0]], float %[[NO_FDIV1]], i64 1
-; CHECK: store volatile <2 x float> %no.md, ptr addrspace(1) %out
-
-; CHECK: %[[HALF_A0:[0-9]+]] = extractelement <2 x float> %a, i64 0
-; CHECK: %[[HALF_B0:[0-9]+]] = extractelement <2 x float> %b, i64 0
-; CHECK: %[[HALF_FDIV0:[0-9]+]] = fdiv float %[[HALF_A0]], %[[HALF_B0]]
-; CHECK: %[[HALF_INS0:[0-9]+]] = insertelement <2 x float> poison, float %[[HALF_FDIV0]], i64 0
-; CHECK: %[[HALF_A1:[0-9]+]] = extractelement <2 x float> %a, i64 1
-; CHECK: %[[HALF_B1:[0-9]+]] = extractelement <2 x float> %b, i64 1
-; CHECK: %[[HALF_FDIV1:[0-9]+]] = fdiv float %[[HALF_A1]], %[[HALF_B1]]
-; CHECK: %md.half.ulp = insertelement <2 x float> %[[HALF_INS0]], float %[[HALF_FDIV1]], i64 1
-; CHECK: store volatile <2 x float> %md.half.ulp, ptr addrspace(1) %out
-
-; CHECK: %[[ONE_A0:[0-9]+]] = extractelement <2 x float> %a, i64 0
-; CHECK: %[[ONE_B0:[0-9]+]] = extractelement <2 x float> %b, i64 0
-; CHECK: %[[ONE_FDIV0:[0-9]+]] = fdiv float %[[ONE_A0]], %[[ONE_B0]]
-; CHECK: %[[ONE_INS0:[0-9]+]] = insertelement <2 x float> poison, float %[[ONE_FDIV0]], i64 0
-; CHECK: %[[ONE_A1:[0-9]+]] = extractelement <2 x float> %a, i64 1
-; CHECK: %[[ONE_B1:[0-9]+]] = extractelement <2 x float> %b, i64 1
-; CHECK: %[[ONE_FDIV1:[0-9]+]] = fdiv float %[[ONE_A1]], %[[ONE_B1]]
-; CHECK: %md.1ulp = insertelement <2 x float> %[[ONE_INS0]], float %[[ONE_FDIV1]], i64 1
-; CHECK: store volatile <2 x float> %md.1ulp, ptr addrspace(1) %out
-
-; CHECK: %[[A0:[0-9]+]] = extractelement <2 x float> %a, i64 0
-; CHECK: %[[B0:[0-9]+]] = extractelement <2 x float> %b, i64 0
-; CHECK: %[[FDIV0:[0-9]+]] = call float @llvm.amdgcn.fdiv.fast(float %[[A0]], float %[[B0]])
-; CHECK: %[[INS0:[0-9]+]] = insertelement <2 x float> poison, float %[[FDIV0]], i64 0
-; CHECK: %[[A1:[0-9]+]] = extractelement <2 x float> %a, i64 1
-; CHECK: %[[B1:[0-9]+]] = extractelement <2 x float> %b, i64 1
-; CHECK: %[[FDIV1:[0-9]+]] = call float @llvm.amdgcn.fdiv.fast(float %[[A1]], float %[[B1]])
-; CHECK: %md.25ulp = insertelement <2 x float> %[[INS0]], float %[[FDIV1]], i64 1
-define amdgpu_kernel void @fdiv_fpmath_vector(ptr addrspace(1) %out, <2 x float> %a, <2 x float> %b) #1 {
+define amdgpu_kernel void @fdiv_fpmath_f32_vector(ptr addrspace(1) %out, <2 x float> %a, <2 x float> %b) {
+; IEEE-LABEL: define amdgpu_kernel void @fdiv_fpmath_f32_vector
+; IEEE-SAME: (ptr addrspace(1) [[OUT:%.*]], <2 x float> [[A:%.*]], <2 x float> [[B:%.*]]) #[[ATTR1]] {
+; IEEE-NEXT:    [[TMP1:%.*]] = extractelement <2 x float> [[A]], i64 0
+; IEEE-NEXT:    [[TMP2:%.*]] = extractelement <2 x float> [[B]], i64 0
+; IEEE-NEXT:    [[TMP3:%.*]] = fdiv float [[TMP1]], [[TMP2]]
+; IEEE-NEXT:    [[TMP4:%.*]] = insertelement <2 x float> poison, float [[TMP3]], i64 0
+; IEEE-NEXT:    [[TMP5:%.*]] = extractelement <2 x float> [[A]], i64 1
+; IEEE-NEXT:    [[TMP6:%.*]] = extractelement <2 x float> [[B]], i64 1
+; IEEE-NEXT:    [[TMP7:%.*]] = fdiv float [[TMP5]], [[TMP6]]
+; IEEE-NEXT:    [[NO_MD:%.*]] = insertelement <2 x float> [[TMP4]], float [[TMP7]], i64 1
+; IEEE-NEXT:    store volatile <2 x float> [[NO_MD]], ptr addrspace(1) [[OUT]], align 8
+; IEEE-NEXT:    [[TMP8:%.*]] = extractelement <2 x float> [[A]], i64 0
+; IEEE-NEXT:    [[TMP9:%.*]] = extractelement <2 x float> [[B]], i64 0
+; IEEE-NEXT:    [[TMP10:%.*]] = fdiv float [[TMP8]], [[TMP9]]
+; IEEE-NEXT:    [[TMP11:%.*]] = insertelement <2 x float> poison, float [[TMP10]], i64 0
+; IEEE-NEXT:    [[TMP12:%.*]] = extractelement <2 x float> [[A]], i64 1
+; IEEE-NEXT:    [[TMP13:%.*]] = extractelement <2 x float> [[B]], i64 1
+; IEEE-NEXT:    [[TMP14:%.*]] = fdiv float [[TMP12]], [[TMP13]]
+; IEEE-NEXT:    [[MD_HALF_ULP:%.*]] = insertelement <2 x float> [[TMP11]], float [[TMP14]], i64 1
+; IEEE-NEXT:    store volatile <2 x float> [[MD_HALF_ULP]], ptr addrspace(1) [[OUT]], align 8
+; IEEE-NEXT:    [[TMP15:%.*]] = extractelement <2 x float> [[A]], i64 0
+; IEEE-NEXT:    [[TMP16:%.*]] = extractelement <2 x float> [[B]], i64 0
+; IEEE-NEXT:    [[TMP17:%.*]] = fdiv float [[TMP15]], [[TMP16]]
+; IEEE-NEXT:    [[TMP18:%.*]] = insertelement <2 x float> poison, float [[TMP17]], i64 0
+; IEEE-NEXT:    [[TMP19:%.*]] = extractelement <2 x float> [[A]], i64 1
+; IEEE-NEXT:    [[TMP20:%.*]] = extractelement <2 x float> [[B]], i64 1
+; IEEE-NEXT:    [[TMP21:%.*]] = fdiv float [[TMP19]], [[TMP20]]
+; IEEE-NEXT:    [[MD_1ULP:%.*]] = insertelement <2 x float> [[TMP18]], float [[TMP21]], i64 1
+; IEEE-NEXT:    store volatile <2 x float> [[MD_1ULP]], ptr addrspace(1) [[OUT]], align 8
+; IEEE-NEXT:    [[TMP22:%.*]] = extractelement <2 x float> [[A]], i64 0
+; IEEE-NEXT:    [[TMP23:%.*]] = extractelement <2 x float> [[B]], i64 0
+; IEEE-NEXT:    [[TMP24:%.*]] = fdiv float [[TMP22]], [[TMP23]]
+; IEEE-NEXT:    [[TMP25:%.*]] = insertelement <2 x float> poison, float [[TMP24]], i64 0
+; IEEE-NEXT:    [[TMP26:%.*]] = extractelement <2 x float> [[A]], i64 1
+; IEEE-NEXT:    [[TMP27:%.*]] = extractelement <2 x float> [[B]], i64 1
+; IEEE-NEXT:    [[TMP28:%.*]] = fdiv float [[TMP26]], [[TMP27]]
+; IEEE-NEXT:    [[MD_25ULP:%.*]] = insertelement <2 x float> [[TMP25]], float [[TMP28]], i64 1
+; IEEE-NEXT:    store volatile <2 x float> [[MD_25ULP]], ptr addrspace(1) [[OUT]], align 8
+; IEEE-NEXT:    ret void
+;
+; DAZ-LABEL: define amdgpu_kernel void @fdiv_fpmath_f32_vector
+; DAZ-SAME: (ptr addrspace(1) [[OUT:%.*]], <2 x float> [[A:%.*]], <2 x float> [[B:%.*]]) #[[ATTR1]] {
+; DAZ-NEXT:    [[TMP1:%.*]] = extractelement <2 x float> [[A]], i64 0
+; DAZ-NEXT:    [[TMP2:%.*]] = extractelement <2 x float> [[B]], i64 0
+; DAZ-NEXT:    [[TMP3:%.*]] = fdiv float [[TMP1]], [[TMP2]]
+; DAZ-NEXT:    [[TMP4:%.*]] = insertelement <2 x float> poison, float [[TMP3]], i64 0
+; DAZ-NEXT:    [[TMP5:%.*]] = extractelement <2 x float> [[A]], i64 1
+; DAZ-NEXT:    [[TMP6:%.*]] = extractelement <2 x float> [[B]], i64 1
+; DAZ-NEXT:    [[TMP7:%.*]] = fdiv float [[TMP5]], [[TMP6]]
+; DAZ-NEXT:    [[NO_MD:%.*]] = insertelement <2 x float> [[TMP4]], float [[TMP7]], i64 1
+; DAZ-NEXT:    store volatile <2 x float> [[NO_MD]], ptr addrspace(1) [[OUT]], align 8
+; DAZ-NEXT:    [[TMP8:%.*]] = extractelement <2 x float> [[A]], i64 0
+; DAZ-NEXT:    [[TMP9:%.*]] = extractelement <2 x float> [[B]], i64 0
+; DAZ-NEXT:    [[TMP10:%.*]] = fdiv float [[TMP8]], [[TMP9]]
+; DAZ-NEXT:    [[TMP11:%.*]] = insertelement <2 x float> poison, float [[TMP10]], i64 0
+; DAZ-NEXT:    [[TMP12:%.*]] = extractelement <2 x float> [[A]], i64 1
+; DAZ-NEXT:    [[TMP13:%.*]] = extractelement <2 x float> [[B]], i64 1
+; DAZ-NEXT:    [[TMP14:%.*]] = fdiv float [[TMP12]], [[TMP13]]
+; DAZ-NEXT:    [[MD_HALF_ULP:%.*]] = insertelement <2 x float> [[TMP11]], float [[TMP14]], i64 1
+; DAZ-NEXT:    store volatile <2 x float> [[MD_HALF_ULP]], ptr addrspace(1) [[OUT]], align 8
+; DAZ-NEXT:    [[TMP15:%.*]] = extractelement <2 x float> [[A]], i64 0
+; DAZ-NEXT:    [[TMP16:%.*]] = extractelement <2 x float> [[B]], i64 0
+; DAZ-NEXT:    [[TMP17:%.*]] = fdiv float [[TMP15]], [[TMP16]]
+; DAZ-NEXT:    [[TMP18:%.*]] = insertelement <2 x float> poison, float [[TMP17]], i64 0
+; DAZ-NEXT:    [[TMP19:%.*]] = extractelement <2 x float> [[A]], i64 1
+; DAZ-NEXT:    [[TMP20:%.*]] = extractelement <2 x float> [[B]], i64 1
+; DAZ-NEXT:    [[TMP21:%.*]] = fdiv float [[TMP19]], [[TMP20]]
+; DAZ-NEXT:    [[MD_1ULP:%.*]] = insertelement <2 x float> [[TMP18]], float [[TMP21]], i64 1
+; DAZ-NEXT:    store volatile <2 x float> [[MD_1ULP]], ptr addrspace(1) [[OUT]], align 8
+; DAZ-NEXT:    [[TMP22:%.*]] = extractelement <2 x float> [[A]], i64 0
+; DAZ-NEXT:    [[TMP23:%.*]] = extractelement <2 x float> [[B]], i64 0
+; DAZ-NEXT:    [[TMP24:%.*]] = call float @llvm.amdgcn.fdiv.fast(float [[TMP22]], float [[TMP23]])
+; DAZ-NEXT:    [[TMP25:%.*]] = insertelement <2 x float> poison, float [[TMP24]], i64 0
+; DAZ-NEXT:    [[TMP26:%.*]] = extractelement <2 x float> [[A]], i64 1
+; DAZ-NEXT:    [[TMP27:%.*]] = extractelement <2 x float> [[B]], i64 1
+; DAZ-NEXT:    [[TMP28:%.*]] = call float @llvm.amdgcn.fdiv.fast(float [[TMP26]], float [[TMP27]])
+; DAZ-NEXT:    [[MD_25ULP:%.*]] = insertelement <2 x float> [[TMP25]], float [[TMP28]], i64 1
+; DAZ-NEXT:    store volatile <2 x float> [[MD_25ULP]], ptr addrspace(1) [[OUT]], align 8
+; DAZ-NEXT:    ret void
+;
   %no.md = fdiv <2 x float> %a, %b
-  store volatile <2 x float> %no.md, ptr addrspace(1) %out
-
+  store volatile <2 x float> %no.md, ptr addrspace(1) %out, align 8
   %md.half.ulp = fdiv <2 x float> %a, %b, !fpmath !1
-  store volatile <2 x float> %md.half.ulp, ptr addrspace(1) %out
-
+  store volatile <2 x float> %md.half.ulp, ptr addrspace(1) %out, align 8
   %md.1ulp = fdiv <2 x float> %a, %b, !fpmath !2
-  store volatile <2 x float> %md.1ulp, ptr addrspace(1) %out
-
+  store volatile <2 x float> %md.1ulp, ptr addrspace(1) %out, align 8
   %md.25ulp = fdiv <2 x float> %a, %b, !fpmath !0
-  store volatile <2 x float> %md.25ulp, ptr addrspace(1) %out
+  store volatile <2 x float> %md.25ulp, ptr addrspace(1) %out, align 8
+  ret void
+}
 
