Specify nvidia env throughout puzzle 09

ehsanmok · ehsanmok · commit d2e0dc998eac · 2025-11-14T21:01:25.000Z
diff --git a/book/src/puzzle_09/first_case.md b/book/src/puzzle_09/first_case.md
@@ -7,7 +7,7 @@ This puzzle presents a crashing GPU program where your task is to identify the i
 **Prerequisites**: Complete [Mojo GPU Debugging Essentials](./essentials.md) to understand CUDA-GDB setup and basic debugging commands. Make sure you've run:
 
 ```bash
-pixi run setup-cuda-gdb
+pixi run -e nvidia setup-cuda-gdb
 ```
 
 This auto-detects your CUDA installation and sets up the necessary links for GPU debugging.
@@ -35,31 +35,15 @@ To experience the bug firsthand, run the following command in your terminal (`pi
 pixi run -e nvidia p09 --first-case
 ```
 
-You'll see output like when the program crashes with this error:
+You'll see output like this when the program crashes:
 
 ```txt
-CUDA call failed: CUDA_ERROR_ILLEGAL_ADDRESS (an illegal memory access was encountered)
-[24326:24326:20250801,180816.333593:ERROR file_io_posix.cc:144] open /sys/devices/system/cpu/cpu0/cpufreq/scaling_cur_freq: No such file or directory (2)
-[24326:24326:20250801,180816.333653:ERROR file_io_posix.cc:144] open /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq: No such file or directory (2)
-Please submit a bug report to https://github.com/modular/modular/issues and include the crash backtrace along with all the relevant source codes.
-Stack dump:
-0.      Program arguments: /home/ubuntu/workspace/mojo-gpu-puzzles/.pixi/envs/default/bin/mojo problems/p09/p09.mojo
-Stack dump without symbol names (ensure you have llvm-symbolizer in your PATH or set the environment var `LLVM_SYMBOLIZER_PATH` to point to it):
-0  mojo                      0x0000653a338d3d2b
-1  mojo                      0x0000653a338d158a
-2  mojo                      0x0000653a338d48d7
-3  libc.so.6                 0x00007cbc08442520
-4  libc.so.6                 0x00007cbc0851e88d syscall + 29
-5  libAsyncRTMojoBindings.so 0x00007cbc0ab68653
-6  libc.so.6                 0x00007cbc08442520
-7  libc.so.6                 0x00007cbc084969fc pthread_kill + 300
-8  libc.so.6                 0x00007cbc08442476 raise + 22
-9  libc.so.6                 0x00007cbc084287f3 abort + 211
-10 libAsyncRTMojoBindings.so 0x00007cbc097c7c7b
-11 libAsyncRTMojoBindings.so 0x00007cbc097c7c9e
-12 (error)                   0x00007cbb5c00600f
-mojo crashed!
-Please file a bug report.
+First Case: Try to identify what's wrong without looking at the code!
+
+stack trace was not collected. Enable stack trace collection with environment variable `MOJO_ENABLE_STACK_TRACE_ON_ERROR`
+Unhandled exception caught during execution: At open-source/max/mojo/stdlib/stdlib/gpu/host/device_context.mojo:2082:17: CUDA call failed: CUDA_ERROR_INVALID_IMAGE (device kernel image is invalid)
+To get more accurate error information, set MODULAR_DEVICE_CONTEXT_SYNC_MODE=true.
+/home/ubuntu/workspace/mojo-gpu-puzzles/.pixi/envs/nvidia/bin/mojo: error: execution exited with a non-zero result: 1
 ```
 
 ## Your task: detective work
diff --git a/book/src/puzzle_09/second_case.md b/book/src/puzzle_09/second_case.md
@@ -11,11 +11,10 @@ Building on your [crash debugging skills from the First Case](./first_case.md),
 
 This intermediate-level debugging challenge covers investigating **algorithmic errors** using `LayoutTensor` operations, where the program runs successfully but produces wrong output - a much more common (and trickier) real-world debugging scenario.
 
-**Prerequisites**: Complete [Mojo GPU Debugging Essentials](./essentials.md) and [Detective Work: First Case](./first_case.md) to understand CUDA-GDB workflow and systematic debugging techniques. Make sure you've run `pixi run setup-cuda-gdb` or similar symlink is available
+**Prerequisites**: Complete [Mojo GPU Debugging Essentials](./essentials.md) and [Detective Work: First Case](./first_case.md) to understand CUDA-GDB workflow and systematic debugging techniques. Make sure you run the setup:
 
 ```bash
-ln -sf /usr/local/cuda/bin/cuda-gdb-minimal $CONDA_PREFIX/bin/cuda-gdb-minimal
-ln -sf /usr/local/cuda/bin/cuda-gdb-python3.12-tui $CONDA_PREFIX/bin/cuda-gdb-python3.12-tui
+pixi run -e nvidia setup-cuda-gdb
 ```
 
 ## Key concepts
@@ -38,7 +37,7 @@ First, examine the kernel without looking at the complete code:
 To experience the bug firsthand, run the following command in your terminal (`pixi` only):
 
