From a0cfd8da6b07f25dbdf4bc1a4afbcffd29178388 Mon Sep 17 00:00:00 2001
From: "Chow, Marcus" <marcus.chow@amd.com>
Date: Mon, 10 Dec 2018 15:40:39 -0500
Subject: [PATCH] arch-gcn3: Add handling for Inf/overflow in CVT insts

Change-Id: I0fddffdeaebd9f45fe89f44d536f80a43de63ff5
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29953
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>
---
 src/arch/gcn3/insts/instructions.cc | 78 ++++++++++++++++++++++++++++-
 1 file changed, 77 insertions(+), 1 deletion(-)

diff --git a/src/arch/gcn3/insts/instructions.cc b/src/arch/gcn3/insts/instructions.cc
index e93278ab7..a7b89234a 100644
--- a/src/arch/gcn3/insts/instructions.cc
+++ b/src/arch/gcn3/insts/instructions.cc
@@ -7260,8 +7260,16 @@ namespace Gcn3ISA
 
         for (int lane = 0; lane < NumVecElemPerVecReg; ++lane) {
             if (wf->execMask(lane)) {
+                int exp;
+                std::frexp(src[lane],&exp);
                 if (std::isnan(src[lane])) {
                     vdst[lane] = 0;
+                } else if (std::isinf(src[lane]) || exp > 30) {
+                    if (std::signbit(src[lane])) {
+                        vdst[lane] = INT_MIN;
+                    } else {
+                        vdst[lane] = INT_MAX;
+                    }
                 } else {
                     vdst[lane] = (VecElemI32)src[lane];
                 }
@@ -7386,8 +7394,18 @@ namespace Gcn3ISA
 
         for (int lane = 0; lane < NumVecElemPerVecReg; ++lane) {
             if (wf->execMask(lane)) {
+                int exp;
+                std::frexp(src[lane],&exp);
                 if (std::isnan(src[lane])) {
                     vdst[lane] = 0;
+                } else if (std::isinf(src[lane])) {
+                    if (std::signbit(src[lane])) {
+                        vdst[lane] = 0;
+                    } else {
+                        vdst[lane] = UINT_MAX;
+                    }
+                } else if (exp > 31) {
+                    vdst[lane] = UINT_MAX;
                 } else {
                     vdst[lane] = (VecElemU32)src[lane];
                 }
@@ -7422,8 +7440,16 @@ namespace Gcn3ISA
 
         for (int lane = 0; lane < NumVecElemPerVecReg; ++lane) {
             if (wf->execMask(lane)) {
+                int exp;
+                std::frexp(src[lane],&exp);
                 if (std::isnan(src[lane])) {
                     vdst[lane] = 0;
+                } else if (std::isinf(src[lane]) || exp > 30) {
+                    if (std::signbit(src[lane])) {
+                        vdst[lane] = INT_MIN;
+                    } else {
+                        vdst[lane] = INT_MAX;
+                    }
                 } else {
                     vdst[lane] = (VecElemI32)src[lane];
                 }
@@ -7772,8 +7798,18 @@ namespace Gcn3ISA
 
         for (int lane = 0; lane < NumVecElemPerVecReg; ++lane) {
             if (wf->execMask(lane)) {
+                int exp;
+                std::frexp(src[lane],&exp);
                 if (std::isnan(src[lane])) {
                     vdst[lane] = 0;
+                } else if (std::isinf(src[lane])) {
+                    if (std::signbit(src[lane])) {
+                        vdst[lane] = 0;
+                    } else {
+                        vdst[lane] = UINT_MAX;
+                    }
+                } else if (exp > 31) {
+                    vdst[lane] = UINT_MAX;
                 } else {
                     vdst[lane] = (VecElemU32)src[lane];
                 }
@@ -25075,8 +25111,16 @@ namespace Gcn3ISA
 
         for (int lane = 0; lane < NumVecElemPerVecReg; ++lane) {
             if (wf->execMask(lane)) {
+                int exp;
+                std::frexp(src[lane],&exp);
                 if (std::isnan(src[lane])) {
                     vdst[lane] = 0;
+                } else if (std::isinf(src[lane]) || exp > 30) {
+                    if (std::signbit(src[lane])) {
+                        vdst[lane] = INT_MIN;
+                    } else {
+                        vdst[lane] = INT_MAX;
+                    }
                 } else {
                     vdst[lane] = (VecElemI32)src[lane];
                 }
@@ -25235,8 +25279,18 @@ namespace Gcn3ISA
 
         for (int lane = 0; lane < NumVecElemPerVecReg; ++lane) {
             if (wf->execMask(lane)) {
+                int exp;
+                std::frexp(src[lane],&exp);
                 if (std::isnan(src[lane])) {
                     vdst[lane] = 0;
+                } else if (std::isinf(src[lane])) {
+                    if (std::signbit(src[lane])) {
+                        vdst[lane] = 0;
+                    } else {
+                        vdst[lane] = UINT_MAX;
+                    }
+                } else if (exp > 31) {
+                    vdst[lane] = UINT_MAX;
                 } else {
                     vdst[lane] = (VecElemU32)src[lane];
                 }
@@ -25287,8 +25341,16 @@ namespace Gcn3ISA
 
         for (int lane = 0; lane < NumVecElemPerVecReg; ++lane) {
             if (wf->execMask(lane)) {
+                int exp;
+                std::frexp(src[lane],&exp);
                 if (std::isnan(src[lane])) {
                     vdst[lane] = 0;
+                } else if (std::isinf(src[lane]) || exp > 30) {
+                    if (std::signbit(src[lane])) {
+                        vdst[lane] = INT_MIN;
+                    } else {
+                        vdst[lane] = INT_MAX;
+                    }
                 } else {
                     vdst[lane] = (VecElemI32)src[lane];
                 }
@@ -25725,7 +25787,21 @@ namespace Gcn3ISA
 
         for (int lane = 0; lane < NumVecElemPerVecReg; ++lane) {
             if (wf->execMask(lane)) {
-                vdst[lane] = (VecElemU32)src[lane];
+                int exp;
+                std::frexp(src[lane],&exp);
+                if (std::isnan(src[lane])) {
+                    vdst[lane] = 0;
+                } else if (std::isinf(src[lane])) {
+                    if (std::signbit(src[lane])) {
+                        vdst[lane] = 0;
+                    } else {
+                        vdst[lane] = UINT_MAX;
+                    }
+                } else if (exp > 31) {
+                    vdst[lane] = UINT_MAX;
+                } else {
+                    vdst[lane] = (VecElemU32)src[lane];
+                }
             }
         }
 
-- 
2.30.2