+define amdgpu_kernel void @rcp_fdiv_f32_vector_fpmath(ptr addrspace(1) %out, <2 x float> %x) {
+; CHECK-LABEL: define amdgpu_kernel void @rcp_fdiv_f32_vector_fpmath
+; CHECK-SAME: (ptr addrspace(1) [[OUT:%.*]], <2 x float> [[X:%.*]]) #[[ATTR1:[0-9]+]] {
+; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <2 x float> [[X]], i64 0
+; CHECK-NEXT:    [[TMP2:%.*]] = fdiv float 1.000000e+00, [[TMP1]]
+; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <2 x float> poison, float [[TMP2]], i64 0
+; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x float> [[X]], i64 1
+; CHECK-NEXT:    [[TMP5:%.*]] = fdiv float 1.000000e+00, [[TMP4]]
+; CHECK-NEXT:    [[NO_MD:%.*]] = insertelement <2 x float> [[TMP3]], float [[TMP5]], i64 1
+; CHECK-NEXT:    store volatile <2 x float> [[NO_MD]], ptr addrspace(1) [[OUT]], align 8
+; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <2 x float> [[X]], i64 0
+; CHECK-NEXT:    [[TMP7:%.*]] = fdiv float 1.000000e+00, [[TMP6]]
+; CHECK-NEXT:    [[TMP8:%.*]] = insertelement <2 x float> poison, float [[TMP7]], i64 0
+; CHECK-NEXT:    [[TMP9:%.*]] = extractelement <2 x float> [[X]], i64 1
+; CHECK-NEXT:    [[TMP10:%.*]] = fdiv float 1.000000e+00, [[TMP9]]
+; CHECK-NEXT:    [[MD_HALF_ULP:%.*]] = insertelement <2 x float> [[TMP8]], float [[TMP10]], i64 1
+; CHECK-NEXT:    store volatile <2 x float> [[MD_HALF_ULP]], ptr addrspace(1) [[OUT]], align 8
+; CHECK-NEXT:    [[TMP11:%.*]] = extractelement <2 x float> [[X]], i64 0
+; CHECK-NEXT:    [[TMP12:%.*]] = call afn float @llvm.amdgcn.rcp.f32(float [[TMP11]])
+; CHECK-NEXT:    [[TMP13:%.*]] = insertelement <2 x float> poison, float [[TMP12]], i64 0
+; CHECK-NEXT:    [[TMP14:%.*]] = extractelement <2 x float> [[X]], i64 1
+; CHECK-NEXT:    [[TMP15:%.*]] = call afn float @llvm.amdgcn.rcp.f32(float [[TMP14]])
+; CHECK-NEXT:    [[AFN_NO_MD:%.*]] = insertelement <2 x float> [[TMP13]], float [[TMP15]], i64 1
+; CHECK-NEXT:    store volatile <2 x float> [[AFN_NO_MD]], ptr addrspace(1) [[OUT]], align 8
+; CHECK-NEXT:    [[TMP16:%.*]] = extractelement <2 x float> [[X]], i64 0
+; CHECK-NEXT:    [[TMP17:%.*]] = call fast float @llvm.amdgcn.rcp.f32(float [[TMP16]])
+; CHECK-NEXT:    [[TMP18:%.*]] = insertelement <2 x float> poison, float [[TMP17]], i64 0
+; CHECK-NEXT:    [[TMP19:%.*]] = extractelement <2 x float> [[X]], i64 1
+; CHECK-NEXT:    [[TMP20:%.*]] = call fast float @llvm.amdgcn.rcp.f32(float [[TMP19]])
+; CHECK-NEXT:    [[FAST_NO_MD:%.*]] = insertelement <2 x float> [[TMP18]], float [[TMP20]], i64 1
+; CHECK-NEXT:    store volatile <2 x float> [[FAST_NO_MD]], ptr addrspace(1) [[OUT]], align 8
+; CHECK-NEXT:    [[TMP21:%.*]] = extractelement <2 x float> [[X]], i64 0
+; CHECK-NEXT:    [[TMP22:%.*]] = call afn float @llvm.amdgcn.rcp.f32(float [[TMP21]])
+; CHECK-NEXT:    [[TMP23:%.*]] = insertelement <2 x float> poison, float [[TMP22]], i64 0
+; CHECK-NEXT:    [[TMP24:%.*]] = extractelement <2 x float> [[X]], i64 1
+; CHECK-NEXT:    [[TMP25:%.*]] = call afn float @llvm.amdgcn.rcp.f32(float [[TMP24]])
+; CHECK-NEXT:    [[AFN_25ULP:%.*]] = insertelement <2 x float> [[TMP23]], float [[TMP25]], i64 1
+; CHECK-NEXT:    store volatile <2 x float> [[AFN_25ULP]], ptr addrspace(1) [[OUT]], align 8
+; CHECK-NEXT:    [[TMP26:%.*]] = extractelement <2 x float> [[X]], i64 0
+; CHECK-NEXT:    [[TMP27:%.*]] = call fast float @llvm.amdgcn.rcp.f32(float [[TMP26]])
+; CHECK-NEXT:    [[TMP28:%.*]] = insertelement <2 x float> poison, float [[TMP27]], i64 0
+; CHECK-NEXT:    [[TMP29:%.*]] = extractelement <2 x float> [[X]], i64 1
+; CHECK-NEXT:    [[TMP30:%.*]] = call fast float @llvm.amdgcn.rcp.f32(float [[TMP29]])
+; CHECK-NEXT:    [[FAST_25ULP:%.*]] = insertelement <2 x float> [[TMP28]], float [[TMP30]], i64 1
+; CHECK-NEXT:    store volatile <2 x float> [[FAST_25ULP]], ptr addrspace(1) [[OUT]], align 8
+; CHECK-NEXT:    ret void
+;
+  %no.md = fdiv <2 x float> <float 1.000000e+00, float 1.000000e+00>, %x
+  store volatile <2 x float> %no.md, ptr addrspace(1) %out, align 8
+  %md.half.ulp = fdiv <2 x float> <float 1.000000e+00, float 1.000000e+00>, %x, !fpmath !1
+  store volatile <2 x float> %md.half.ulp, ptr addrspace(1) %out, align 8
+  %afn.no.md = fdiv afn <2 x float> <float 1.000000e+00, float 1.000000e+00>, %x
+  store volatile <2 x float> %afn.no.md, ptr addrspace(1) %out, align 8
+  %fast.no.md = fdiv fast <2 x float> <float 1.000000e+00, float 1.000000e+00>, %x
+  store volatile <2 x float> %fast.no.md, ptr addrspace(1) %out, align 8
+  %afn.25ulp = fdiv afn <2 x float> <float 1.000000e+00, float 1.000000e+00>, %x, !fpmath !0
+  store volatile <2 x float> %afn.25ulp, ptr addrspace(1) %out, align 8
+  %fast.25ulp = fdiv fast <2 x float> <float 1.000000e+00, float 1.000000e+00>, %x, !fpmath !0
+  store volatile <2 x float> %fast.25ulp, ptr addrspace(1) %out, align 8
   ret void
 }
 