 ```bash
-pixi run p09 --second-case
+pixi run -e nvidia p09 --second-case
 ```
 
 You'll see output like this - **no crash, but wrong results**:
@@ -49,10 +48,10 @@ This program computes sliding window sums for each position...
 Input array: [0, 1, 2, 3]
 Computing sliding window sums (window size = 3)...
 Each position should sum its neighbors: [left + center + right]
-Actual result: HostBuffer([0.0, 1.0, 3.0, 5.0])
-Expected: [1.0, 3.0, 6.0, 5.0]
-❌ Test FAILED - Sliding window sums are incorrect!
-Check the window indexing logic...
+stack trace was not collected. Enable stack trace collection with environment variable `MOJO_ENABLE_STACK_TRACE_ON_ERROR`
+Unhandled exception caught during execution: At open-source/max/mojo/stdlib/stdlib/gpu/host/device_context.mojo:2082:17: CUDA call failed: CUDA_ERROR_INVALID_IMAGE (device kernel image is invalid)
+To get more accurate error information, set MODULAR_DEVICE_CONTEXT_SYNC_MODE=true.
+/home/ubuntu/workspace/mojo-gpu-puzzles/.pixi/envs/nvidia/bin/mojo: error: execution exited with a non-zero result: 1
 ```
 
 ## Your task: detective work
@@ -69,7 +68,7 @@ Check the window indexing logic...
 Start with:
 
 ```bash
-pixi run mojo debug --cuda-gdb --break-on-launch problems/p09/p09.mojo --second-case
+pixi run -e nvidia mojo debug --cuda-gdb --break-on-launch problems/p09/p09.mojo --second-case
 ```
 
 ### GDB command shortcuts (faster debugging)
@@ -116,7 +115,7 @@ pixi run mojo debug --cuda-gdb --break-on-launch problems/p09/p09.mojo --second-
 #### Step 1: Start the debugger
 
 ```bash
-pixi run mojo debug --cuda-gdb --break-on-launch problems/p09/p09.mojo --second-case
+pixi run -e nvidia mojo debug --cuda-gdb --break-on-launch problems/p09/p09.mojo --second-case
 ```
 
 #### Step 2: analyze the symptoms first
@@ -360,7 +359,7 @@ for offset in range(ITER):           # ← Only 2 iterations: [0, 1]
 - Focus on the algorithm logic rather than trying to inspect tensor contents
 - Use systematic reasoning to trace what each thread should vs actually accesses
 
-**💡 Key Insight**: This type of off-by-one loop bug is extremely common in GPU programming. The systematic approach you learned here - combining limited debugger info with mathematical analysis and pattern recognition - is exactly how professional GPU developers debug when tools have limitations.
+**Key Insight**: This type of off-by-one loop bug is extremely common in GPU programming. The systematic approach you learned here - combining limited debugger info with mathematical analysis and pattern recognition - is exactly how professional GPU developers debug when tools have limitations.
 
 </div>
 </details>
diff --git a/book/src/puzzle_09/third_case.md b/book/src/puzzle_09/third_case.md
@@ -11,11 +11,10 @@ You've learned debugging [memory crashes](./first_case.md) and [logic bugs](./se
 
 This advanced-level debugging challenge teaches you to investigate **thread coordination failures** using shared memory, LayoutTensor operations, and barrier synchronization - combining all the systematic investigation skills from the previous cases.
 
-**Prerequisites**: Complete [Mojo GPU Debugging Essentials](./essentials.md), [Detective Work: First Case](./first_case.md), and [Detective Work: Second Case](./second_case.md) to understand CUDA-GDB workflow, variable inspection limitations, and systematic debugging approaches. Make sure you've run `pixi run setup-cuda-gdb` or similar symlink is available
+**Prerequisites**: Complete [Mojo GPU Debugging Essentials](./essentials.md), [Detective Work: First Case](./first_case.md), and [Detective Work: Second Case](./second_case.md) to understand CUDA-GDB workflow, variable inspection limitations, and systematic debugging approaches. Make sure you run the setup:
 
 ```bash
-ln -sf /usr/local/cuda/bin/cuda-gdb-minimal $CONDA_PREFIX/bin/cuda-gdb-minimal
-ln -sf /usr/local/cuda/bin/cuda-gdb-python3.12-tui $CONDA_PREFIX/bin/cuda-gdb-python3.12-tui
+pixi run -e nvidia setup-cuda-gdb
 ```
 
 ## Key concepts
@@ -37,7 +36,7 @@ First, examine the kernel without looking at the complete code:
 To experience the bug firsthand, run the following command in your terminal (`pixi` only):
 
 ```bash
-pixi run p09 --third-case
+pixi -e nvidia run p09 --third-case
 ```
 
 You'll see output like this - **the program hangs indefinitely**:
@@ -68,7 +67,7 @@ Waiting for GPU computation to complete...
 Start with:
 
 ```bash
-pixi run mojo debug --cuda-gdb --break-on-launch problems/p09/p09.mojo --third-case
+pixi run -e nvidia mojo debug --cuda-gdb --break-on-launch problems/p09/p09.mojo --third-case
 ```
 
 ### GDB command shortcuts (faster debugging)
@@ -114,7 +113,7 @@ pixi run mojo debug --cuda-gdb --break-on-launch problems/p09/p09.mojo --third-c
 
 #### Step 1: start the debugger
 ```bash
-pixi run mojo debug --cuda-gdb --break-on-launch problems/p09/p09.mojo --third-case
+pixi run -e nvidia mojo debug --cuda-gdb --break-on-launch problems/p09/p09.mojo --third-case
 ```
 
 #### Step 2: analyze the hanging behavior