-; CHECK-LABEL: @rcp_fdiv_fpmath_vector(
-; CHECK: %[[NO0:[0-9]+]] =  extractelement <2 x float> %x, i64 0
-; CHECK: %[[NO_FDIV0:[0-9]+]] = fdiv float 1.000000e+00, %[[NO0]]
-; CHECK: %[[NO_INS0:[0-9]+]] = insertelement <2 x float> poison, float %[[NO_FDIV0]], i64 0
-; CHECK: %[[NO1:[0-9]+]] = extractelement <2 x float> %x, i64 1
-; CHECK: %[[NO_FDIV1:[0-9]+]] = fdiv float 1.000000e+00, %[[NO1]]
-; CHECK: %no.md = insertelement <2 x float> %[[NO_INS0]], float %[[NO_FDIV1]], i64 1
-; CHECK: store volatile <2 x float> %no.md, ptr addrspace(1) %out
-
-; CHECK: %[[HALF0:[0-9]+]] =  extractelement <2 x float> %x, i64 0
-; CHECK: %[[HALF_FDIV0:[0-9]+]] = fdiv float 1.000000e+00, %[[HALF0]]
-; CHECK: %[[HALF_INS0:[0-9]+]] = insertelement <2 x float> poison, float %[[HALF_FDIV0]], i64 0
-; CHECK: %[[HALF1:[0-9]+]] = extractelement <2 x float> %x, i64 1
-; CHECK: %[[HALF_FDIV1:[0-9]+]] =  fdiv float 1.000000e+00, %[[HALF1]]
-; CHECK: %md.half.ulp = insertelement <2 x float> %[[HALF_INS0]], float %[[HALF_FDIV1]], i64 1
-; CHECK: store volatile <2 x float> %md.half.ulp, ptr addrspace(1) %out
-
-; CHECK: %[[AFN_NO0:[0-9]+]] =  extractelement <2 x float> %x, i64 0
-; CHECK: %[[AFN_NO_FDIV0:[0-9]+]] = call afn float @llvm.amdgcn.rcp.f32(float %[[AFN_NO0]])
-; CHECK: %[[AFN_NO_INS0:[0-9]+]] = insertelement <2 x float> poison, float %[[AFN_NO_FDIV0]], i64 0
-; CHECK: %[[AFN_NO1:[0-9]+]] = extractelement <2 x float> %x, i64 1
-; CHECK: %[[AFN_NO_FDIV1:[0-9]+]] =  call afn float @llvm.amdgcn.rcp.f32(float %[[AFN_NO1]])
-; CHECK: %afn.no.md = insertelement <2 x float> %[[AFN_NO_INS0]], float %[[AFN_NO_FDIV1]], i64 1
-; CHECK: store volatile <2 x float> %afn.no.md, ptr addrspace(1) %out
-
-; CHECK: %[[FAST_NO0:[0-9]+]] =  extractelement <2 x float> %x, i64 0
-; CHECK: %[[FAST_NO_RCP0:[0-9]+]] = call fast float @llvm.amdgcn.rcp.f32(float %[[FAST_NO0]])
-; CHECK: %[[FAST_NO_INS0:[0-9]+]] = insertelement <2 x float> poison, float %[[FAST_NO_RCP0]], i64 0
-; CHECK: %[[FAST_NO1:[0-9]+]] = extractelement <2 x float> %x, i64 1
-; CHECK: %[[FAST_NO_RCP1:[0-9]+]] =  call fast float @llvm.amdgcn.rcp.f32(float %[[FAST_NO1]])
-; CHECK: %fast.no.md = insertelement <2 x float> %[[FAST_NO_INS0]], float %[[FAST_NO_RCP1]], i64 1
-; CHECK: store volatile <2 x float> %fast.no.md, ptr addrspace(1) %out
-
-; CHECK: %[[AFN_250:[0-9]+]] =  extractelement <2 x float> %x, i64 0
-; CHECK: %[[AFN_25_RCP0:[0-9]+]] = call afn float @llvm.amdgcn.rcp.f32(float %[[AFN_250]])
-; CHECK: %[[AFN_25_INS0:[0-9]+]] = insertelement <2 x float> poison, float %[[AFN_25_RCP0]], i64 0
-; CHECK: %[[AFN_251:[0-9]+]] = extractelement <2 x float> %x, i64 1
-; CHECK: %[[AFN_25_RCP1:[0-9]+]] =  call afn float @llvm.amdgcn.rcp.f32(float %[[AFN_251]])
-; CHECK: %afn.25ulp = insertelement <2 x float> %[[AFN_25_INS0]], float %[[AFN_25_RCP1]], i64 1
-; CHECK: store volatile <2 x float> %afn.25ulp, ptr addrspace(1) %out
-
-; CHECK: %[[FAST_250:[0-9]+]] =  extractelement <2 x float> %x, i64 0
-; CHECK: %[[FAST_25_RCP0:[0-9]+]] = call fast float @llvm.amdgcn.rcp.f32(float %[[FAST_250]])
-; CHECK: %[[FAST_25_INS0:[0-9]+]] = insertelement <2 x float> poison, float %[[FAST_25_RCP0]], i64 0
-; CHECK: %[[FAST_251:[0-9]+]] = extractelement <2 x float> %x, i64 1
-; CHECK: %[[FAST_25_RCP1:[0-9]+]] =  call fast float @llvm.amdgcn.rcp.f32(float %[[FAST_251]])
-; CHECK: %fast.25ulp = insertelement <2 x float> %[[FAST_25_INS0]], float %[[FAST_25_RCP1]], i64 1
-; CHECK: store volatile <2 x float> %fast.25ulp, ptr addrspace(1) %out
-define amdgpu_kernel void @rcp_fdiv_fpmath_vector(ptr addrspace(1) %out, <2 x float> %x) #1 {
-  %no.md = fdiv <2 x float> <float 1.0, float 1.0>, %x
-  store volatile <2 x float> %no.md, ptr addrspace(1) %out
-
-  %md.half.ulp = fdiv <2 x float> <float 1.0, float 1.0>, %x, !fpmath !1
-  store volatile <2 x float> %md.half.ulp, ptr addrspace(1) %out
-
-  %afn.no.md = fdiv afn <2 x float> <float 1.0, float 1.0>, %x
-  store volatile <2 x float> %afn.no.md, ptr addrspace(1) %out
-
-  %fast.no.md = fdiv fast <2 x float> <float 1.0, float 1.0>, %x
-  store volatile <2 x float> %fast.no.md, ptr addrspace(1) %out
-
-  %afn.25ulp = fdiv afn <2 x float> <float 1.0, float 1.0>, %x, !fpmath !0
-  store volatile <2 x float> %afn.25ulp, ptr addrspace(1) %out
-
-  %fast.25ulp = fdiv fast <2 x float> <float 1.0, float 1.0>, %x, !fpmath !0
-  store volatile <2 x float> %fast.25ulp, ptr addrspace(1) %out
-
-  ret void
-}
-
-; CHECK-LABEL: @rcp_fdiv_fpmath_vector_nonsplat(
-; CHECK: %[[NO0:[0-9]+]] =  extractelement <2 x float> %x, i64 0
-; CHECK: %[[NO_FDIV0:[0-9]+]] = fdiv float 1.000000e+00, %[[NO0]]
-; CHECK: %[[NO_INS0:[0-9]+]] = insertelement <2 x float> poison, float %[[NO_FDIV0]], i64 0
-; CHECK: %[[NO1:[0-9]+]] = extractelement <2 x float> %x, i64 1
-; CHECK: %[[NO_FDIV1:[0-9]+]] = fdiv float 2.000000e+00, %[[NO1]]
-; CHECK: %no.md = insertelement <2 x float> %[[NO_INS0]], float %[[NO_FDIV1]], i64 1
-; CHECK: store volatile <2 x float> %no.md, ptr addrspace(1) %out
-
-; CHECK: %[[AFN_NO0:[0-9]+]] =  extractelement <2 x float> %x, i64 0
-; CHECK: %[[AFN_NO_FDIV0:[0-9]+]] = call afn float @llvm.amdgcn.rcp.f32(float %[[AFN_NO0]])
-; CHECK: %[[AFN_NO_INS0:[0-9]+]] = insertelement <2 x float> poison, float %[[AFN_NO_FDIV0]], i64 0
-; CHECK: %[[AFN_NO1:[0-9]+]] = extractelement <2 x float> %x, i64 1
-; CHECK: %[[AFN_NO_FDIV1:[0-9]+]] =  call afn float @llvm.amdgcn.rcp.f32(float %[[AFN_NO1]])
-; CHECK: %[[AFN_NO_MUL1:[0-9]+]] = fmul afn float 2.000000e+00, %[[AFN_NO_FDIV1]]
-; CHECK: %afn.no.md = insertelement <2 x float> %[[AFN_NO_INS0]], float %[[AFN_NO_MUL1]], i64 1
-; CHECK: store volatile <2 x float> %afn.no.md, ptr addrspace(1) %out
-
-; CHECK: %[[FAST_NO0:[0-9]+]] =  extractelement <2 x float> %x, i64 0
-; CHECK: %[[FAST_NO_RCP0:[0-9]+]] = call fast float @llvm.amdgcn.rcp.f32(float %[[FAST_NO0]])
-; CHECK: %[[FAST_NO_INS0:[0-9]+]] = insertelement <2 x float> poison, float %[[FAST_NO_RCP0]], i64 0
-; CHECK: %[[FAST_NO1:[0-9]+]] = extractelement <2 x float> %x, i64 1
-; CHECK: %[[FAST_NO_RCP1:[0-9]+]] =  call fast float @llvm.amdgcn.rcp.f32(float %[[FAST_NO1]])
-; CHECK: %[[FAST_NO_MUL1:[0-9]+]] = fmul fast float 2.000000e+00, %[[FAST_NO_RCP1]]
-; CHECK: %fast.no.md = insertelement <2 x float> %[[FAST_NO_INS0]], float %[[FAST_NO_MUL1]], i64 1
-; CHECK: store volatile <2 x float> %fast.no.md, ptr addrspace(1) %out
-
-; CHECK: %[[AFN_250:[0-9]+]] =  extractelement <2 x float> %x, i64 0
-; CHECK: %[[AFN_25_RCP0:[0-9]+]] = call afn float @llvm.amdgcn.rcp.f32(float %[[AFN_250]])
-; CHECK: %[[AFN_25_INS0:[0-9]+]] = insertelement <2 x float> poison, float %[[AFN_25_RCP0]], i64 0
-; CHECK: %[[AFN_251:[0-9]+]] = extractelement <2 x float> %x, i64 1
-; CHECK: %[[AFN_25_RCP1:[0-9]+]] =  call afn float @llvm.amdgcn.rcp.f32(float %[[AFN_251]])
-; CHECK: %[[AFN_25_MUL1:[0-9]+]] = fmul afn float 2.000000e+00, %[[AFN_25_RCP1]]
-; CHECK: %afn.25ulp = insertelement <2 x float> %[[AFN_25_INS0]], float %[[AFN_25_MUL1]], i64 1
-; CHECK: store volatile <2 x float> %afn.25ulp, ptr addrspace(1) %out
-
-; CHECK: %[[FAST_250:[0-9]+]] =  extractelement <2 x float> %x, i64 0
-; CHECK: %[[FAST_25_RCP0:[0-9]+]] = call fast float @llvm.amdgcn.rcp.f32(float %[[FAST_250]])
-; CHECK: %[[FAST_25_INS0:[0-9]+]] = insertelement <2 x float> poison, float %[[FAST_25_RCP0]], i64 0
-; CHECK: %[[FAST_251:[0-9]+]] = extractelement <2 x float> %x, i64 1
-; CHECK: %[[FAST_25_RCP1:[0-9]+]] =  call fast float @llvm.amdgcn.rcp.f32(float %[[FAST_251]])
-; CHECK: %[[FAST_25_MUL1:[0-9]+]] = fmul fast float 2.000000e+00, %[[FAST_25_RCP1]]
-; CHECK: %fast.25ulp = insertelement <2 x float> %[[FAST_25_INS0]], float %[[FAST_25_MUL1]], i64 1
-; CHECK: store volatile <2 x float> %fast.25ulp, ptr addrspace(1) %out
-define amdgpu_kernel void @rcp_fdiv_fpmath_vector_nonsplat(ptr addrspace(1) %out, <2 x float> %x) #1 {
-  %no.md = fdiv <2 x float> <float 1.0, float 2.0>, %x
-  store volatile <2 x float> %no.md, ptr addrspace(1) %out
-
-  %afn.no.md = fdiv afn <2 x float> <float 1.0, float 2.0>, %x
-  store volatile <2 x float> %afn.no.md, ptr addrspace(1) %out
-
-  %fast.no.md = fdiv fast <2 x float> <float 1.0, float 2.0>, %x
-  store volatile <2 x float> %fast.no.md, ptr addrspace(1) %out
-
-  %afn.25ulp = fdiv afn <2 x float> <float 1.0, float 2.0>, %x, !fpmath !0
-  store volatile <2 x float> %afn.25ulp, ptr addrspace(1) %out
-
-  %fast.25ulp = fdiv fast <2 x float> <float 1.0, float 2.0>, %x, !fpmath !0
-  store volatile <2 x float> %fast.25ulp, ptr addrspace(1) %out
-
-  ret void
-}
-
-; CHECK-LABEL: @rcp_fdiv_fpmath_vector_partial_constant(
-; CHECK: %[[AFN_A0:[0-9]+]] = extractelement <2 x float> %x.insert, i64 0
-; CHECK: %[[AFN_B0:[0-9]+]] = extractelement <2 x float> %y, i64 0
-; CHECK: %[[AFN_RCP0:[0-9]+]] = call afn float @llvm.amdgcn.rcp.f32(float %[[AFN_B0]])
-; CHECK: %[[AFN_MUL0:[0-9]+]] = fmul afn float %[[AFN_A0]], %[[AFN_RCP0]]
-; CHECK: %[[AFN_INS0:[0-9]+]] = insertelement <2 x float> poison, float %[[AFN_MUL0]], i64 0
-; CHECK: %[[AFN_A1:[0-9]+]] = extractelement <2 x float> %x.insert, i64 1
-; CHECK: %[[AFN_B1:[0-9]+]] = extractelement <2 x float> %y, i64 1
-; CHECK: %[[AFN_RCP1:[0-9]+]] = call afn float @llvm.amdgcn.rcp.f32(float %[[AFN_B1]])
-; CHECK: %[[AFN_MUL1:[0-9]+]] = fmul afn float %[[AFN_A1]], %[[AFN_RCP1]]
-; CHECK: %afn.25ulp = insertelement <2 x float> %[[AFN_INS0]], float %[[AFN_MUL1]], i64 1
-; CHECK: store volatile <2 x float> %afn.25ulp
-
-; CHECK: %[[FAST_A0:[0-9]+]] = extractelement <2 x float> %x.insert, i64 0
-; CHECK: %[[FAST_B0:[0-9]+]] = extractelement <2 x float> %y, i64 0
-; CHECK: %[[FAST_RCP0:[0-9]+]] = call fast float @llvm.amdgcn.rcp.f32(float %[[FAST_B0]])
-; CHECK: %[[FAST_MUL0:[0-9]+]] = fmul fast float %[[FAST_A0]], %[[FAST_RCP0]]
-; CHECK: %[[FAST_INS0:[0-9]+]] = insertelement <2 x float> poison, float %[[FAST_MUL0]], i64 0
-; CHECK: %[[FAST_A1:[0-9]+]] = extractelement <2 x float> %x.insert, i64 1
-; CHECK: %[[FAST_B1:[0-9]+]] = extractelement <2 x float> %y, i64 1
-; CHECK: %[[FAST_RCP1:[0-9]+]] = call fast float @llvm.amdgcn.rcp.f32(float %[[FAST_B1]])
-; CHECK: %[[FAST_MUL1:[0-9]+]] = fmul fast float %[[FAST_A1]], %[[FAST_RCP1]]
-; CHECK: %fast.25ulp = insertelement <2 x float> %[[FAST_INS0]], float %[[FAST_MUL1]], i64 1
-; CHECK: store volatile <2 x float> %fast.25ulp
-define amdgpu_kernel void @rcp_fdiv_fpmath_vector_partial_constant(ptr addrspace(1) %out, <2 x float> %x, <2 x float> %y) #1 {
-  %x.insert = insertelement <2 x float> %x, float 1.0, i32 0
+define amdgpu_kernel void @rcp_fdiv_f32_fpmath_vector_nonsplat(ptr addrspace(1) %out, <2 x float> %x) {
+; CHECK-LABEL: define amdgpu_kernel void @rcp_fdiv_f32_fpmath_vector_nonsplat
+; CHECK-SAME: (ptr addrspace(1) [[OUT:%.*]], <2 x float> [[X:%.*]]) #[[ATTR1]] {
+; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <2 x float> [[X]], i64 0
+; CHECK-NEXT:    [[TMP2:%.*]] = fdiv float 1.000000e+00, [[TMP1]]
+; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <2 x float> poison, float [[TMP2]], i64 0
+; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x float> [[X]], i64 1
+; CHECK-NEXT:    [[TMP5:%.*]] = fdiv float 2.000000e+00, [[TMP4]]
+; CHECK-NEXT:    [[NO_MD:%.*]] = insertelement <2 x float> [[TMP3]], float [[TMP5]], i64 1
+; CHECK-NEXT:    store volatile <2 x float> [[NO_MD]], ptr addrspace(1) [[OUT]], align 8
+; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <2 x float> [[X]], i64 0
+; CHECK-NEXT:    [[TMP7:%.*]] = call afn float @llvm.amdgcn.rcp.f32(float [[TMP6]])
+; CHECK-NEXT:    [[TMP8:%.*]] = insertelement <2 x float> poison, float [[TMP7]], i64 0
+; CHECK-NEXT:    [[TMP9:%.*]] = extractelement <2 x float> [[X]], i64 1
+; CHECK-NEXT:    [[TMP10:%.*]] = call afn float @llvm.amdgcn.rcp.f32(float [[TMP9]])
+; CHECK-NEXT:    [[TMP11:%.*]] = fmul afn float 2.000000e+00, [[TMP10]]
+; CHECK-NEXT:    [[AFN_NO_MD:%.*]] = insertelement <2 x float> [[TMP8]], float [[TMP11]], i64 1
+; CHECK-NEXT:    store volatile <2 x float> [[AFN_NO_MD]], ptr addrspace(1) [[OUT]], align 8
+; CHECK-NEXT:    [[TMP12:%.*]] = extractelement <2 x float> [[X]], i64 0
+; CHECK-NEXT:    [[TMP13:%.*]] = call fast float @llvm.amdgcn.rcp.f32(float [[TMP12]])
+; CHECK-NEXT:    [[TMP14:%.*]] = insertelement <2 x float> poison, float [[TMP13]], i64 0
+; CHECK-NEXT:    [[TMP15:%.*]] = extractelement <2 x float> [[X]], i64 1
+; CHECK-NEXT:    [[TMP16:%.*]] = call fast float @llvm.amdgcn.rcp.f32(float [[TMP15]])
+; CHECK-NEXT:    [[TMP17:%.*]] = fmul fast float 2.000000e+00, [[TMP16]]
+; CHECK-NEXT:    [[FAST_NO_MD:%.*]] = insertelement <2 x float> [[TMP14]], float [[TMP17]], i64 1
+; CHECK-NEXT:    store volatile <2 x float> [[FAST_NO_MD]], ptr addrspace(1) [[OUT]], align 8
+; CHECK-NEXT:    [[TMP18:%.*]] = extractelement <2 x float> [[X]], i64 0
+; CHECK-NEXT:    [[TMP19:%.*]] = call afn float @llvm.amdgcn.rcp.f32(float [[TMP18]])
+; CHECK-NEXT:    [[TMP20:%.*]] = insertelement <2 x float> poison, float [[TMP19]], i64 0
+; CHECK-NEXT:    [[TMP21:%.*]] = extractelement <2 x float> [[X]], i64 1
+; CHECK-NEXT:    [[TMP22:%.*]] = call afn float @llvm.amdgcn.rcp.f32(float [[TMP21]])
+; CHECK-NEXT:    [[TMP23:%.*]] = fmul afn float 2.000000e+00, [[TMP22]]
+; CHECK-NEXT:    [[AFN_25ULP:%.*]] = insertelement <2 x float> [[TMP20]], float [[TMP23]], i64 1
+; CHECK-NEXT:    store volatile <2 x float> [[AFN_25ULP]], ptr addrspace(1) [[OUT]], align 8
+; CHECK-NEXT:    [[TMP24:%.*]] = extractelement <2 x float> [[X]], i64 0
+; CHECK-NEXT:    [[TMP25:%.*]] = call fast float @llvm.amdgcn.rcp.f32(float [[TMP24]])
+; CHECK-NEXT:    [[TMP26:%.*]] = insertelement <2 x float> poison, float [[TMP25]], i64 0
+; CHECK-NEXT:    [[TMP27:%.*]] = extractelement <2 x float> [[X]], i64 1
+; CHECK-NEXT:    [[TMP28:%.*]] = call fast float @llvm.amdgcn.rcp.f32(float [[TMP27]])
+; CHECK-NEXT:    [[TMP29:%.*]] = fmul fast float 2.000000e+00, [[TMP28]]
+; CHECK-NEXT:    [[FAST_25ULP:%.*]] = insertelement <2 x float> [[TMP26]], float [[TMP29]], i64 1
+; CHECK-NEXT:    store volatile <2 x float> [[FAST_25ULP]], ptr addrspace(1) [[OUT]], align 8
+; CHECK-NEXT:    ret void
+;
+  %no.md = fdiv <2 x float> <float 1.000000e+00, float 2.000000e+00>, %x
+  store volatile <2 x float> %no.md, ptr addrspace(1) %out, align 8
+  %afn.no.md = fdiv afn <2 x float> <float 1.000000e+00, float 2.000000e+00>, %x
+  store volatile <2 x float> %afn.no.md, ptr addrspace(1) %out, align 8
+  %fast.no.md = fdiv fast <2 x float> <float 1.000000e+00, float 2.000000e+00>, %x
+  store volatile <2 x float> %fast.no.md, ptr addrspace(1) %out, align 8
+  %afn.25ulp = fdiv afn <2 x float> <float 1.000000e+00, float 2.000000e+00>, %x, !fpmath !0
+  store volatile <2 x float> %afn.25ulp, ptr addrspace(1) %out, align 8
+  %fast.25ulp = fdiv fast <2 x float> <float 1.000000e+00, float 2.000000e+00>, %x, !fpmath !0
+  store volatile <2 x float> %fast.25ulp, ptr addrspace(1) %out, align 8
+  ret void
+}
 
+define amdgpu_kernel void @rcp_fdiv_f32_vector_fpmath_partial_constant(ptr addrspace(1) %out, <2 x float> %x, <2 x float> %y) {
+; CHECK-LABEL: define amdgpu_kernel void @rcp_fdiv_f32_vector_fpmath_partial_constant
+; CHECK-SAME: (ptr addrspace(1) [[OUT:%.*]], <2 x float> [[X:%.*]], <2 x float> [[Y:%.*]]) #[[ATTR1]] {
+; CHECK-NEXT:    [[X_INSERT:%.*]] = insertelement <2 x float> [[X]], float 1.000000e+00, i32 0
+; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <2 x float> [[X_INSERT]], i64 0
+; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <2 x float> [[Y]], i64 0
+; CHECK-NEXT:    [[TMP3:%.*]] = call afn float @llvm.amdgcn.rcp.f32(float [[TMP2]])
+; CHECK-NEXT:    [[TMP4:%.*]] = fmul afn float [[TMP1]], [[TMP3]]
+; CHECK-NEXT:    [[TMP5:%.*]] = insertelement <2 x float> poison, float [[TMP4]], i64 0
+; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <2 x float> [[X_INSERT]], i64 1
+; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <2 x float> [[Y]], i64 1
+; CHECK-NEXT:    [[TMP8:%.*]] = call afn float @llvm.amdgcn.rcp.f32(float [[TMP7]])
+; CHECK-NEXT:    [[TMP9:%.*]] = fmul afn float [[TMP6]], [[TMP8]]
+; CHECK-NEXT:    [[AFN_25ULP:%.*]] = insertelement <2 x float> [[TMP5]], float [[TMP9]], i64 1
+; CHECK-NEXT:    store volatile <2 x float> [[AFN_25ULP]], ptr addrspace(1) [[OUT]], align 8
+; CHECK-NEXT:    [[TMP10:%.*]] = extractelement <2 x float> [[X_INSERT]], i64 0
+; CHECK-NEXT:    [[TMP11:%.*]] = extractelement <2 x float> [[Y]], i64 0
+; CHECK-NEXT:    [[TMP12:%.*]] = call fast float @llvm.amdgcn.rcp.f32(float [[TMP11]])
+; CHECK-NEXT:    [[TMP13:%.*]] = fmul fast float [[TMP10]], [[TMP12]]
+; CHECK-NEXT:    [[TMP14:%.*]] = insertelement <2 x float> poison, float [[TMP13]], i64 0
+; CHECK-NEXT:    [[TMP15:%.*]] = extractelement <2 x float> [[X_INSERT]], i64 1
+; CHECK-NEXT:    [[TMP16:%.*]] = extractelement <2 x float> [[Y]], i64 1
+; CHECK-NEXT:    [[TMP17:%.*]] = call fast float @llvm.amdgcn.rcp.f32(float [[TMP16]])
+; CHECK-NEXT:    [[TMP18:%.*]] = fmul fast float [[TMP15]], [[TMP17]]
+; CHECK-NEXT:    [[FAST_25ULP:%.*]] = insertelement <2 x float> [[TMP14]], float [[TMP18]], i64 1
+; CHECK-NEXT:    store volatile <2 x float> [[FAST_25ULP]], ptr addrspace(1) [[OUT]], align 8
+; CHECK-NEXT:    ret void
+;
+  %x.insert = insertelement <2 x float> %x, float 1.000000e+00, i32 0
   %afn.25ulp = fdiv afn <2 x float> %x.insert, %y, !fpmath !0
-  store volatile <2 x float> %afn.25ulp, ptr addrspace(1) %out
-
+  store volatile <2 x float> %afn.25ulp, ptr addrspace(1) %out, align 8
   %fast.25ulp = fdiv fast <2 x float> %x.insert, %y, !fpmath !0
-  store volatile <2 x float> %fast.25ulp, ptr addrspace(1) %out
+  store volatile <2 x float> %fast.25ulp, ptr addrspace(1) %out, align 8
+  ret void
+}
+
+define amdgpu_kernel void @rsq_f32_fpmath(ptr addrspace(1) %out, float %x) {
+; IEEE-LABEL: define amdgpu_kernel void @rsq_f32_fpmath
+; IEEE-SAME: (ptr addrspace(1) [[OUT:%.*]], float [[X:%.*]]) #[[ATTR1]] {
+; IEEE-NEXT:    [[SQRT_X_NO_MD:%.*]] = call float @llvm.sqrt.f32(float [[X]])
+; IEEE-NEXT:    [[NO_MD:%.*]] = fdiv float 1.000000e+00, [[SQRT_X_NO_MD]]
+; IEEE-NEXT:    store volatile float [[NO_MD]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    [[SQRT_MD_1ULP:%.*]] = call float @llvm.sqrt.f32(float [[X]]), !fpmath [[META2]]
+; IEEE-NEXT:    [[MD_1ULP:%.*]] = fdiv float 1.000000e+00, [[SQRT_MD_1ULP]], !fpmath [[META2]]
+; IEEE-NEXT:    store volatile float [[MD_1ULP]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    [[SQRT_MD_1ULP_MULTI_USE:%.*]] = call float @llvm.sqrt.f32(float [[X]]), !fpmath [[META2]]
+; IEEE-NEXT:    store volatile float [[SQRT_MD_1ULP_MULTI_USE]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    [[MD_1ULP_MULTI_USE:%.*]] = fdiv float 1.000000e+00, [[SQRT_MD_1ULP_MULTI_USE]], !fpmath [[META2]]
+; IEEE-NEXT:    store volatile float [[MD_1ULP_MULTI_USE]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    [[SQRT_MD_25ULP:%.*]] = call float @llvm.sqrt.f32(float [[X]]), !fpmath [[META0]]
+; IEEE-NEXT:    [[MD_25ULP:%.*]] = call float @llvm.amdgcn.fdiv.fast(float 1.000000e+00, float [[SQRT_MD_25ULP]])
+; IEEE-NEXT:    store volatile float [[MD_25ULP]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    [[SQRT_MD_HALF_ULP:%.*]] = call float @llvm.sqrt.f32(float [[X]]), !fpmath [[META1]]
+; IEEE-NEXT:    [[MD_HALF_ULP:%.*]] = fdiv float 1.000000e+00, [[SQRT_MD_HALF_ULP]], !fpmath [[META1]]
+; IEEE-NEXT:    store volatile float [[MD_HALF_ULP]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    [[SQRT_X_AFN_NO_MD:%.*]] = call afn float @llvm.sqrt.f32(float [[X]])
+; IEEE-NEXT:    [[AFN_NO_MD:%.*]] = call afn float @llvm.amdgcn.rcp.f32(float [[SQRT_X_AFN_NO_MD]])
+; IEEE-NEXT:    store volatile float [[AFN_NO_MD]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    [[SQRT_X_AFN_25ULP:%.*]] = call afn float @llvm.sqrt.f32(float [[X]]), !fpmath [[META0]]
+; IEEE-NEXT:    [[AFN_25ULP:%.*]] = call afn float @llvm.amdgcn.rcp.f32(float [[SQRT_X_AFN_25ULP]])
+; IEEE-NEXT:    store volatile float [[AFN_25ULP]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    [[SQRT_X_FAST_NO_MD:%.*]] = call fast float @llvm.sqrt.f32(float [[X]])
+; IEEE-NEXT:    [[FAST_NO_MD:%.*]] = call fast float @llvm.amdgcn.rcp.f32(float [[SQRT_X_FAST_NO_MD]])
+; IEEE-NEXT:    store volatile float [[FAST_NO_MD]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    [[SQRT_X_FAST_25ULP:%.*]] = call fast float @llvm.sqrt.f32(float [[X]]), !fpmath [[META0]]
+; IEEE-NEXT:    [[FAST_25ULP:%.*]] = call fast float @llvm.amdgcn.rcp.f32(float [[SQRT_X_FAST_25ULP]])
+; IEEE-NEXT:    store volatile float [[FAST_25ULP]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    [[SQRT_X_3ULP:%.*]] = call float @llvm.sqrt.f32(float [[X]]), !fpmath [[META3]]
+; IEEE-NEXT:    [[FDIV_OPENCL:%.*]] = call float @llvm.amdgcn.fdiv.fast(float 1.000000e+00, float [[SQRT_X_3ULP]])
+; IEEE-NEXT:    store volatile float [[FDIV_OPENCL]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    [[SQRT_X_HALF_ULP:%.*]] = call float @llvm.sqrt.f32(float [[X]]), !fpmath [[META1]]
+; IEEE-NEXT:    [[FDIV_SQRT_MISMATCH_MD0:%.*]] = call float @llvm.amdgcn.fdiv.fast(float 1.000000e+00, float [[SQRT_X_HALF_ULP]])
+; IEEE-NEXT:    store volatile float [[FDIV_SQRT_MISMATCH_MD0]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    [[SQRT_MISMATCH_MD1:%.*]] = call afn float @llvm.sqrt.f32(float [[X]])
+; IEEE-NEXT:    [[FDIV_SQRT_MISMATCH_MD1:%.*]] = call float @llvm.amdgcn.fdiv.fast(float 1.000000e+00, float [[SQRT_MISMATCH_MD1]])
+; IEEE-NEXT:    store volatile float [[FDIV_SQRT_MISMATCH_MD1]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    [[SQRT_MISMATCH_MD2:%.*]] = call float @llvm.sqrt.f32(float [[X]]), !fpmath [[META3]]
+; IEEE-NEXT:    [[FDIV_SQRT_MISMATCH_MD2:%.*]] = call afn float @llvm.amdgcn.rcp.f32(float [[SQRT_MISMATCH_MD2]])
+; IEEE-NEXT:    store volatile float [[FDIV_SQRT_MISMATCH_MD2]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    ret void
+;
+; DAZ-LABEL: define amdgpu_kernel void @rsq_f32_fpmath
+; DAZ-SAME: (ptr addrspace(1) [[OUT:%.*]], float [[X:%.*]]) #[[ATTR1]] {
+; DAZ-NEXT:    [[SQRT_X_NO_MD:%.*]] = call float @llvm.sqrt.f32(float [[X]])
+; DAZ-NEXT:    [[NO_MD:%.*]] = fdiv float 1.000000e+00, [[SQRT_X_NO_MD]]
+; DAZ-NEXT:    store volatile float [[NO_MD]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    [[SQRT_MD_1ULP:%.*]] = call float @llvm.sqrt.f32(float [[X]]), !fpmath [[META2]]
+; DAZ-NEXT:    [[MD_1ULP:%.*]] = call float @llvm.amdgcn.rcp.f32(float [[SQRT_MD_1ULP]])
+; DAZ-NEXT:    store volatile float [[MD_1ULP]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    [[SQRT_MD_1ULP_MULTI_USE:%.*]] = call float @llvm.sqrt.f32(float [[X]]), !fpmath [[META2]]
+; DAZ-NEXT:    store volatile float [[SQRT_MD_1ULP_MULTI_USE]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    [[MD_1ULP_MULTI_USE:%.*]] = call float @llvm.amdgcn.rcp.f32(float [[SQRT_MD_1ULP_MULTI_USE]])
+; DAZ-NEXT:    store volatile float [[MD_1ULP_MULTI_USE]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    [[SQRT_MD_25ULP:%.*]] = call float @llvm.sqrt.f32(float [[X]]), !fpmath [[META0]]
+; DAZ-NEXT:    [[MD_25ULP:%.*]] = call float @llvm.amdgcn.rcp.f32(float [[SQRT_MD_25ULP]])
+; DAZ-NEXT:    store volatile float [[MD_25ULP]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    [[SQRT_MD_HALF_ULP:%.*]] = call float @llvm.sqrt.f32(float [[X]]), !fpmath [[META1]]
+; DAZ-NEXT:    [[MD_HALF_ULP:%.*]] = fdiv float 1.000000e+00, [[SQRT_MD_HALF_ULP]], !fpmath [[META1]]
+; DAZ-NEXT:    store volatile float [[MD_HALF_ULP]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    [[SQRT_X_AFN_NO_MD:%.*]] = call afn float @llvm.sqrt.f32(float [[X]])
+; DAZ-NEXT:    [[AFN_NO_MD:%.*]] = call afn float @llvm.amdgcn.rcp.f32(float [[SQRT_X_AFN_NO_MD]])
+; DAZ-NEXT:    store volatile float [[AFN_NO_MD]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    [[SQRT_X_AFN_25ULP:%.*]] = call afn float @llvm.sqrt.f32(float [[X]]), !fpmath [[META0]]
+; DAZ-NEXT:    [[AFN_25ULP:%.*]] = call afn float @llvm.amdgcn.rcp.f32(float [[SQRT_X_AFN_25ULP]])
+; DAZ-NEXT:    store volatile float [[AFN_25ULP]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    [[SQRT_X_FAST_NO_MD:%.*]] = call fast float @llvm.sqrt.f32(float [[X]])
+; DAZ-NEXT:    [[FAST_NO_MD:%.*]] = call fast float @llvm.amdgcn.rcp.f32(float [[SQRT_X_FAST_NO_MD]])
+; DAZ-NEXT:    store volatile float [[FAST_NO_MD]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    [[SQRT_X_FAST_25ULP:%.*]] = call fast float @llvm.sqrt.f32(float [[X]]), !fpmath [[META0]]
+; DAZ-NEXT:    [[FAST_25ULP:%.*]] = call fast float @llvm.amdgcn.rcp.f32(float [[SQRT_X_FAST_25ULP]])
+; DAZ-NEXT:    store volatile float [[FAST_25ULP]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    [[SQRT_X_3ULP:%.*]] = call float @llvm.sqrt.f32(float [[X]]), !fpmath [[META3:![0-9]+]]
+; DAZ-NEXT:    [[FDIV_OPENCL:%.*]] = call float @llvm.amdgcn.rcp.f32(float [[SQRT_X_3ULP]])
+; DAZ-NEXT:    store volatile float [[FDIV_OPENCL]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    [[SQRT_X_HALF_ULP:%.*]] = call float @llvm.sqrt.f32(float [[X]]), !fpmath [[META1]]
+; DAZ-NEXT:    [[FDIV_SQRT_MISMATCH_MD0:%.*]] = call float @llvm.amdgcn.rcp.f32(float [[SQRT_X_HALF_ULP]])
+; DAZ-NEXT:    store volatile float [[FDIV_SQRT_MISMATCH_MD0]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    [[SQRT_MISMATCH_MD1:%.*]] = call afn float @llvm.sqrt.f32(float [[X]])
+; DAZ-NEXT:    [[FDIV_SQRT_MISMATCH_MD1:%.*]] = call float @llvm.amdgcn.rcp.f32(float [[SQRT_MISMATCH_MD1]])
+; DAZ-NEXT:    store volatile float [[FDIV_SQRT_MISMATCH_MD1]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    [[SQRT_MISMATCH_MD2:%.*]] = call float @llvm.sqrt.f32(float [[X]]), !fpmath [[META3]]
+; DAZ-NEXT:    [[FDIV_SQRT_MISMATCH_MD2:%.*]] = call afn float @llvm.amdgcn.rcp.f32(float [[SQRT_MISMATCH_MD2]])
+; DAZ-NEXT:    store volatile float [[FDIV_SQRT_MISMATCH_MD2]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    ret void
+;
+  %sqrt.x.no.md = call float @llvm.sqrt.f32(float %x)
+  %no.md = fdiv float 1.000000e+00, %sqrt.x.no.md
+  store volatile float %no.md, ptr addrspace(1) %out, align 4
+
+  ; Matches the rsq instruction accuracy
+  %sqrt.md.1ulp = call float @llvm.sqrt.f32(float %x), !fpmath !2
+  %md.1ulp = fdiv float 1.000000e+00, %sqrt.md.1ulp, !fpmath !2
+  store volatile float %md.1ulp, ptr addrspace(1) %out, align 4
+
+  %sqrt.md.1ulp.multi.use = call float @llvm.sqrt.f32(float %x), !fpmath !2
+  store volatile float %sqrt.md.1ulp.multi.use, ptr addrspace(1) %out, align 4
+  %md.1ulp.multi.use = fdiv float 1.000000e+00, %sqrt.md.1ulp.multi.use, !fpmath !2
+  store volatile float %md.1ulp.multi.use, ptr addrspace(1) %out, align 4
+
+  %sqrt.md.25ulp = call float @llvm.sqrt.f32(float %x), !fpmath !0
+  %md.25ulp = fdiv float 1.000000e+00, %sqrt.md.25ulp, !fpmath !0
+  store volatile float %md.25ulp, ptr addrspace(1) %out, align 4
+
+  %sqrt.md.half.ulp = call float @llvm.sqrt.f32(float %x), !fpmath !1
+  %md.half.ulp = fdiv float 1.000000e+00, %sqrt.md.half.ulp, !fpmath !1
+  store volatile float %md.half.ulp, ptr addrspace(1) %out, align 4
+
+  %sqrt.x.afn.no.md = call afn float @llvm.sqrt.f32(float %x)
+  %afn.no.md = fdiv afn float 1.000000e+00, %sqrt.x.afn.no.md
+  store volatile float %afn.no.md, ptr addrspace(1) %out, align 4
+
+  %sqrt.x.afn.25ulp = call afn float @llvm.sqrt.f32(float %x), !fpmath !0
+  %afn.25ulp = fdiv afn float 1.000000e+00, %sqrt.x.afn.25ulp, !fpmath !0
+  store volatile float %afn.25ulp, ptr addrspace(1) %out, align 4
+
+  %sqrt.x.fast.no.md = call fast float @llvm.sqrt.f32(float %x)
+  %fast.no.md = fdiv fast float 1.000000e+00, %sqrt.x.fast.no.md
+  store volatile float %fast.no.md, ptr addrspace(1) %out, align 4
+
+  %sqrt.x.fast.25ulp = call fast float @llvm.sqrt.f32(float %x), !fpmath !0
+  %fast.25ulp = fdiv fast float 1.000000e+00, %sqrt.x.fast.25ulp, !fpmath !0
+  store volatile float %fast.25ulp, ptr addrspace(1) %out, align 4
+
+
+  ; Test mismatched metadata/flags between the sqrt and fdiv
+
+  ; Test the expected opencl default pattern
+  %sqrt.x.3ulp = call float @llvm.sqrt.f32(float %x), !fpmath !3  ; OpenCL default requires 3 for sqrt and 2.5 for fdiv
+  %fdiv.opencl = fdiv float 1.0, %sqrt.x.3ulp, !fpmath !0
+  store volatile float %fdiv.opencl, ptr addrspace(1) %out, align 4
+
+  ; sqrt demands higher precision than fdiv
+  %sqrt.x.half.ulp = call float @llvm.sqrt.f32(float %x), !fpmath !1
+  %fdiv.sqrt.mismatch.md0 = fdiv float 1.0, %sqrt.x.half.ulp, !fpmath !0
+  store volatile float %fdiv.sqrt.mismatch.md0, ptr addrspace(1) %out, align 4
+
+  ; sqrt demands full precision but has afn
+  %sqrt.mismatch.md1 = call afn float @llvm.sqrt.f32(float %x)
+  %fdiv.sqrt.mismatch.md1 = fdiv float 1.0, %sqrt.mismatch.md1, !fpmath !0
+  store volatile float %fdiv.sqrt.mismatch.md1, ptr addrspace(1) %out, align 4
+
+  ; sqrt has relaxed precision fdiv has afn only
+  %sqrt.mismatch.md2 = call float @llvm.sqrt.f32(float %x), !fpmath !3
+  %fdiv.sqrt.mismatch.md2 = fdiv afn float 1.0, %sqrt.mismatch.md2
+  store volatile float %fdiv.sqrt.mismatch.md2, ptr addrspace(1) %out, align 4
 
   ret void
 }
 
-; CHECK-LABEL: @fdiv_fpmath_f32_denormals(
-; CHECK: %no.md = fdiv float %a, %b{{$}}
-; CHECK: %md.half.ulp = fdiv float %a, %b
-; CHECK: %md.1ulp = fdiv float %a, %b
-; CHECK: %md.25ulp = fdiv float %a, %b
-; CHECK: %md.3ulp = fdiv float %a, %b
-; CHECK: %[[RCP_FAST:[0-9]+]] = call fast float @llvm.amdgcn.rcp.f32(float %b)
-; CHECK: %fast.md.25ulp = fmul fast float %a, %[[RCP_FAST]]
-; CHECK: %[[RCP_AFN:[0-9]+]] = call afn float @llvm.amdgcn.rcp.f32(float %b)
-; CHECK: %afn.md.25ulp  = fmul afn float %a, %[[RCP_AFN]]
-define amdgpu_kernel void @fdiv_fpmath_f32_denormals(ptr addrspace(1) %out, float %a, float %b) #2 {
-  %no.md = fdiv float %a, %b
-  store volatile float %no.md, ptr addrspace(1) %out
+define amdgpu_kernel void @rsq_f32_vector_fpmath(ptr addrspace(1) %out, <2 x float> %x) {
+; IEEE-LABEL: define amdgpu_kernel void @rsq_f32_vector_fpmath
+; IEEE-SAME: (ptr addrspace(1) [[OUT:%.*]], <2 x float> [[X:%.*]]) #[[ATTR1]] {
+; IEEE-NEXT:    [[SQRT_X_NO_MD:%.*]] = call <2 x float> @llvm.sqrt.v2f32(<2 x float> [[X]])
+; IEEE-NEXT:    [[TMP1:%.*]] = extractelement <2 x float> [[SQRT_X_NO_MD]], i64 0
+; IEEE-NEXT:    [[TMP2:%.*]] = fdiv float 1.000000e+00, [[TMP1]]
+; IEEE-NEXT:    [[TMP3:%.*]] = insertelement <2 x float> poison, float [[TMP2]], i64 0
+; IEEE-NEXT:    [[TMP4:%.*]] = extractelement <2 x float> [[SQRT_X_NO_MD]], i64 1
+; IEEE-NEXT:    [[TMP5:%.*]] = fdiv float 1.000000e+00, [[TMP4]]
+; IEEE-NEXT:    [[NO_MD:%.*]] = insertelement <2 x float> [[TMP3]], float [[TMP5]], i64 1
+; IEEE-NEXT:    store volatile <2 x float> [[NO_MD]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    [[SQRT_MD_1ULP:%.*]] = call <2 x float> @llvm.sqrt.v2f32(<2 x float> [[X]]), !fpmath [[META2]]
+; IEEE-NEXT:    [[TMP6:%.*]] = extractelement <2 x float> [[SQRT_MD_1ULP]], i64 0
+; IEEE-NEXT:    [[TMP7:%.*]] = fdiv float 1.000000e+00, [[TMP6]]
+; IEEE-NEXT:    [[TMP8:%.*]] = insertelement <2 x float> poison, float [[TMP7]], i64 0
+; IEEE-NEXT:    [[TMP9:%.*]] = extractelement <2 x float> [[SQRT_MD_1ULP]], i64 1
+; IEEE-NEXT:    [[TMP10:%.*]] = fdiv float 1.000000e+00, [[TMP9]]
+; IEEE-NEXT:    [[MD_1ULP:%.*]] = insertelement <2 x float> [[TMP8]], float [[TMP10]], i64 1
+; IEEE-NEXT:    store volatile <2 x float> [[MD_1ULP]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    [[SQRT_MD_1ULP_UNDEF:%.*]] = call <2 x float> @llvm.sqrt.v2f32(<2 x float> [[X]]), !fpmath [[META2]]
+; IEEE-NEXT:    [[TMP11:%.*]] = extractelement <2 x float> [[SQRT_MD_1ULP_UNDEF]], i64 0
+; IEEE-NEXT:    [[TMP12:%.*]] = fdiv float 1.000000e+00, [[TMP11]]
+; IEEE-NEXT:    [[TMP13:%.*]] = insertelement <2 x float> poison, float [[TMP12]], i64 0
+; IEEE-NEXT:    [[TMP14:%.*]] = extractelement <2 x float> [[SQRT_MD_1ULP_UNDEF]], i64 1
+; IEEE-NEXT:    [[TMP15:%.*]] = fdiv float undef, [[TMP14]]
+; IEEE-NEXT:    [[MD_1ULP_UNDEF:%.*]] = insertelement <2 x float> [[TMP13]], float [[TMP15]], i64 1
+; IEEE-NEXT:    store volatile <2 x float> [[MD_1ULP_UNDEF]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    [[SQRT_X_3ULP:%.*]] = call <2 x float> @llvm.sqrt.v2f32(<2 x float> [[X]]), !fpmath [[META3]]
+; IEEE-NEXT:    [[TMP16:%.*]] = extractelement <2 x float> [[SQRT_X_3ULP]], i64 0
+; IEEE-NEXT:    [[TMP17:%.*]] = call float @llvm.amdgcn.fdiv.fast(float 1.000000e+00, float [[TMP16]])
+; IEEE-NEXT:    [[TMP18:%.*]] = insertelement <2 x float> poison, float [[TMP17]], i64 0
+; IEEE-NEXT:    [[TMP19:%.*]] = extractelement <2 x float> [[SQRT_X_3ULP]], i64 1
+; IEEE-NEXT:    [[TMP20:%.*]] = call float @llvm.amdgcn.fdiv.fast(float 1.000000e+00, float [[TMP19]])
+; IEEE-NEXT:    [[FDIV_OPENCL:%.*]] = insertelement <2 x float> [[TMP18]], float [[TMP20]], i64 1
+; IEEE-NEXT:    store volatile <2 x float> [[FDIV_OPENCL]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    ret void
+;
+; DAZ-LABEL: define amdgpu_kernel void @rsq_f32_vector_fpmath
+; DAZ-SAME: (ptr addrspace(1) [[OUT:%.*]], <2 x float> [[X:%.*]]) #[[ATTR1]] {
+; DAZ-NEXT:    [[SQRT_X_NO_MD:%.*]] = call <2 x float> @llvm.sqrt.v2f32(<2 x float> [[X]])
+; DAZ-NEXT:    [[TMP1:%.*]] = extractelement <2 x float> [[SQRT_X_NO_MD]], i64 0
+; DAZ-NEXT:    [[TMP2:%.*]] = fdiv float 1.000000e+00, [[TMP1]]
+; DAZ-NEXT:    [[TMP3:%.*]] = insertelement <2 x float> poison, float [[TMP2]], i64 0
+; DAZ-NEXT:    [[TMP4:%.*]] = extractelement <2 x float> [[SQRT_X_NO_MD]], i64 1
+; DAZ-NEXT:    [[TMP5:%.*]] = fdiv float 1.000000e+00, [[TMP4]]
+; DAZ-NEXT:    [[NO_MD:%.*]] = insertelement <2 x float> [[TMP3]], float [[TMP5]], i64 1
+; DAZ-NEXT:    store volatile <2 x float> [[NO_MD]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    [[SQRT_MD_1ULP:%.*]] = call <2 x float> @llvm.sqrt.v2f32(<2 x float> [[X]]), !fpmath [[META2]]
+; DAZ-NEXT:    [[TMP6:%.*]] = extractelement <2 x float> [[SQRT_MD_1ULP]], i64 0
+; DAZ-NEXT:    [[TMP7:%.*]] = call float @llvm.amdgcn.rcp.f32(float [[TMP6]])
+; DAZ-NEXT:    [[TMP8:%.*]] = insertelement <2 x float> poison, float [[TMP7]], i64 0
+; DAZ-NEXT:    [[TMP9:%.*]] = extractelement <2 x float> [[SQRT_MD_1ULP]], i64 1
+; DAZ-NEXT:    [[TMP10:%.*]] = call float @llvm.amdgcn.rcp.f32(float [[TMP9]])
+; DAZ-NEXT:    [[MD_1ULP:%.*]] = insertelement <2 x float> [[TMP8]], float [[TMP10]], i64 1
+; DAZ-NEXT:    store volatile <2 x float> [[MD_1ULP]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    [[SQRT_MD_1ULP_UNDEF:%.*]] = call <2 x float> @llvm.sqrt.v2f32(<2 x float> [[X]]), !fpmath [[META2]]
+; DAZ-NEXT:    [[TMP11:%.*]] = extractelement <2 x float> [[SQRT_MD_1ULP_UNDEF]], i64 0
+; DAZ-NEXT:    [[TMP12:%.*]] = call float @llvm.amdgcn.rcp.f32(float [[TMP11]])
+; DAZ-NEXT:    [[TMP13:%.*]] = insertelement <2 x float> poison, float [[TMP12]], i64 0
+; DAZ-NEXT:    [[TMP14:%.*]] = extractelement <2 x float> [[SQRT_MD_1ULP_UNDEF]], i64 1
+; DAZ-NEXT:    [[TMP15:%.*]] = fdiv float undef, [[TMP14]]
+; DAZ-NEXT:    [[MD_1ULP_UNDEF:%.*]] = insertelement <2 x float> [[TMP13]], float [[TMP15]], i64 1
+; DAZ-NEXT:    store volatile <2 x float> [[MD_1ULP_UNDEF]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    [[SQRT_X_3ULP:%.*]] = call <2 x float> @llvm.sqrt.v2f32(<2 x float> [[X]]), !fpmath [[META3]]
+; DAZ-NEXT:    [[TMP16:%.*]] = extractelement <2 x float> [[SQRT_X_3ULP]], i64 0
+; DAZ-NEXT:    [[TMP17:%.*]] = call float @llvm.amdgcn.rcp.f32(float [[TMP16]])
+; DAZ-NEXT:    [[TMP18:%.*]] = insertelement <2 x float> poison, float [[TMP17]], i64 0
+; DAZ-NEXT:    [[TMP19:%.*]] = extractelement <2 x float> [[SQRT_X_3ULP]], i64 1
+; DAZ-NEXT:    [[TMP20:%.*]] = call float @llvm.amdgcn.rcp.f32(float [[TMP19]])
+; DAZ-NEXT:    [[FDIV_OPENCL:%.*]] = insertelement <2 x float> [[TMP18]], float [[TMP20]], i64 1
+; DAZ-NEXT:    store volatile <2 x float> [[FDIV_OPENCL]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    ret void
+;
+  %sqrt.x.no.md = call <2 x float> @llvm.sqrt.v2f32(<2 x float> %x)
+  %no.md = fdiv <2 x float> <float 1.0, float 1.0>, %sqrt.x.no.md
+  store volatile <2 x float> %no.md, ptr addrspace(1) %out, align 4
+
+  ; Matches the rsq instruction accuracy
+  %sqrt.md.1ulp = call <2 x float> @llvm.sqrt.v2f32(<2 x float> %x), !fpmath !2
+  %md.1ulp = fdiv <2 x float> <float 1.0, float 1.0>, %sqrt.md.1ulp, !fpmath !2
+  store volatile <2 x float> %md.1ulp, ptr addrspace(1) %out, align 4
+
+  ; Matches the rsq instruction accuracy
+  %sqrt.md.1ulp.undef = call <2 x float> @llvm.sqrt.v2f32(<2 x float> %x), !fpmath !2
+  %md.1ulp.undef = fdiv <2 x float> <float 1.0, float undef>, %sqrt.md.1ulp.undef, !fpmath !2
+  store volatile <2 x float> %md.1ulp.undef, ptr addrspace(1) %out, align 4
+
+  ; Test mismatched metadata/flags between the sqrt and fdiv
+
+  ; Test the expected opencl default pattern
+  %sqrt.x.3ulp = call <2 x float> @llvm.sqrt.v2f32(<2 x float> %x), !fpmath !3  ; OpenCL default requires 3 for sqrt and 2.5 for fdiv
+  %fdiv.opencl = fdiv <2 x float> <float 1.0, float 1.0>, %sqrt.x.3ulp, !fpmath !0
+  store volatile <2 x float> %fdiv.opencl, ptr addrspace(1) %out, align 4
 
-  %md.half.ulp = fdiv float %a, %b, !fpmath !1
-  store volatile float %md.half.ulp, ptr addrspace(1) %out
+  ret void
+}
 
-  %md.1ulp = fdiv float %a, %b, !fpmath !2
-  store volatile float %md.1ulp, ptr addrspace(1) %out
+define amdgpu_kernel void @multiple_arcp_fdiv_denom_nomd(ptr addrspace(1) %out, float %x, float %y, float %denom) {
+; CHECK-LABEL: define amdgpu_kernel void @multiple_arcp_fdiv_denom_nomd
+; CHECK-SAME: (ptr addrspace(1) [[OUT:%.*]], float [[X:%.*]], float [[Y:%.*]], float [[DENOM:%.*]]) #[[ATTR1]] {
+; CHECK-NEXT:    [[ARCP0:%.*]] = fdiv arcp float [[X]], [[DENOM]]
+; CHECK-NEXT:    [[ARCP1:%.*]] = fdiv arcp float [[Y]], [[DENOM]]
+; CHECK-NEXT:    store volatile float [[ARCP0]], ptr addrspace(1) [[OUT]], align 4
+; CHECK-NEXT:    store volatile float [[ARCP1]], ptr addrspace(1) [[OUT]], align 4
+; CHECK-NEXT:    ret void
+;
+  %arcp0 = fdiv arcp float %x, %denom
+  %arcp1 = fdiv arcp float %y, %denom
+  store volatile float %arcp0, ptr addrspace(1) %out
+  store volatile float %arcp1, ptr addrspace(1) %out
+  ret void
+}
 
-  %md.25ulp = fdiv float %a, %b, !fpmath !0
-  store volatile float %md.25ulp, ptr addrspace(1) %out
+define amdgpu_kernel void @multiple_arcp_fdiv_denom_25ulp(ptr addrspace(1) %out, float %x, float %y, float %denom) {
+; IEEE-LABEL: define amdgpu_kernel void @multiple_arcp_fdiv_denom_25ulp
+; IEEE-SAME: (ptr addrspace(1) [[OUT:%.*]], float [[X:%.*]], float [[Y:%.*]], float [[DENOM:%.*]]) #[[ATTR1]] {
+; IEEE-NEXT:    [[ARCP0:%.*]] = fdiv arcp float [[X]], [[DENOM]], !fpmath [[META0]]
+; IEEE-NEXT:    [[ARCP1:%.*]] = fdiv arcp float [[Y]], [[DENOM]], !fpmath [[META0]]
+; IEEE-NEXT:    store volatile float [[ARCP0]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    store volatile float [[ARCP1]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    ret void
+;
+; DAZ-LABEL: define amdgpu_kernel void @multiple_arcp_fdiv_denom_25ulp
+; DAZ-SAME: (ptr addrspace(1) [[OUT:%.*]], float [[X:%.*]], float [[Y:%.*]], float [[DENOM:%.*]]) #[[ATTR1]] {
+; DAZ-NEXT:    [[ARCP0:%.*]] = call arcp float @llvm.amdgcn.fdiv.fast(float [[X]], float [[DENOM]])
+; DAZ-NEXT:    [[ARCP1:%.*]] = call arcp float @llvm.amdgcn.fdiv.fast(float [[Y]], float [[DENOM]])
+; DAZ-NEXT:    store volatile float [[ARCP0]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    store volatile float [[ARCP1]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    ret void
+;
+  %arcp0 = fdiv arcp float %x, %denom, !fpmath !0
+  %arcp1 = fdiv arcp float %y, %denom, !fpmath !0
+  store volatile float %arcp0, ptr addrspace(1) %out
+  store volatile float %arcp1, ptr addrspace(1) %out
+  ret void
+}
 
-  %md.3ulp = fdiv float %a, %b, !fpmath !3
-  store volatile float %md.3ulp, ptr addrspace(1) %out
+define amdgpu_kernel void @multiple_arcp_fdiv_denom_25ulp_x3(ptr addrspace(1) %out, float %x, float %y, float %z, float %denom) {
+; IEEE-LABEL: define amdgpu_kernel void @multiple_arcp_fdiv_denom_25ulp_x3
+; IEEE-SAME: (ptr addrspace(1) [[OUT:%.*]], float [[X:%.*]], float [[Y:%.*]], float [[Z:%.*]], float [[DENOM:%.*]]) #[[ATTR1]] {
+; IEEE-NEXT:    [[ARCP0:%.*]] = fdiv arcp float [[X]], [[DENOM]], !fpmath [[META0]]
+; IEEE-NEXT:    [[ARCP1:%.*]] = fdiv arcp float [[Y]], [[DENOM]], !fpmath [[META0]]
+; IEEE-NEXT:    [[ARCP2:%.*]] = fdiv arcp float [[Z]], [[DENOM]], !fpmath [[META0]]
+; IEEE-NEXT:    store volatile float [[ARCP0]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    store volatile float [[ARCP1]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    store volatile float [[ARCP2]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    ret void
+;
+; DAZ-LABEL: define amdgpu_kernel void @multiple_arcp_fdiv_denom_25ulp_x3
+; DAZ-SAME: (ptr addrspace(1) [[OUT:%.*]], float [[X:%.*]], float [[Y:%.*]], float [[Z:%.*]], float [[DENOM:%.*]]) #[[ATTR1]] {
+; DAZ-NEXT:    [[ARCP0:%.*]] = call arcp float @llvm.amdgcn.fdiv.fast(float [[X]], float [[DENOM]])
+; DAZ-NEXT:    [[ARCP1:%.*]] = call arcp float @llvm.amdgcn.fdiv.fast(float [[Y]], float [[DENOM]])
+; DAZ-NEXT:    [[ARCP2:%.*]] = call arcp float @llvm.amdgcn.fdiv.fast(float [[Z]], float [[DENOM]])
+; DAZ-NEXT:    store volatile float [[ARCP0]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    store volatile float [[ARCP1]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    store volatile float [[ARCP2]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    ret void
+;
+  %arcp0 = fdiv arcp float %x, %denom, !fpmath !0
+  %arcp1 = fdiv arcp float %y, %denom, !fpmath !0
+  %arcp2 = fdiv arcp float %z, %denom, !fpmath !0
+  store volatile float %arcp0, ptr addrspace(1) %out
+  store volatile float %arcp1, ptr addrspace(1) %out
+  store volatile float %arcp2, ptr addrspace(1) %out
+  ret void
+}
 
-  %fast.md.25ulp = fdiv fast float %a, %b, !fpmath !0
-  store volatile float %fast.md.25ulp, ptr addrspace(1) %out
+define amdgpu_kernel void @multiple_arcp_fdiv_denom_25ulp_nomd(ptr addrspace(1) %out, float %x, float %y, float %denom) {
+; IEEE-LABEL: define amdgpu_kernel void @multiple_arcp_fdiv_denom_25ulp_nomd
+; IEEE-SAME: (ptr addrspace(1) [[OUT:%.*]], float [[X:%.*]], float [[Y:%.*]], float [[DENOM:%.*]]) #[[ATTR1]] {
+; IEEE-NEXT:    [[ARCP0:%.*]] = fdiv arcp float [[X]], [[DENOM]], !fpmath [[META0]]
+; IEEE-NEXT:    [[ARCP1:%.*]] = fdiv arcp float [[Y]], [[DENOM]]
+; IEEE-NEXT:    store volatile float [[ARCP0]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    store volatile float [[ARCP1]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    ret void
+;
+; DAZ-LABEL: define amdgpu_kernel void @multiple_arcp_fdiv_denom_25ulp_nomd
+; DAZ-SAME: (ptr addrspace(1) [[OUT:%.*]], float [[X:%.*]], float [[Y:%.*]], float [[DENOM:%.*]]) #[[ATTR1]] {
+; DAZ-NEXT:    [[ARCP0:%.*]] = call arcp float @llvm.amdgcn.fdiv.fast(float [[X]], float [[DENOM]])
+; DAZ-NEXT:    [[ARCP1:%.*]] = fdiv arcp float [[Y]], [[DENOM]]
+; DAZ-NEXT:    store volatile float [[ARCP0]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    store volatile float [[ARCP1]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    ret void
+;
+  %arcp0 = fdiv arcp float %x, %denom, !fpmath !0
+  %arcp1 = fdiv arcp float %y, %denom
+  store volatile float %arcp0, ptr addrspace(1) %out
+  store volatile float %arcp1, ptr addrspace(1) %out
+  ret void
+}
 
-  %afn.md.25ulp = fdiv afn float %a, %b, !fpmath !0
-  store volatile float %afn.md.25ulp, ptr addrspace(1) %out
+define amdgpu_kernel void @multiple_arcp_fdiv_denom_nomd_25ulp(ptr addrspace(1) %out, float %x, float %y, float %denom) {
+; IEEE-LABEL: define amdgpu_kernel void @multiple_arcp_fdiv_denom_nomd_25ulp
+; IEEE-SAME: (ptr addrspace(1) [[OUT:%.*]], float [[X:%.*]], float [[Y:%.*]], float [[DENOM:%.*]]) #[[ATTR1]] {
+; IEEE-NEXT:    [[ARCP0:%.*]] = fdiv arcp float [[X]], [[DENOM]]
+; IEEE-NEXT:    [[ARCP1:%.*]] = fdiv arcp float [[Y]], [[DENOM]], !fpmath [[META0]]
+; IEEE-NEXT:    store volatile float [[ARCP0]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    store volatile float [[ARCP1]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    ret void
+;
+; DAZ-LABEL: define amdgpu_kernel void @multiple_arcp_fdiv_denom_nomd_25ulp
+; DAZ-SAME: (ptr addrspace(1) [[OUT:%.*]], float [[X:%.*]], float [[Y:%.*]], float [[DENOM:%.*]]) #[[ATTR1]] {
+; DAZ-NEXT:    [[ARCP0:%.*]] = fdiv arcp float [[X]], [[DENOM]]
+; DAZ-NEXT:    [[ARCP1:%.*]] = call arcp float @llvm.amdgcn.fdiv.fast(float [[Y]], float [[DENOM]])
+; DAZ-NEXT:    store volatile float [[ARCP0]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    store volatile float [[ARCP1]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    ret void
+;
+  %arcp0 = fdiv arcp float %x, %denom
+  %arcp1 = fdiv arcp float %y, %denom, !fpmath !0
+  store volatile float %arcp0, ptr addrspace(1) %out
+  store volatile float %arcp1, ptr addrspace(1) %out
+  ret void
+}
+
+define amdgpu_kernel void @multiple_arcp_fdiv_denom_1ulp(ptr addrspace(1) %out, float %x, float %y, float %denom) {
+; CHECK-LABEL: define amdgpu_kernel void @multiple_arcp_fdiv_denom_1ulp
+; CHECK-SAME: (ptr addrspace(1) [[OUT:%.*]], float [[X:%.*]], float [[Y:%.*]], float [[DENOM:%.*]]) #[[ATTR1]] {
+; CHECK-NEXT:    [[ARCP0:%.*]] = fdiv arcp float [[X]], [[DENOM]], !fpmath [[META2:![0-9]+]]
+; CHECK-NEXT:    [[ARCP1:%.*]] = fdiv arcp float [[Y]], [[DENOM]], !fpmath [[META2]]
+; CHECK-NEXT:    store volatile float [[ARCP0]], ptr addrspace(1) [[OUT]], align 4
+; CHECK-NEXT:    store volatile float [[ARCP1]], ptr addrspace(1) [[OUT]], align 4
+; CHECK-NEXT:    ret void
+;
+  %arcp0 = fdiv arcp float %x, %denom, !fpmath !2
+  %arcp1 = fdiv arcp float %y, %denom, !fpmath !2
+  store volatile float %arcp0, ptr addrspace(1) %out
+  store volatile float %arcp1, ptr addrspace(1) %out
+  ret void
+}
 
+define amdgpu_kernel void @multiple_arcp_fdiv_denom_1ulp_vector(ptr addrspace(1) %out, <2 x float> %x, <2 x float> %y, <2 x float> %denom) {
+; CHECK-LABEL: define amdgpu_kernel void @multiple_arcp_fdiv_denom_1ulp_vector
+; CHECK-SAME: (ptr addrspace(1) [[OUT:%.*]], <2 x float> [[X:%.*]], <2 x float> [[Y:%.*]], <2 x float> [[DENOM:%.*]]) #[[ATTR1]] {
+; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <2 x float> [[X]], i64 0
+; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <2 x float> [[DENOM]], i64 0
+; CHECK-NEXT:    [[TMP3:%.*]] = fdiv arcp float [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <2 x float> poison, float [[TMP3]], i64 0
+; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x float> [[X]], i64 1
+; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <2 x float> [[DENOM]], i64 1
+; CHECK-NEXT:    [[TMP7:%.*]] = fdiv arcp float [[TMP5]], [[TMP6]]
+; CHECK-NEXT:    [[ARCP0:%.*]] = insertelement <2 x float> [[TMP4]], float [[TMP7]], i64 1
+; CHECK-NEXT:    [[TMP8:%.*]] = extractelement <2 x float> [[Y]], i64 0
+; CHECK-NEXT:    [[TMP9:%.*]] = extractelement <2 x float> [[DENOM]], i64 0
+; CHECK-NEXT:    [[TMP10:%.*]] = fdiv arcp float [[TMP8]], [[TMP9]]
+; CHECK-NEXT:    [[TMP11:%.*]] = insertelement <2 x float> poison, float [[TMP10]], i64 0
+; CHECK-NEXT:    [[TMP12:%.*]] = extractelement <2 x float> [[Y]], i64 1
+; CHECK-NEXT:    [[TMP13:%.*]] = extractelement <2 x float> [[DENOM]], i64 1
+; CHECK-NEXT:    [[TMP14:%.*]] = fdiv arcp float [[TMP12]], [[TMP13]]
+; CHECK-NEXT:    [[ARCP1:%.*]] = insertelement <2 x float> [[TMP11]], float [[TMP14]], i64 1
+; CHECK-NEXT:    store volatile <2 x float> [[ARCP0]], ptr addrspace(1) [[OUT]], align 8
+; CHECK-NEXT:    store volatile <2 x float> [[ARCP1]], ptr addrspace(1) [[OUT]], align 8
+; CHECK-NEXT:    ret void
+;
+  %arcp0 = fdiv arcp <2 x float> %x, %denom, !fpmath !2
+  %arcp1 = fdiv arcp <2 x float> %y, %denom, !fpmath !2
+  store volatile <2 x float> %arcp0, ptr addrspace(1) %out
+  store volatile <2 x float> %arcp1, ptr addrspace(1) %out
   ret void
 }
 
-; CHECK-LABEL: @rcp_fpmath_dynamic_denorm(
-; CHECK: %md.25ulp = fdiv float 1.000000e+00, %x, !fpmath !2
-define float @rcp_fpmath_dynamic_denorm(float %x) #3 {
-  %md.25ulp = fdiv float 1.0, %x, !fpmath !2
-  ret float %md.25ulp
+define amdgpu_kernel void @multiple_arcp_fdiv_sqrt_denom_25ulp(ptr addrspace(1) %out, float %x, float %y, float %sqr.denom) {
+; IEEE-LABEL: define amdgpu_kernel void @multiple_arcp_fdiv_sqrt_denom_25ulp
+; IEEE-SAME: (ptr addrspace(1) [[OUT:%.*]], float [[X:%.*]], float [[Y:%.*]], float [[SQR_DENOM:%.*]]) #[[ATTR1]] {
+; IEEE-NEXT:    [[DENOM:%.*]] = call float @llvm.sqrt.f32(float [[SQR_DENOM]]), !fpmath [[META3]]
+; IEEE-NEXT:    [[ARCP0:%.*]] = fdiv arcp float [[X]], [[DENOM]], !fpmath [[META0]]
+; IEEE-NEXT:    [[ARCP1:%.*]] = fdiv arcp float [[Y]], [[DENOM]], !fpmath [[META0]]
+; IEEE-NEXT:    store volatile float [[ARCP0]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    store volatile float [[ARCP1]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    ret void
+;
+; DAZ-LABEL: define amdgpu_kernel void @multiple_arcp_fdiv_sqrt_denom_25ulp
+; DAZ-SAME: (ptr addrspace(1) [[OUT:%.*]], float [[X:%.*]], float [[Y:%.*]], float [[SQR_DENOM:%.*]]) #[[ATTR1]] {
+; DAZ-NEXT:    [[DENOM:%.*]] = call float @llvm.sqrt.f32(float [[SQR_DENOM]]), !fpmath [[META3]]
+; DAZ-NEXT:    [[ARCP0:%.*]] = call arcp float @llvm.amdgcn.fdiv.fast(float [[X]], float [[DENOM]])
+; DAZ-NEXT:    [[ARCP1:%.*]] = call arcp float @llvm.amdgcn.fdiv.fast(float [[Y]], float [[DENOM]])
+; DAZ-NEXT:    store volatile float [[ARCP0]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    store volatile float [[ARCP1]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    ret void
+;
+  %denom = call float @llvm.sqrt.f32(float %sqr.denom), !fpmath !3
+  %arcp0 = fdiv arcp float %x, %denom, !fpmath !0
+  %arcp1 = fdiv arcp float %y, %denom, !fpmath !0
+  store volatile float %arcp0, ptr addrspace(1) %out
+  store volatile float %arcp1, ptr addrspace(1) %out
+  ret void
+}
+
+define amdgpu_kernel void @multiple_arcp_fdiv_sqrt_denom_vector_25ulp(ptr addrspace(1) %out, <2 x float> %x, <2 x float> %y, <2 x float> %sqr.denom) {
+; IEEE-LABEL: define amdgpu_kernel void @multiple_arcp_fdiv_sqrt_denom_vector_25ulp
+; IEEE-SAME: (ptr addrspace(1) [[OUT:%.*]], <2 x float> [[X:%.*]], <2 x float> [[Y:%.*]], <2 x float> [[SQR_DENOM:%.*]]) #[[ATTR1]] {
+; IEEE-NEXT:    [[DENOM:%.*]] = call <2 x float> @llvm.sqrt.v2f32(<2 x float> [[SQR_DENOM]]), !fpmath [[META3]]
+; IEEE-NEXT:    [[TMP1:%.*]] = extractelement <2 x float> [[X]], i64 0
+; IEEE-NEXT:    [[TMP2:%.*]] = extractelement <2 x float> [[DENOM]], i64 0
+; IEEE-NEXT:    [[TMP3:%.*]] = fdiv arcp float [[TMP1]], [[TMP2]]
+; IEEE-NEXT:    [[TMP4:%.*]] = insertelement <2 x float> poison, float [[TMP3]], i64 0
+; IEEE-NEXT:    [[TMP5:%.*]] = extractelement <2 x float> [[X]], i64 1
+; IEEE-NEXT:    [[TMP6:%.*]] = extractelement <2 x float> [[DENOM]], i64 1
+; IEEE-NEXT:    [[TMP7:%.*]] = fdiv arcp float [[TMP5]], [[TMP6]]
+; IEEE-NEXT:    [[ARCP0:%.*]] = insertelement <2 x float> [[TMP4]], float [[TMP7]], i64 1
+; IEEE-NEXT:    [[TMP8:%.*]] = extractelement <2 x float> [[Y]], i64 0
+; IEEE-NEXT:    [[TMP9:%.*]] = extractelement <2 x float> [[DENOM]], i64 0
+; IEEE-NEXT:    [[TMP10:%.*]] = fdiv arcp float [[TMP8]], [[TMP9]]
+; IEEE-NEXT:    [[TMP11:%.*]] = insertelement <2 x float> poison, float [[TMP10]], i64 0
+; IEEE-NEXT:    [[TMP12:%.*]] = extractelement <2 x float> [[Y]], i64 1
+; IEEE-NEXT:    [[TMP13:%.*]] = extractelement <2 x float> [[DENOM]], i64 1
+; IEEE-NEXT:    [[TMP14:%.*]] = fdiv arcp float [[TMP12]], [[TMP13]]
+; IEEE-NEXT:    [[ARCP1:%.*]] = insertelement <2 x float> [[TMP11]], float [[TMP14]], i64 1
+; IEEE-NEXT:    store volatile <2 x float> [[ARCP0]], ptr addrspace(1) [[OUT]], align 8
+; IEEE-NEXT:    store volatile <2 x float> [[ARCP1]], ptr addrspace(1) [[OUT]], align 8
+; IEEE-NEXT:    ret void
+;
+; DAZ-LABEL: define amdgpu_kernel void @multiple_arcp_fdiv_sqrt_denom_vector_25ulp
+; DAZ-SAME: (ptr addrspace(1) [[OUT:%.*]], <2 x float> [[X:%.*]], <2 x float> [[Y:%.*]], <2 x float> [[SQR_DENOM:%.*]]) #[[ATTR1]] {
+; DAZ-NEXT:    [[DENOM:%.*]] = call <2 x float> @llvm.sqrt.v2f32(<2 x float> [[SQR_DENOM]]), !fpmath [[META3]]
+; DAZ-NEXT:    [[TMP1:%.*]] = extractelement <2 x float> [[X]], i64 0
+; DAZ-NEXT:    [[TMP2:%.*]] = extractelement <2 x float> [[DENOM]], i64 0
+; DAZ-NEXT:    [[TMP3:%.*]] = call arcp float @llvm.amdgcn.fdiv.fast(float [[TMP1]], float [[TMP2]])
+; DAZ-NEXT:    [[TMP4:%.*]] = insertelement <2 x float> poison, float [[TMP3]], i64 0
+; DAZ-NEXT:    [[TMP5:%.*]] = extractelement <2 x float> [[X]], i64 1
+; DAZ-NEXT:    [[TMP6:%.*]] = extractelement <2 x float> [[DENOM]], i64 1
+; DAZ-NEXT:    [[TMP7:%.*]] = call arcp float @llvm.amdgcn.fdiv.fast(float [[TMP5]], float [[TMP6]])
+; DAZ-NEXT:    [[ARCP0:%.*]] = insertelement <2 x float> [[TMP4]], float [[TMP7]], i64 1
+; DAZ-NEXT:    [[TMP8:%.*]] = extractelement <2 x float> [[Y]], i64 0
+; DAZ-NEXT:    [[TMP9:%.*]] = extractelement <2 x float> [[DENOM]], i64 0
+; DAZ-NEXT:    [[TMP10:%.*]] = call arcp float @llvm.amdgcn.fdiv.fast(float [[TMP8]], float [[TMP9]])
+; DAZ-NEXT:    [[TMP11:%.*]] = insertelement <2 x float> poison, float [[TMP10]], i64 0
+; DAZ-NEXT:    [[TMP12:%.*]] = extractelement <2 x float> [[Y]], i64 1
+; DAZ-NEXT:    [[TMP13:%.*]] = extractelement <2 x float> [[DENOM]], i64 1
+; DAZ-NEXT:    [[TMP14:%.*]] = call arcp float @llvm.amdgcn.fdiv.fast(float [[TMP12]], float [[TMP13]])
+; DAZ-NEXT:    [[ARCP1:%.*]] = insertelement <2 x float> [[TMP11]], float [[TMP14]], i64 1
+; DAZ-NEXT:    store volatile <2 x float> [[ARCP0]], ptr addrspace(1) [[OUT]], align 8
+; DAZ-NEXT:    store volatile <2 x float> [[ARCP1]], ptr addrspace(1) [[OUT]], align 8
+; DAZ-NEXT:    ret void
+;
+  %denom = call <2 x float> @llvm.sqrt.v2f32(<2 x float> %sqr.denom), !fpmath !3
+  %arcp0 = fdiv arcp <2 x float> %x, %denom, !fpmath !0
+  %arcp1 = fdiv arcp <2 x float> %y, %denom, !fpmath !0
+  store volatile <2 x float> %arcp0, ptr addrspace(1) %out
+  store volatile <2 x float> %arcp1, ptr addrspace(1) %out
+  ret void
 }
 
-; CHECK-LABEL: @rcp_dynamic_denorm(
-; CHECK: %md.25ulp = fdiv float 1.000000e+00, %x
-define float @rcp_dynamic_denorm(float %x) #3 {
-  %md.25ulp = fdiv float 1.0, %x
-  ret float %md.25ulp
+define amdgpu_kernel void @multiple_arcp_fdiv_sqrt_denom_25ulp_x3(ptr addrspace(1) %out, float %x, float %y, float %z, float %sqr.denom) {
+; IEEE-LABEL: define amdgpu_kernel void @multiple_arcp_fdiv_sqrt_denom_25ulp_x3
+; IEEE-SAME: (ptr addrspace(1) [[OUT:%.*]], float [[X:%.*]], float [[Y:%.*]], float [[Z:%.*]], float [[SQR_DENOM:%.*]]) #[[ATTR1]] {
+; IEEE-NEXT:    [[DENOM:%.*]] = call float @llvm.sqrt.f32(float [[SQR_DENOM]]), !fpmath [[META3]]
+; IEEE-NEXT:    [[ARCP0:%.*]] = fdiv arcp float [[X]], [[DENOM]], !fpmath [[META0]]
+; IEEE-NEXT:    [[ARCP1:%.*]] = fdiv arcp float [[Y]], [[DENOM]], !fpmath [[META0]]
+; IEEE-NEXT:    [[ARCP2:%.*]] = fdiv arcp float [[Z]], [[DENOM]], !fpmath [[META0]]
+; IEEE-NEXT:    store volatile float [[ARCP0]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    store volatile float [[ARCP1]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    store volatile float [[ARCP2]], ptr addrspace(1) [[OUT]], align 4
+; IEEE-NEXT:    ret void
+;
+; DAZ-LABEL: define amdgpu_kernel void @multiple_arcp_fdiv_sqrt_denom_25ulp_x3
+; DAZ-SAME: (ptr addrspace(1) [[OUT:%.*]], float [[X:%.*]], float [[Y:%.*]], float [[Z:%.*]], float [[SQR_DENOM:%.*]]) #[[ATTR1]] {
+; DAZ-NEXT:    [[DENOM:%.*]] = call float @llvm.sqrt.f32(float [[SQR_DENOM]]), !fpmath [[META3]]
+; DAZ-NEXT:    [[ARCP0:%.*]] = call arcp float @llvm.amdgcn.fdiv.fast(float [[X]], float [[DENOM]])
+; DAZ-NEXT:    [[ARCP1:%.*]] = call arcp float @llvm.amdgcn.fdiv.fast(float [[Y]], float [[DENOM]])
+; DAZ-NEXT:    [[ARCP2:%.*]] = call arcp float @llvm.amdgcn.fdiv.fast(float [[Z]], float [[DENOM]])
+; DAZ-NEXT:    store volatile float [[ARCP0]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    store volatile float [[ARCP1]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    store volatile float [[ARCP2]], ptr addrspace(1) [[OUT]], align 4
+; DAZ-NEXT:    ret void
+;
+  %denom = call float @llvm.sqrt.f32(float %sqr.denom), !fpmath !3
+  %arcp0 = fdiv arcp float %x, %denom, !fpmath !0
+  %arcp1 = fdiv arcp float %y, %denom, !fpmath !0
+  %arcp2 = fdiv arcp float %z, %denom, !fpmath !0
+  store volatile float %arcp0, ptr addrspace(1) %out
+  store volatile float %arcp1, ptr addrspace(1) %out
+  store volatile float %arcp2, ptr addrspace(1) %out
+  ret void
 }
 
-attributes #0 = { nounwind optnone noinline }
-attributes #1 = { nounwind "denormal-fp-math-f32"="preserve-sign,preserve-sign" }
-attributes #2 = { nounwind "denormal-fp-math-f32"="ieee,ieee" }
-attributes #3 = { nounwind "denormal-fp-math-f32"="dynamic,dynamic" }
+declare float @llvm.sqrt.f32(float)
+declare <2 x float> @llvm.sqrt.v2f32(<2 x float>)
+
+attributes #0 = { optnone noinline }
 
 !0 = !{float 2.500000e+00}
 !1 = !{float 5.000000e-01}
 !2 = !{float 1.000000e+00}
 !3 = !{float 3.000000e+00}
+;.
+; CHECK: [[META0]] = !{float 2.500000e+00}
+; CHECK: [[META2]] = !{float 1.000000e+00}
+;.